chapter 1 - ResearchSpace@Auckland - The University of Auckland

Copyright Statement
The digital copy of this thesis is protected by the Copyright Act 1994 (New
Zealand). This thesis may be consulted by you, provided you comply with the
provisions of the Act and the following conditions of use:
• Any use you make of these documents or images must be for
research or private study purposes only, and you may not make
them available to any other person.
• Authors control the copyright of their thesis. You will recognise the
author's right to be identified as the author of this thesis, and due
acknowledgement will be made to the author where appropriate.
• You will obtain the author's permission before publishing any
material from their thesis.
To request permissions please use the Feedback form on our webpage.
http://researchspace.auckland.ac.nz/feedback
General copyright and disclaimer
In addition to the above conditions, authors give their consent for the digital
copy of their work to be used subject to the conditions specified on the Library
Thesis Consent Form

READING LAPITA IN NEAR OCEANIA:
INTERTIDAL AND SHALLOW-WATER POTTERY
SCATTERS, ROVIANA LAGOON, NEW GEORGIA,
SOLOMON ISLANDS
MATTHEW WALTER FELGATE
A thesis submitted in partial fulfilment of the requirements for the degree of
Doctor of Philosophy in Anthropology, University of Auckland, 2003

ABSTRACT
Lapita is the name given by archaeologists to a material culture complex distributed from
Papua New Guinea to Samoa about 3000 years ago, which marks major economic changes
in Near Oceania and the first settlement by humans of Remote Oceania. Those parts of
Solomon Islands that lie in Near Oceania, together with Bougainville, comprise a large gap
in the recorded distribution of Lapita, which the current research seeks to explain. At
Roviana Lagoon, centrally located in this gap, scatters of pottery, stone artefacts, and other
stone items are found in shallow water in this sheltered, landlocked lagoon, initially thought
to be late derivatives of Lapita. This research seeks method and theory to aid in the
interpretation of this type of archaeological record.
Intensive littoral survey discovered a wider chronological range of pottery styles
than had previously been recorded, including materials attributable directly to the Lapita
material culture complex. A study of vessel brokenness and completeness enabled sample
evaluation, estimation of a parent population from which the sample derived, assessment
of the state of preservation of the sample, and systematic choice of unit of quantification.
Studies of wave exposure of collection sites and taphonomic evidence from sherds
concluded that the cultural formation process of these sites was stilt house settlement (as
found elsewhere in Near Oceania for Lapita) over deeper water than today. Falling relative
sea levels and consequent increasing effects of swash-zone processes have resulted in high
archaeological visibility and poor state of preservation at Roviana Lagoon.
Analysis of ceramic and lithic variability and spatial analysis allowed the
construction of a provisional chronology in need of further testing. Indications are that
there is good potential to construct a robust, high-resolution ceramic chronology by
focussing on carefully controlled surface collection from this sort of location, ceramic
seriation and testing/calibration using direct dating by AMS radiocarbon and
Thermoluminescence.
Data on preservation and archaeological visibility of stilt house settlements along
a sheltered emerging coastline allows preservation and visibility for this type of settlement
to be modeled elsewhere. When such a model is applied to other areas of the Lapita gap,
which are predominantly either less favourable for preservation or less favourable for
archaeological visibility, the gap in the distribution of Lapita can be seen to be an area of
low probability of detection by archaeologists, meaning there is currently no evidence for
absence of settlement in the past, and good reason to think that Lapita was continuously
distributed across Near Oceania as a network of stilt village settlement. This finding
highlights the need for explicit models of probability of detection to discover or read the
Lapita archaeological record.
Keywords: pottery; Lapita; formation processes; surface archaeology; tidal archaeology;
Oceania

ACKNOWLEDGEMENTS
Primary supervisor, Dr. Peter Sheppard, got the New Georgia Archaeological Survey off
the ground (and into the sea), an achievement that made this research possible. As a
supervisor he gave me a lot of rope, as well as the occasional well-timed tug back to the
topic, and has always been available when needed. Peter was instrumental in obtaining
institutional financial support in the form of a University of Auckland Doctoral
Scholarship, and a research stipend from the Marsden Fund. Field expenses and substantial
analytical expenses were funded primarily from grants to the NGAS by the University of
Auckland, National Geographic Research and the Marsden Fund. Peter has also been a
steady companion and mentor in the field and at all stages of the research, and he and
Debbie have made my family and I welcome in their home on numerous occasions.
This research in effect continues an academic program begun in the 1970s by
Professor Roger C. Green, as the Southeast Solomons Culture History Project, which
eventually brought Peter Sheppard to this part of the world to work on Lapita lithics from
Solomon Islands. It was Roger Green who first focused my attention on Oceanic
prehistory, initially with an ample shirt emblazoned with pacific canoes, and later through
an intense series of lectures on Oceanic Prehistory, capped by his last 3-week fieldschool
before retirement from active fieldwork in the mid 1980s. Roger predicted in 1978 that
Lapita would be found at New Georgia, and like Peter Sheppard, has been a mentor
throughout. He also organized and personally funded petrographic analysis of ceramic
tempers, and made his extensive personal library available.
Dr. Simon Holdaway generously accepted the task of secondary supervision in midresearch
when the rules changed, and has, through suggestions in the lab, comments on
drafts, and by convening an archaeology theory reading group, contributed substantially
to the outcome.
Material and moral support from family were essential to successful completion of
the research, especially from my parents, who largely financed the final years, and elder
brother, who was always ready to help pack up the house and store incredible
accumulations of household effects every time we decamped to the Solomons, and is still
doing so.
In Solomon Islands, support and encouragement from Mr Lawrence Foanoata,
Director of the National Museum has been constant and essential. Permissions from the
Research committee of the Ministry of Education, and from the Western Province
Administration were key to the success of the research. Mr John Keopo of the National
Museum was instrumental in the smooth running of the first field season. Rhys and
Margaret Richards provided hospitality in Honiara on more than one occasion, and Rhys
was instrumental as the New Zealand High Comissioner in obtaining a residence permit
during extended fieldwork.
II

Mr Kenneth Roga, Western Province Field Archaeologist, has been a primary architect of
fieldwork throughout. Unflagging energy in the field, his encyclopaedic network of wantok
which has ensured a warm welcome on so many occasions, commitment to archaeology
despite difficult political circumstances, and a range of talents too numerous to mention
have been invaluable. A rock-steady figure at the helm of a canoe over many long journeys
through sometimes dangerous seas, his and Janet Roga’s hospitality in Gizo on so many
occasions is deeply appreciated.
At Roviana Lagoon, permission for research was received from the late Chief
Johnathan Roni. Chief Joseph Kama of Kaliquongu and Chief Nathan Kera of Saikile each
consented to research in areas under their jurisdiction. Thanks are also due to Mr. Solomon
Roni and his family, especially all at Miho in Sasavele: Sam Roni and Waring, Martha
Roni, Aggie Roni and Abel Kae, my father-in-law Sale Maebule and late mother-in-law
Mabelo Roni, Sitiveni, my wife Noseduri, and many of the next generation too numerous
to mention. Thanks also are due to members of the Elelo community who helped out in so
many ways during our extended stay, especially Hetti Lanni, Barrie and Selina Ford and
George and Visa Sapolo. The late Mr Phillip Lanni generously assisted with his time during
a visit to Gharanga, despite ill health. In Munda, thanks are due to David Kera and family,
Dave and Mariana Cook, and Trevor and Zahi Cumberland, for assistance with
communications and travel on many occasions. In Saikile, John Kororo and family abetted
us on our search for pottery at Mbaraulu, and Sae Oka was a prime mover in the survey
for pottery on Ndora Island.
During analysis and writeup at Auckland many individuals contributed their skills:
Dilys Johns and Dr Rod Wallace supplied technical and conservation advice, Hamish Mc
Donald and Tim Mackrell took all the artefact photographs (or at least all the good ones),
Joan Lawrence illustrated the pottery and lithic artefacts, Dr Simon Bickler collaborated
on the problem of sample interpretation, and wrote a simulation program to estimate
parent population of vessels. Dr Robin Parker of the Geology Department provided thin
section descriptions of lithics. Barry Curnham, also in Geology, showed me how to make
epoxy-impregnated ceramic thin sections. Peter Sheppard examined chert thin sections. Bill
Dickinson of the University of Arizona analysed and reported on ceramic thin sections, and
collaborated on a paper writing up the results. Jim Feathers of the Anthropology
Department, University of Washington, did the thermoluminescence analysis and was
generous with his time in discussing progress and results. Dr Christine Prior of the Rafter
Radiocarbon Lab organized AMS dating of samples from Honiavasa and Hoghoi, including
some careful experimentation with the Honiavasa sample. Professor Jim Allen gave
permission to cite the 1984 and 1985 Lapita Homeland Project Field Reports.
Discussions with Dr Stuart Bedford, Moira Doherty, Dr Simon Best and Dr Simon Bickler,
contributed to the research and identified several errors. Stuart, Simon Best and Moira
provided useful comments on parts of the manuscript. Stuart and Carolyn made me
welcome in their home during the final stages of writing.
III

CHAPTER 1:
CHAPTER 2:
Contents:
RESEARCH QUESTIONS AND METHODOLOGY ................ 1
Introduction: ................................................. 1
The Research Region: .......................................... 6
The Roviana Early Ceramic Archaeological Record: ............... 14
Research Questions: .......................................... 15
Approaches to Ceramic Classification and Analysis: ................ 21
Quantification: ............................................... 34
Seriation Theory : a Review: ................................... 43
Form and Function: a Review : ................................. 57
Chapter summary and conclusions: .............................. 70
A REVIEW OF CONSTRUCTIONS OF LAPITA TEMPORAL
VARIABILITY .............................................. 77
Introduction: ................................................ 77
Green’s 1978 Lapita Ceramic Series: ............................ 79
Anson’s Early Far Western Lapita: .............................. 85
Mussau: ..................................................... 88
The “Changing” Face of Lapita: ................................ 89
Summerhayes, West New Britain, Anir, and a Three-stage Lapita Ceramic
Series: ................................................ 93
The Watom Lapita Series: .....................................103
Wahome’s Seriation of Admiralties Pottery Assemblages: ...........107
Specht on Buka: ..............................................109
Wickler on Buka: .............................................111
Vanuatu: ....................................................116
New Caledonia: ..............................................119
The View from the East: Fiji and Tonga: .........................123

Lapita Temporal Variability -Conclusions: ...................... 125
CHAPTER 3:
CHAPTER 4:
CHAPTER 5:
SCALE AND METHOD OF FIELD SURVEY REQUIRED TO GENERATE
A SAMPLE OF LAPITA SITES FOR SERIATION IN THE NEW
GEORGIA REGION ......................................... 133
Introduction: ............................................... 133
Review of Near-oceanic Survey Methods and Results: ............. 139
Sampling Theory and the Roviana/Kaliquongu Survey Regions: ..... 158
Survey Methods: ............................................ 163
Results: .................................................... 188
Discussion and Conclusions: the Two Surveys. .................... 190
CERAMICS: UNITS OF DESCRIPTION ....................... 193
Introduction: ............................................... 193
Database Structure: .......................................... 195
Thickness, Form and Decoration of Various Vessel Parts Represented on
Sherds: .............................................. 202
Vessel Form: ............................................... 205
Decoration: ................................................. 219
Transforming Relational Data into a Flat Table: .................. 242
Chapter Summary and Conclusions: ............................ 244
SAMPLE EVALUATION AND QUANTIFYING SAMPLE SIZE. ... 247
Introduction ................................................ 247
Methods for Establishing Vessel Brokenness and Completeness: ..... 248
Simulation Approach: ........................................ 253
Statistical Approaches: ....................................... 257
Simulation Results: .......................................... 259
Statistical Results: ........................................... 262

Discussion: ..................................................263
CHAPTER 6:
CHAPTER 7:
CHAPTER 8:
Summary and Conclusions: ....................................263
WAVE EXPOSURE ..........................................269
Introduction: ................................................269
Review: .....................................................270
Roviana Intertidal Ceramics and Waves: .........................272
Solomon Islands Tides: ........................................277
Winds: ......................................................280
Measuring Fetch: .............................................282
Results: .....................................................283
Discussion: ..................................................283
Chapter Summary and Conclusions: ......................286
SITE/ASSEMBLAGE FORMATION PROCESSES (SHERD
TAPHONOMY). .............................................289
Introduction: ................................................289
Review: Middle Range Theory for the Identification of Formation Processes:
......................................................291
Models of Cultural Formation Process: ...........................298
Models of Post-depositional Formation Process: ...................299
Summary of Models: ..........................................301
Qualitative Evidence From Samples: .............................302
Quantitative Evidence: Size/Density Effects and the Structure and Context
of the Deposit: ..........................................305
Sea Level History: ............................................312
Chapter Summary and Conclusions: .............................313

VESSEL FORM VARIABILITY AND VESSEL FUNCTION ....... 319
CHAPTER 9:
Introduction: ............................................... 319
Form: Initial Nominal Classification: ........................... 322
Metric Variability Within Form 6: ............................. 331
Inferring Vessel Form from Body Sherds: ....................... 339
Vessel Function from Evidence for Use-Alteration: ............... 346
Secondary Smoothing of Vessel Interiors, and Interior Neck Form: . . 347
Shoulder Form Classes of Form 6 Vessels: ...................... 351
Sherd Restriction Factor: ..................................... 353
Chapter Summary: .......................................... 355
CERAMIC DECORATIVE CLASSIFICATION AND
DECORATION/FORM VARIABILITY: ....................... 359
Introduction: ............................................... 359
Classification of Linear Motifs: ................................ 360
Part Representation in the Potsherd Sample: ..................... 361
Covariation Between Vessel Part and Decoration: ................. 364
Summary of Decorative Structure Across the Vessel: .............. 371
Interpretation of Decorative Co-variation by Vessel Part: .......... 373
Attribute Frequencies: ....................................... 374
Variability of Impressed Lips: ................................. 375
Decorative Attributes and Vessel Form: ......................... 383
Chapter Summary: .......................................... 391
CHAPTER 10:
LITHICS ................................................... 395
Introduction: ............................................... 395
Review: .................................................... 398
Roviana Lithic Artefacts: ..................................... 401
Chert Flakes/Fragments: ..................................... 411
Analysis of Hoghoi Water-rounded and Fractured Volcanic Manuports:

......................................................412
Chapter Summary: ...........................................417
CHAPTER 11:
INTRASITE SPATIAL STRUCTURE OF CERAMIC AND LITHIC
VARIABILITY: ..............................................421
Introduction: ................................................421
Selection of Sites for Intrasite Spatial Analysis: ....................423
Objectives: ..................................................423
Method .....................................................425
Zangana: ....................................................428
Hoghoi: .....................................................438
Honiavasa: ..................................................443
Chapter Summary and Conclusions. .............................447
CHAPTER 12:
CHRONOLOGY .............................................451
Introduction: ................................................451
AMS Radiocarbon Dates on Potsherds: ..........................453
Thermoluminescence (TL) Dates from Quartz-Calcite Sherds: .......458
Roviana TL Data: ............................................463
Seriation: ....................................................465
Discussion of Correspondence Analyses: the Alternatives: ...........478
Conclusions: Integrating 14 C, TL and Seriation: ...................480
CHAPTER 13:
SUMMARY AND CONCLUSIONS .............................483
Introduction: ................................................483
Summary: ...................................................484
External Comparisons: ........................................498
The Lapita Gap as an Area of Low Probability of Detection : ........503
Intertidal-Zone and Shallow-Water Archaeology: ..................504

REFERENCES .................................................... 509

List of Tables:
Table 1: Sampling, assemblage richness, and the Jaccard coefficient (p=present, a=absent).
............................................................ 52
Table 2: Reef/ Santa Cruz motif counts as given in Anson 1983. .............. 80
Table 3: Sherd counts and MNI assembled from various tables in Parker (1981). . 81
Table 4: Relative proportions of dentate and incised, recalculated from data presented by
Donovan (1973). .............................................. 82
Table 5: Summary of a review of survey methods and Lapita results. ...........155
Table 6: Site densities by period for Roviana and Kaliquongu surveys. .........188
Table 7: Structure of master record for each sherd. .........................196
Table 8: Classes of mineral identified at 10x magnification in reflected light. ....199
Table 9: Examples of descriptive syntax for tempers. .......................200
Table 10: Temper groupings after Dickinson 2000. .........................201
Table 11: Data structure of table of thicknesses for each part of the sherd. .......204
Table 12: Data structure for the table of records of sherd form attributes by vessel part.
............................................................206
Table 13: Data structure for decoration records.............................220
Table 14: Decorative techniques. .......................................221
Table 15: Decoration pattern definitions..................................229
Table 16: Decorative elements..........................................240
Table 17: Data structure for flat table of summary data of sherd properties. ......245
Table 18: Variation in lip brokenness between sites. ........................250
Table 19: Selection of sample for estimating vessel completeness. .............252
Table 20: Breakage population estimate for the combined Roviana highly decorated lip
sample using the statistic of Chao 1984. ............................262
Table 21: Fetch measurements for collection sites as an indicator of wave exposure.
............................................................283
Table 22: Total sherd count and average sherd area by collection site, resulting from
combined effects of collection intensity and wave exposure. ............286
Table 23: Ratio of lip-rim sherds to body sherds as a measure of collector effect.

........................................................... 309
Table 24: Ratios of plain to decorated sherds, controlled by vessel part (lips, rims, necks
and shoulders. ............................................... 310
Table 25: Body sherd to neck sherd ratio as an index of sherd skipping. ........ 311
Table 26: Counts of vessel form classes. ................................ 336
Table 27: The attribute “even/not even” by site. .......................... 348
Table 28: Relative abundance of “hard neck” interior profiles to “not hard”. .... 350
Table 29: Relative abundance of even interiors with hard neck interior profiles to even
interiors without hard profile. ................................... 350
Table 30: Sherd restriction factor comparison of site samples. ............... 353
Table 31: Preservational bias of vessel parts as an explanation for differences in sherd
restriction factor (data has the form average thickness (count) standard deviation).
........................................................... 354
Table 32: Part representation, all sherds including necked sherds but excluding carinated
and/or inverted-rim sherds. ..................................... 362
Table 33: Part representation for inverted-rim sherds (no neck corner point expected from
morphology of sherd). ......................................... 362
Table 34: Carinated sherds (excluding carinated sherds with inverted rims). .... 363
Table 35: Form strength variation for different pottery styles and the effect on part
representation. ............................................... 363
Table 36: Lip decoration and rim decoration.............................. 365
Table 37: Lip decoration and neck decoration............................. 366
Table 38: Cross-tabulation of lip decoration with shoulder decoration. ......... 367
Table 39: Cross-tabulation of rim decoration and neck decoration. ............ 368
Table 40: Cross-tabulation of rim decoration and shoulder decoration. ......... 369
Table 41: Cross-tabulation of neck decoration and shoulder decoration. ........ 370
Table 42: Counts of occurrences of decorative attributes. ................... 375
Table 43: Petrographic descriptions of lithic artefacts. ...................... 407
Table 44: Petrographic classification of manuports collected at Hoghoi. ....... 414
Table 45: Size comparisons between petrographic classes of lithic manuports. . . 415
Table 46: Radiocarbon determinations from pottery (calibrated using OxCal 3.5, Stuiver
et al. 1998 atmospheric data). ................................... 454

Table 47: Thermoluminescence dating results. Precision shown is at confidence limits of
one standard deviation. .........................................464
Table 48: Definitions of attribute codes used in seriation tables and plots. .......465
Table 49: Relative contribution of first and second components of CA using all attributes
and forms. ...................................................468
Table 50: CA diagnostics by attribute, all attributes included. .................468
Table 51: CA diagnostics by site, all attributes included (*=inertia outlier). ......470
Table 52: Eigenvalues and contributions to intertia of CA components, for data excluding
form-correlated attributes. .......................................473
Table 53: CA diagnostic table for attributes, form-correlated attributes omitted (*=intertia
outlier). ......................................................473
Table 54: CA diagnostics by site, form-correlated attributes omitted (*=inertia outlier).
............................................................475
Table 55: CA diagnostics for attributes, omitting Honiavasa data and form-correlated
attributes. ....................................................477
Table 56: CA diagnostics by site, omitting Honiavasa data and form-correlated attributes.
............................................................477
Table 57: CA eigenvalues and component contributions to inertia omitting Honiavasa data
and form-correlated attributes ....................................477
Table 58: Summary of variability .......................................479

List of Figures:
Figure 1: Near Oceania, Remote Oceania and Solomon Islands, showing the location of
the New Georgia Group. ......................................... 3
Figure 2: Map of New Georgia Group showing principal geological formations (after
Coulson, Dunkerly, Hughes and Ridgeway 1987). ..................... 9
Figure 3: The author providing scale for solution notches in Plio-Pleistocene limestone
cliff near Saikile passage, Roviana Lagoon (photograph courtesy of Peter
Sheppard) .................................................... 14
Figure 4: Paniavile at low tide, looking north, with several inhabited islets and the New
Georgia mainland in the distance. ................................. 16
Figure 5: Processes of formation of the Archaeological record (After Felgate and Bickler
n.d., adapted from De Boer 1983): inferring a breakage population from an
archaeological sample........................................... 44
Figure 6: Roviana and Kaliquongu surveys as samples of the New Georgia lagoon and
barrier island system. ...........................................158
Figure 7: Slab-built carinated vessels from Honiavasa initially assigned to the Lapita
period (this assignment is examined in more detail in Chapters 8 and 9): the solid
vertical line in HV.2.464 represents an estimated location of the central vertical
avis (CVA) of the pot. ..........................................167
Figure 8: Carinated vessels from Honiavasa, showing double-line markes on some design
zones, bands of nubbins at the neck in two cases, and a band of fingernail
impression in one case (top). .....................................168
Figure 9: HV.4.202 has an incised motif laid out in double lines; HV.1.314 is dentate-
stamped, with similar design structure and a double carination; HV.2.341 is
carinated, with the design laid out in applied fillets, bounded horizontally by
decorated lap joins between slabs; HV.2.297 and HV.4.379 have bands of single
fingernail impressions and nubbins at the neck respectively. ............169
Figure 10: Incised rims with opposed-pinch fingernail impressed band at the neck,
diagnostic of the Miho subgroup of Post-Lapita styles. ................171
Figure 11: Miho-style post-Lapita sherds. ................................172
Figure 12: Miho-style post-Lapita sherds with incised shoulders. ..............172

Figure 13: Miho-style post-Lapita sherds with CVA measurements shown (MH290 was
measured at to locations on the profile; at the interior of the neck orifice and the
exterior shoulder). ............................................ 173
Figure 14: Miho-style post-Lapita sherds; dashed CVA lines are measurements based on
non-circular (uneven) curvature at the profile locations indicated by arrows.
........................................................... 174
Figure 15: Gharanga-style post-Lapita sherds. (Gharanga is a short-rim subgroup of
Gharanga-Kopo which may have multiple bands of opposed-pinch fingernail
impression on the shoulder. ..................................... 175
Figure 16: Gharanga-style post-Lapita sherds. ............................ 175
Figure 17: Gharanga/Kopo style post-Lapita sherds (top) and a large Gharanga-style
sherd (bottom) (four measurement points used as arrowed to estimate CVA.
........................................................... 176
Figure 18: Intermediate between Gharanga and Kopo styles: all post-Lapita. .... 177
Figure 19: Kopo-style post-Lapita rim sherds. ............................ 177
Figure 20: Another large Gharanga-style post-Lapita sherd from an unrestricted vessel
form........................................................ 178
Figure 21: Gharanga-style small-orifice post-Lapita sherd showing rolled rim and thin wall
common to this style. .......................................... 178
Figure 22: Kopo-Style post-Lapita sherd (taller, less everted rim than Gharanga style)
from a large-orifice vessel, with bands of impressions along both inner and outer
edges of the lip. .............................................. 178
Figure 23: Less decorated variant of Gharanga-Kopo post-Lapita style, without a strong
corner point in vertical section at the neck. ......................... 179
Figure 24: Large-orifice Gharanga/Kopo-style sherd with deformation of the lip into a
wave pattern. ................................................ 179
Figure 25: Large Gharanga-style post-Lapita sherd with typical decoration, including a
band of impressions along the inner edge of the lip. .................. 179
Figure 26: Gharanga/Kopo-style post-Lapita vessels: most are weakly restricted at the
neck, with short, heavily everted rims. Punctate band at the neck is the most
common decoration in this group, while multiple bands of fingernail pinch are
common on the short-rim examples. One sherd (MH.33) had exotic quartz-calcite

hybrid temper (see Chapter 4). ...................................180
Figure 27: Locations mentioned in the text in relation to Roviana and Kaliquongu surveys.
............................................................183
Figure 28: Kaliquongu survey transects: the unfilled symbols represent sites discovered
by informant-prospection during the Roviana survey. .................189
Figure 29: Data structure for each sherd record; each sherd can have many records in the
detail tables pertaining to the various parts of the vessel represented. .....195
Figure 30: Major vessel form variants showing part terminology; L=lip, R=rim, N=neck,
S=shoulder, C=carination, B=body; inverted or unrestricted vessels have no neck,
while for restricted vessels with everted rims (the first seven) the only distinction
in neck types is between the double neck (top centre) and the single neck
(including all unlabelled). .......................................203
Figure 31: Lip form variants showing database codes .......................207
Figure 33: Measurement of rim depth, rim Vcurve, neck angle and neck Vcurve at
different levels of brokenness. ....................................212
Figure 34: Shoulder form measurement. ..................................213
Figure 35: Derivation of conical/cannister ( C ) and spheroidal (G) sherds from various
body forms. ..................................................216
Figure 36: Form codes for vessel interiors. ...............................217
Figure 37: Possible conical base sherds from robust vessels. ..................218
Figure 38: Examples of applied decoration. ...............................223
Figure 39: Examples of applied decoration. ...............................223
Figure 40: Applied decoration on compound rims. .........................224
Figure 41: Examples of deformation of the lip into a discontinuous band. .......224
Figure 42: Horizontal deformation of the lip into a continuous band in a wave pattern.
............................................................225
Figure 43: Examples of discontinuous deformation. ........................225
Figure 44: Excision of outer lip to form a band of notches. ...................226
Figure 45: Compound rims from Nusa Roviana. NR.34 has excised lines forming the
triangle pattern on the upper rim. ..................................226
Figure 46: Impressions on the top of the lip; and spatula impressions on the neck, thought
to indicate forming of the neck using a tool. .........................227

Figure 47: Excision by rotation: one end of a small twig or rod has been poked into the
clay and the other end moved in a circle, leaving a conical hole. ....... 227
Figure 48: Perforation (upper hole). .................................... 227
Figure 49: Examples of wavy stamping and an example of dentate-stamping. . . . 228
Figure 50: Applied decoration (in combination with incised decoration) with detachment
scars indicating a v-pattern, and also possible attached disc. ........... 228
Figure 51: A band of fingernail impressions (opposed pinch), each of which is oriented
diagonal to the CVA rather than vertically, or parallel to the CVA. ...... 230
Figure 52: Deformation, perforation, and the “bnd” incomplete pattern example.
........................................................... 230
Figure 53: “cf” (Crow’s foot) pattern on the vessel rim above a band of pinching.
........................................................... 230
Figure 54: Examples of lip impression in pattern “bpi” (band parallel inner-edge of lip).
........................................................... 231
Figure 55: Examples of lip impressions laid out in pattern “bpo” (band parallel outer-
edge). ...................................................... 231
Figure 56: Fragmented examples with lip impressions assigned to “band parallel outer”
pattern. ..................................................... 232
Figure 57: Examples of lip impressions/incision laid out in patterns “bot” (band opposing
top), “bdt” (band diagonal top) and “bpt” (band parallel top), which were regarded
as equivalent in analysis due to non-exclusive nature of these descriptions.
........................................................... 232
Figure 58: Pattern expressed using linear arrangements of fingernail pinching. . . 233
Figure 59: Miscellaneous incised patterns: MH360 is middle row, left-hand column.
GW258 is bottom-right. ........................................ 233
Figure 60: Examples of applied nubbins and some bounded incised patterns (MH259 is an
unbounded pattern). ........................................... 234
Figure 61: Decorated sherds with quartz-calcite hybrid granitic temper. ........ 234
Figure 62: Unbounded incised patterns. ................................. 235
Figure 63: Thin incised rims from Hoghoi. .............................. 235
Figure 64: Unbounded incised pattern on the shoulder from Paniavile. ......... 235
Figure 65: An example of “chv” pattern, a band of unbounded linear triangles filled by

alternating fields of parallel lines. .................................236
Figure 66: Example of “chv” pattern on tall rim (pinching at neck). ............236
Figure 67 Curvilinear incised patterns. ...................................237
Figure 68: Crow’s foot mark, probably post-deposition scratching. ............237
Figure 69: Example of cross-hatch pattern “ct3”............................237
Figure 70: Cross hatch pattern “cvh”. ...................................238
Figure 71: Example of pattern “gm5". ...................................238
Figure 72: Sole example of pattern “rl1", a double-line arrangement of single repeated
fingernail impressions. .........................................238
Figure 73 Roviana vessel completeness against sampling fraction, sampling fraction
against vessel representation fraction; random assignment of shreds to vessels;
mean EVE of 5.78%. ..........................................260
Figure 74: Roviana vessel completeness against vessel representation fraction; random
assignment of sherds to vessels; mean EVE of 5.78%. .................261
Figure 76: Effect of sherd thickness on sherd strength (controlled for temper variation by
using placered volcanic tempered sherds only). ......................306
Figure 77: Body sherd size for the various temper classes. ...................306
Figure 78: Histogram of body sherd size classes by site. .....................308
Figure 79: Initial classification of vessel forms. ............................321
Figure 80: Examples of unrestricted Form 2 variants: Form 2a (top); Form 2b (upper
middle); Form 2c (lower middle); and Form 2d (bottom). ..............325
Figure 81: Additional Form 2a Sherds (top 6); the two decorated gambrelled vessels from
Nusa Roviana are a short-rim variant of Form 2a; the lower sherd is transitional
between Form 2a and Form 2d, being unrestricted. ...................326
Figure 82: Form 3 inverted restricted vessels. .............................329
Figure 83: Form 3 or Form 4 (top left) and another example of a short-rim variant of
Form 2a (top right); the sole example of Form 3 with loop handle(s) (2 nd to top);
the only confirmed Form 4 carinated bowl, in exotic temper (2 nd to bottom) and a
large base sherd or frying pan, Form 5 (bottom). .....................330
Figure 84: External neck radius, all sites, size intervals 10mm.................331
Figure 85: All sites, sherds >25cm 2 , neck Hcurve intervals 10mm..............332
Figure 86: Form 6 “Neck Hcurve” variation; the two top sherds are too small to get a

hand into (Form 6a), while the two lower sherds have head-sized or larger orifices
(Form 6b). .................................................. 334
Figure 87: Relationship between rim angle and rim height for two sites, all measurements
shown. ..................................................... 337
Figure 88: Rim angle and rim depth, all sites, filtered so that EVE is greater than 9%, to
reduce the effects of measurement error. .......................... 338
Figure 89:Body sherd curvature measurements, showing spheroidal sherds (diagonal
alignment) and other canister/conical forms (e.g. s-shaped pattern of Honiavasa
sherd measurements). ......................................... 340
Figure 90: Comparison of body sherd curvature measurements by site. ........ 342
Figure 91: Robust base sherds and cannister-shaped large body sherd (the latter having
exotic quartz-calcite temper). ................................... 344
Figure 92: Curvature of robust body sherds (thicker than 14mm). ............. 345
Figure 93: Hard interior neck profile and even interior body/shoulder profile (top); a
softer neck interior profile (middle) and uneven interior profile (bottom). . 349
Figure 94: Hard shoulder variants of Form 6 (top and middle) as distinct from the more
common soft shoulder form (bottom). ............................. 352
Figure 95: Tall, fragile everted excurvate rims. ........................... 356
Figure 96: Lip impression, single band on outer edge of lip, labeled by mark section: u=u-
shaped, v=v-shaped, o=oblique v, w=w-shaped, s=flat-bottomed groove. . 379
Figure 97: Lip impression, single band on inner edge of lip, labeled by mark section: u=u-
shaped section, etc.. ........................................... 379
Figure 98: Lip impression, single band on top face of lip, labeled by mark section: u=u-
shaped, etc.. ................................................. 380
Figure 99: Lip impression, bands on both edges of the lip. Labeled by mark section: u=u-
shaped, etc.. ................................................. 380
Figure 100: Calculation of lip orientation angle. .......................... 382
Figure 101: Lip orientation and location of impressions..................... 383
Figure 102: Rim depth by decorative class. .............................. 384
Figure 103: Rim depth and rim angle of undecorated lip-rim-neck sherds. ...... 387
Figure 104: Location of bands of lip impression in relation to rim form variability.
........................................................... 388

Figure 105: Neck thickness comparison of Gharanga/Kopo and Miho styles/types.
............................................................389
Figure 106: Neck Vcurve for Gharanga/Kopo decoration, Miho Decoration, and plain
rim-plain neck-plain shoulder sherds. ..............................390
Figure 107: Planilateral sectioned adze from Hoghoi initial surface collection (top), plano-
convex-sectioned type V adze from Zangana (middle), and planilateral adze
fragment from Zangana (bottom). .................................402
Figure 108: Images of the adzes shown in the preceding illustration. ...........403
Figure 109: Green “type VI” or “type VIII” triangular-section adze fragment from Miho
(top); butt-end of a plano-lateral-sectioned adze from Zangana South (middle) and
a fragment of a trapezoidal-sectioned Green “type IV” adze from Zangana.
............................................................404
Figure 110: Images of adzes illustrated on previous page. ....................405
Figure 111: Canarium hammerstone from H5 ceramic findspot (top) (a similar artefact
was found at Hoghoi); waisted sandstone slab from Zangana (middle); and chert
flakes from Hoghoi (bottom). ....................................406
Figure 112: Artefacts photographed from Oka collection and reported to be from the
Paniavile site: shell and stone adzes (top); stone adzes (middle) and waisted
tools/weapons and a pineapple club (bottom). ........................409
Figure 113: Waisted axes photographed from the Lanni collection, courtesy of the late
Mr. Phillip Lanni, found in the vicinity of Gharanga Stream. ............410
Figure 114: Un-ground adze preform photographed from Lanni collection courtesy of the
late Mr. Phillip Lanni, reportedly found at the Gharanga site. ...........411
Figure 115: Water-rounded lithic manuports, cortex complete. ................416
Figure 116: Size distribution of fractured lithic manuports, either with some cortex or
without. .....................................................416
Figure 117: Intertidal collection units at Zangana: numbers are values in “unit” column in
table “Flat.db” appended on CD. Units without numbers are those which yielded
no sherds. ....................................................427
Figure 118: Spatial distribution of the sherd sample at Zangana................428
Figure 119: Across-shore size sorting at Zangana...........................429
Figure 120: Possible linear settlement patterning in the distribution of large sherds at

Zangana south. ............................................... 431
Figure 121: Initial point-provenanced collection of decorated sherds at Zangana
(subsequent collection transects shown for spatial reference). .......... 432
Figure 122: Lip deformation into a wave present in both Zangana-North and Zangana-
South. ...................................................... 433
Figure 123: Bands of punctation were restricted to Zangana-North. ........... 434
Figure 124: Unbounded linear incision or necks banded with pinching at Zangana.
........................................................... 435
Figure 125: Distribution of temper classes at Zangana. ..................... 436
Figure 126: Detail of distribution of temper classes at Zangana-North.......... 437
Figure 127: Distribution of the sherd sample at Hoghoi. .................... 438
Figure 128: Lack of sherd size variation at Hoghoi, except at 35-40m (n=4). .... 439
Figure 129: Larger average manuport mass from 25m to 60m at Hoghoi. ....... 440
Figure 130: Large unfractured stones and small fractured stones concentrated between
25m and 75m at Hoghoi, with small rounded stones more widely distributed.
........................................................... 440
Figure 131: Size-sorting of manuport petrographic classes at Hoghoi, suggestive of
different size-procurement patterns by source. ...................... 441
Figure 132: Distribution of potsherd temper classes at Hoghoi. .............. 442
Figure 133: Distribution of the sherd sample at Honiavasa................... 444
Figure 134: Effects of wave refraction (and collection intensity?) on sherd size at
Honiavasa: the western margin is exposed to waves from Honiavasa channel,
which expend their energy in a swash zone at about the centre of the site at low
tide. ....................................................... 445
Figure 135: Distribution of pottery tempers at Honiavasa.................... 445
Figure 136: Co-joining sherds from deeper western margin of Honiavasa.. ..... 446
Figure 137: Distribution of pottery decorative attributes at Honiavasa.......... 447
Figure 138: A method of identifying sub-fossil organic inclusions? (Image supplied by
Rafter Radiocarbon Lab) ....................................... 455
Figure 139: Calibration of radiocarbon determination from a charcoal inclusion in a sherd
from Paniavile................................................ 455
Figure 140: Calibration of a radiocarbon determination from smoke-derived carbon on a

sherd from Hoghoi. ............................................456
Figure 141: AMS sample taken from blackened sherd at far left, found on the surface of
unit 12 at Hoghoi. The sherd to the right, found in a subsurface test of unit 15 at
Hoghoi, may be from the same vessel, and also has a sooted surface. .....457
Figure 142: Vessel with surface sooting from Hoghoi dated by AMS radiocarbon.
............................................................458
Figure 143: Attributes used in seriations, groupings explained in chapter conclusions.
............................................................467
Figure 144: Correspondence plot (attributes) using all attributes and forms. .....469
Figure 145: Correspondence plot (sites) using all attributes and forms. .........471
Figure 146: Correspondence plot (attributes) excluding form-correlated attributes
............................................................474
Figure 147: Correspondence plot (sites) excluding form-correlated attributes. ....475
Figure 148: Correspondence plot (sites), Honiavasa sample and form-correllated attributes
omitted from data-set. ..........................................478

CHAPTER 1:
RESEARCH QUESTIONS AND METHODOLOGY
Introduction:
I went to Roviana Lagoon as a student in a research team looking for Lapita pottery,
among other things. We were faced with a conundrum, in that anything that looked
vaguely like Lapita was in the sea. How to proceed? It became clear that this pattern,
beyond being a nuisance, posed key research questions fundamental to archaeological
interpretation in the region for the Lapita period.
Lapita:
Lapita pottery is a component of an archaeological horizon-style found from the Bismarck
Archipelago to Samoa, dating to approximately 1300BC-800BC. Lapita pottery marks the
first human colonization of Remote Oceania, that part of the Pacific that cannot be
reached other than by making lengthy ocean crossings (Green 1991b) indicating that this
pottery style was correlated in Remote Oceania with a maritime colonizing cultural
adaptation and a period of rapid expansion (Green 1991a). Near-Oceania has by contrast
been occupied for about 30 000 years, at least as far to the southeast as Buka, to the north
of Bougainville (Allen et al. 1989, Wickler 1995, 2001, Wickler & Spriggs 1988), thus
requiring a more complex model of local formation of the maritime colonizing adaptation,
incorporating intrusive Island-Southeast-Asian Neolithic elements, local innovations, and
integration with pre-existing indigenous populations (Green 1991a).
1

Lapita and Solomon Islands: The Lapita Gap:
Establishing the distribution of the Lapita Pottery horizon has been a major goal of Pacific
archaeology since the 1950s. The Near Oceanic Solomon Islands (the Solomon Islands
including Bougainville and excluding Ontong Java, Te Motu Province and Rennell-
Bellona) (Figure 1) comprise a major and puzzling gap in the recorded distribution of
early-Lapita pottery sites (Green 1978, Kirch 1997:53, Kirch & Hunt 1988, Roe 1992,
1993, Spriggs 1997: 128). Ultimately at issue is whether the distribution of early-Lapita
pottery was continuous or discontinuous in Near Oceania. This issue has significant
implications for our ideas about what Lapita represents. A continuous distribution of
early-Lapita pottery across this region would favour a model of the largely indigenous
development of Lapita pottery, or at least raise the probability of integration into local
populations of any migrants from Island Southeast Asia at an early stage in the
development of the Lapita cultural complex. A discontinuous distribution would favour
an avoidance model of Lapita colonization (e.g. Sergeantson & Gao 1995: 169), where
Lapita represents “foreign” intrusion, by an expansionist society living at the fringes of
an already occupied hostile and/or malarious Near-Oceanic Solomon Islands. In this sort
of model Lapita expansion “colonies” are seen as confined mostly to offshore islands in
the Bismarck Archipelago, and remain culturally and genetically relatively distinct from
earlier occupants of Near Oceania for an extended period; generally bypassing the bulk
of the Near Oceanic Solomon Islands in favour of a previously unoccupied or otherwise
healthier Remote Oceania.
2

Figure 1: Near Oceania, Remote Oceania and Solomon Islands, showing the location of the New Georgia Group.

Opinion amongst archaeologists with an interest in Lapita is divided as to the reasons for
the gap in the distribution of early Lapita sites. Two possible explanations commonly
discussed previously are (a) that the gap is purely an artefact of insufficient survey in this
region (Green 1978, Spriggs 1997:128) and (b) that the gap directly reflects a
discontinuous distribution of Lapita in this region or even complete avoidance of this
region by Lapita peoples in the past (Gorecki 1992, Roe 1992, 1993:185, Sheppard et al.
1999). Some take an equivocal position in relation to these possibilities (e.g. Kirch
1997:53) (but see also Kirch 1997:283). An additional possibility (c), less often considered
is that tectonic instability has reduced coastal site visibility/preservation in the Near-
Oceanic Solomon Islands (Dickinson & Green 1998, Kirch & Hunt 1988:18). To these a
fourth possibility is added here (d): that the primarily terrestrial archaeological survey
which has occurred in the Solomon Islands has failed to locate Lapita sites due to a pattern
in the past in this region of Lapita settlements located exclusively over the intertidal zone,
comprising stilt houses or small artificial islets. Implicit in this proposition is the hypothesis,
related to (c) above, that archaeological deposits resulting from intertidal-zone settlement
in tropical waters are easily destroyed or hidden by even slight changes in sea-level, and
are subject to destruction by wave action along most coastlines. A case is made for this
fourth possibility in this thesis.
The Lapita Ceramic Series:
While Lapita as a defined stylistic horizon has its uses, a longitudinal (temporal) view of
Lapita as an evolving ceramic series with regional expressions is a major research objective
for many. Although considerable effort has been expended to construct and compare
regional chronologies for the Lapita period (Anson 1983, 1986, 2000, Best 1984, 2002,
Green 1978, 1979, Sand 2000, Specht 1969, Spriggs 1990, Summerhayes 2000a, 2002,
Wickler 1995, 2001) the temporal dimension of the Lapita horizon and its aftermath
remains poorly defined in many areas, and this is especially true for the Western Province
4

of Solomon Islands. A major focus of any work concerned with the Lapita period must
therefore be understanding ceramic variability, including the definition of temporal
variability. The reliability and resolution of such constructs of ceramic variability are also
key issues. These subjects are treated in more detail in Chapter 2.
Archaeology of the Intertidal Zone and Subtidal Reef Flat:
The first questions that tend to arise in conversations about an archaeological record
written mainly from the intertidal-zone concern cultural formation processes: whether the
materials in the sea got there as a result of discard from settlement over water, or whether
they were re-deposited in the sea, either as a result of human agency or as a result of
erosion or subsidence. A further question contingent on the answer to these is whether
cultural or activity patterning is retained in the spatial structure of intertidal scatters, or
whether purely geomorphological processes such as sediment size-sorting are evidenced.
An understanding of both cultural and natural formation processes of intertidal
sites is critical to interpretation of the Lapita gap. Models formulated in this regard,
especially those concerning the state of preservation of sites in relation to degree of wave
exposure, have fundamental implications for all archaeologists working in Near Oceania
with an interest in the Lapita period.
A related question on which higher level cultural or behavioural inferences depend
is “what sort of samples are obtainable from these sites?” If we wish to obtain a saturated
sample of production diversity, for example (as might be used in a seriation analysis), we
might want to know whether these site-samples are representative of production in the
past, or whether they are subject to some systematic or predictable biases. What is the
sample size in terms of vessels, as represented by sherds? Is there evidence of historical
and heritable continuity linking all site samples, or are there discontinuities in the overall
sample in either of these respects? How can sample size be quantified from a collection
of sherds?
5

It is in the nature of surface archaeology as opposed to stratigraphic excavation to
go out rather than down, in search of horizontally structured variability rather than
superposition-structured variability (although intertidal-zone materials are, strictly
speaking, benthic-collected rather than surface-collected, the terrestrial terms surface-
collection and surface archaeology will be widely used in this thesis, to avoid unnecessary
jargon). The benefits of going out rather than down are manifold when the aim is to make
the region the primary unit of analysis (as advocated by Binford 1964): samples from large
areas are easily obtainable, in turn allowing more sites to be collected within the resources
available, which in turn permits a closer approach to the ideal of a saturated sample,
particularly if people in the past tended to move periodically rather than stay in one place.
When doing surface archaeology in the sea, dates by stratigraphic association are
not to be had, requiring direct dating of materials adhering to artefacts or incorporated
within artefacts. While this has its difficulties, in other ways it is no bad thing, since dating
by stratigraphic association can be problematic (Feathers 1997), particularly for small
carbonized fragments potentially affected by turbation. In stratigraphic excavation there
is no remedy in collecting large numbers of samples from chronostratigraphic units/strata:
the problem is only worsened, with no way to know whether a date should be accepted as
associated with the materials in the unit or rejected unless its age is outrageously
incongruous. A number of direct dates of various sorts were obtained in the present study,
which is encouraging, suggesting the prospects of surface archaeology are much better
than would have been imagined a few decades ago.
The Research Region:
The New Georgia Group is situated in the northwest of the Near Oceanic Solomon
Islands, in the Western Province. At a latitude of eight degrees south, climate at low
6

altitude is torrid-equatorial in the summer months, relieved on the coast by sea-breezes
during the day, while the southeast trade-wind blows during the winter months, bringing
comparatively mild showery weather. It is rare for cyclones to track directly through the
Western Province (Bath & Deguara 2003), but strong winds are often experienced during
the late summer when cyclones typically pass to the south in the region of Rennel-Bellona.
A fuller discussion of winds, tides and climate is given in Chapter 6.
Most of the New Georgia Group is covered in lowland primary rainforest, which
has been the focus of extensive log extraction in recent decades. Inland areas were
apparently intensively settled in past centuries and irrigation terraces can still be seen. The
modern population of around 20 000 is concentrated in coastal villages of a few hundred
people, and in the main towns of Gizo, Munda and Noro. Coastal areas with good
gardening soils (alluvium or uplifted backreef sediments) have a mosaic of gardens and
regenerating bush. Various forms of indigenous plantation and subsistence arboriculture
occupy significant tracts of land near the coast. A commercial plantation timber company
operates on Kolombangara. Most local transport is by outboard-powered canoe, except
within local centres or in logging areas.
Geology, Geomorphology and Related Human Geography:
Geological exploration has been comprehensive, beginning in the 1880s, with general
descriptions and with the details filled in by the Solomon Islands Geological Survey from
the 1950s till independence. Interests first in bauxite prospecting and more recently in gold
prospecting have seen many prospecting-licence surveys in the 1970s and 1980s.
Additional field survey and a synthesis of existing information were undertaken by the
British Geological Survey between 1976 and 1983 (Dunkley 1986).
The New Georgia Group was formed during an episode of volcanic activity which
began in the upper-Miocene and continues today (Dunkley 1986:7), and there is rapid and
substantial uplift of a forearc or outer arc region comprising Ranongga, Tetepare and the
7

southern part of Rendova (Figure 2), beneath which very young, warm and ductile
buoyant lithosphere of the spreading Woodlark basin is being subducted. A notable feature
of this process relevant to the archaeology is the marked reduction in seismicity relative
to Bougainville or Guadalcanal (Dunkley 1986:9). although a serious earthquake caused
shoreline changes on Vella Lavella in 1959 (Stoddart 1969b:378).
The group comprises a large number of islands trending NW-SE over a distance
of 230km with a total landmass of 5,060 km 2 (Dunkley 1986:abstract), in an area of land
and sea totaling 15,000 km 2 . The group is within 24-hour paddling distance (50km) of the
main Choiseul/Ysabel chain, 50km distant across the New Georgia Sound to the northeast.
The group is exposed to the Solomon Sea and Coral Sea to the southwest, across which
lie the Louisiade Archipelago and the Queensland coast, at distances of 540km and
1600km respectively, on what would have been an ideal return-voyaging course in relation
to the Southeast trade-wind (Irwin 1992).
The main volcanic chain (Vella Lavella, Kolombangara, New Georgia, Vangunu,
Nggatokae and Northern Rendova) comprises a series of remnant composite volcanic
cones of basaltic composition, with various small intrusive complexes, fringed by an
intricate system of uplifted Pleistocene coral reefs and lagoons (Dunkley 1986: 12). The
largest of the volcanic cones is Kolombangara, rising to 1760m, while others are in an
advanced state of erosion, exposing central sub-volcanic intrusion complexes. Upon
examining charts in 1928, W.M Davis noted that “....none of the elevated reefs of the of
the Solomon Islands is more remarkable than the emerged barrier reef which skirts...New
Georgia” (Davis 1928:397-398), and this complex was the subject of a more detailed field
study by the Royal Society Expedition to the British Solomon Islands in the mid 1960s
(Stoddart 1969a). A more recent project explicates a detailed sea-level history for this
formation (Mann et al. 1998).
8

9
Figure 2: Map of New Georgia Group showing principal geological formations (after Coulson, Dunkerly, Hughes and Ridgeway 1987).

The Roviana Formation:
The archaeological materials that form the data of this thesis are all found on intertidal or
shallow-water reef flats. “The differential uplift of the New Georgia barrier reefs has led
to the development of a wide series of reef shores, from vertical and overhanging cliffs to
horizontal intertidal flats. (Stoddart 1969b:371).” The height of the raised-reef chains
around New Georgia vary along their length and with proximity to the mainland, with local
variations in uplift due to tilting and displacement along faults (Dunkley 1986:36). Of the
upraised reefs forming lagoons around New Georgia, only the barrier reefs of the Roviana
Lagoon are suitable for gardening, with extensive areas of Pleistocene upraised backreef
or lagoonal sediments (see geological map and Dunkley 1986:37). These may have
originated as pools formed by double-barrier reef systems, which are an unusual
characteristic of the New Georgia barrier system, and which testify to a complex interplay
between tectonic subsidence followed by uplift and fluctuating relative sea-levels (Mann
et al. 1998).
There are extensive areas of uplifted backreef sediments in some parts of the New
Georgia mainland also, which are elevated to a height of 190m above sea level in the
Munda and Kazukuru areas, unlike the barrier-reef chain. It is more usual elsewhere
around New Georgia for the tops of the upraised barrier reef to be bare of soil, with
solution erosion providing a jagged micro topography, on which only trees will grow. This
seems to be the case for all of the upraised barriers around Marovo, Ngerrasi and
VonaVona lagoons. The raised reefs of the Roviana formation are typically indurated and
recrystallized, with karstic features resulting from partial dissolution. Ages are uppermost
Pliocene in some cases on fossil evidence (Dunkley 1986:38)
Deltaic fans of volcanic alluvium are found at the mouths of the larger rivers that
drain the New Georgia landmass, the higher well-drained central areas of which provide
favoured garden areas to the modern population. The actively prograding seaward margins
10

of these deltas are soft mud, and are not used at present for agriculture, other than Sago
plantation for roofing and walling material. Gardens are also located at the mouths of
smaller streams along the mainland shore, where the shore is more accessible by canoe, and
where swamp-taro may be grown in pits dug inland of the shoreline.
In Roviana lagoon, extensive areas of coastal saline flats with fresher uplifted
sandy/coralline lagoonal sediments containing fragmented Acropora corals are cyclically
hoed, and planted in sago, coconuts and bananas. These flats are roughly 30cm above the
high tide mark, and may be inundated by saltwater during storm surges or other extreme
tides. These are most likely evidence for a recent high-stand of sea-levels (late Holocene
sea level changes are quite well documented by Mann et al. 1998). Pandanus grows thickly
in untended areas. Calophyllum species prosper on this flat and are an important modern
economic resource, while a number of tree species used in manufacture of canoe parts and
fishing spears are harvested from this flat also. Modern housing is located mostly on the
uplifted backreef sediments, among gardens, but some kitchens and canoe houses are
located on the lower flat adjacent to canoe wharves. Crocodiles may traverse this lower
flat at night in search of penned livestock or unwary dogs or small humans. As this flat was
probably inundated in the pre-1000 bp period, lagoon shorelines would have been quite
steep in many places compared to their present-day conformation, making canoe
landing/storage difficult in many areas, which may have been an important influence on
settlement type in the Lapita period.
There are many sand keys to the west of Nusa Roviana, where tectonic uplift is less
than elsewhere in the Roviana lagoon, probably as a result of faulting/tilting of the Munda-
Noro reefal formation, and also to the southeast of Vangunu, at the southern margin of
Marovo Lagoon. These sand keys, while picturesque in a tropical travel-brochure sort of
way, are sometimes swept by waves during storms, and are not favoured for settlement by
local people.
11

Archaeological survey for the present thesis was confined to Roviana lagoon,
between Munda and Kalena bay, with intensive intertidal survey restricted to the
Kaliquongu region. While there is some variation in the height above sea level of the
uplifted Pleistocene reef through the survey region from Nusa Roviana to Kalena Bay, the
uplifted backreef sediments and lower coastal flats are common throughout the survey
region, although there are localized cliffed shores also in some places. This constancy
suggests a common recent relative sea-level history to the entire survey region, borne out
by the research of Mann et al. (1998). Elsewhere around New Georgia, outside the survey
frame of this research, recent sea-level history is more varied.
Regarding long-term (post-Pliocene) differences in uplift, to the east of the survey
region, between Kalena bay and Viru Harbour, the situation differs, with uplift obviously
greater, and backreef/lagoon sediments extending from the reef to the mainland, with no
modern lagoon remaining. To the West of the survey region, uplift is less, with the
Pleistocene reef barely emerged, forming the base of the sand keys that shelter the Munda
coastline. West of the Honiavasa passage the lagoon-side geomorphological record of the
uplifted barrier chain is complicated by extensive road works during the Second World
War (WWII), which in places substantially modified the shore profile to create roads and
industrial installations (Bethlehem Island and Zangana Point are examples of major
earthmoving during WWII).
The outer-arc region has more sedimentary rocks and low-grade metamorphic
rocks than the main volcanic chain, where these are virtually absent other than as a small
formation on Kolombangara. These sedimentary rocks and tuffs are of interest due to the
presence in intertidal archaeological scatters of numerous abrader fragments and adze
fragments that are either sedimentary or possibly metamorphosed sedimentary materials
(see Chapter 10). The outer-arc region is also of interest due to higher rates of tectonic
uplift, that have outstripped the Holocene marine transgression by hundreds of metres.
12

The sea-level history of the outer-arc coastlines thus contrasts markedly with that of the
New Georgia uplifted fringing reef and comparison of the archaeological record there with
Roviana Lagoon is desirable, although this has not yet been undertaken. The lower two-
thirds of the Kiorosi section on Tetepare is of Early Pleistocene age, beginning about 1.65
Ma. at the base of the exposure, and the upper third is of late-Pleistocene age, postdating
0.27Ma. (Dunkley 1986:28), and has undergone exceptionally high rates of uplift in the
Late Pleistocene. It is not clear to what extent remnant uplifted shorelines are exposed on
the outer arc, as massive slumping has occurred in places along the exposed southern coast
of Rendova/Tetepare, although Plio-Pleistocene raised fringing reefs are well preserved as
a series of terraces to the southwest of the Rendova volcanic cone.
Summary of Geology:
In summary the geology of the New Georgia Group presents a complex tapestry of rock
types, structural geology/plate tectonics and sea-level change, dominated from the
archaeologist’s point of view by raised reef formations of considerable antiquity. There are
some remaining uncertainties about the timing and magnitude of Holocene changes in
relative sea-level, but there is no evidence for significant change in the overall
conformation of the Roviana Lagoon during the late Holocene, in terms of exposure to
ocean storm waves at least. The Roviana Lagoon existed in approximately its present form
since the Holocene marine transgression, but inundation of the coastal flat during a recent
high-stand would have made many of the shorelines more difficult to occupy and land
canoes on. There seems to be no difference across the survey area in this respect, other
than the down-faulted or down-warped sand keys west of Nusa Roviana, which would
have been near-submerged during a 1m high-stand, making them even more exposed to
storm surge than at present and probably undesirable for any form of settlement. Some
cliffed lagoon-side coasts in the Aroroso Passage area that do not have
13

Figure 3: The author providing scale for solution notches in Plio-Pleistocene limestone
cliff near Saikile passage, Roviana Lagoon (photograph courtesy of Peter Sheppard)
the slightly-raised shoreline flat are examples of local topography unsuitable for the growth
of coral, along which the more usual series of shore terraces has not formed (Figure 3).
The Roviana Early Ceramic Archaeological Record:
Reeve reported four archaeological sites in the Western province, most notably the
Paniavile intertidal site in Roviana Lagoon (Figure 4),
“Containing a distinctive and possibly Lapita-related style of decorated
pottery.... While the absence of dentate-stamping makes it impossible to
consider the Paniavile material as belonging within the classic Lapita
tradition, its decorative techniques and its wide range of vessel
shapes...suggest that it represents a related tradition, possibly derived or
descended from Lapita” (Reeve 1989)
14

Reeve suggested that these ceramics closely post-date Lapita, and further suggested
similarities to a number of ceramic assemblages from elsewhere in Melanesia. Reeve
suggested that the Paniavile site in its design system is more similar to late Lapita material
at Watom (Anson 1983, Green & Anson 1991). This suggestion is examined in the course
of this thesis, and the Paniavile pottery is found to be quite dissimilar to Watom material
(see Chapter 13), although his statement, “It may be that the Paniavile ceramics in some
way represent a link between the late Lapita material... and the Mangaasi/Sohano ceramic
traditions.” (Reeve 1989) is closer to the conclusions drawn in Chapter 13.
Research Questions:
Diverse specific research questions arise in relation to these materials in the sea. The
overall question being, “How to proceed?”, leads to a series of “How to proceed”
questions and answers, rather than a Roviana prehistory.
How can the formation of the Roviana intertidal-zone and reef-flat record be
researched? What scale, method and intensity of survey is required to obtain a useful
sample set? What cultural and post-depositional formation processes can be inferred to
have produced the record? Do the Roviana materials show heritable continuity from
Lapita? Were settlements over water or on land? Can we see settlements in the distribution
of materials, or are artefact/manuport density distributions a product of post-depositional
factors? What systematic biases are there in the record? How should the record be written
(what collection/recording methods are appropriate; how much detail is needed)? How
fragile are these sites ( for example, in what state of preservation are they on discovery,
and what is the impact of surface collection)?
How can we infer the distribution of Lapita-phase pottery use in the past in Near Oceania?
What state of preservation are the Roviana materials in, and what are the taphonomic
processes affecting them? Can we model preservation of similar sites elsewhere using
these data?
15

Figure 4: Paniavile at low tide, looking north, with several inhabited islets and the New
Georgia mainland in the distance.
How might a ceramic chronology be constructed? What classificatory units fit the
sample and the period? How should the properties of samples be quantified? How can the
various dimensions of time, style, space and function be defined and identified or controlled
for? Are the available samples adequate for the task? If not, how might a better sample be
obtained? What are the prospects for obtaining direct dates, and what methodological
development research is needed in this respect?
What can we infer about social interaction from transport of lithic raw materials and
pottery temper, and how might these results be extended in future? What are the priorities
for sampling of geological source rocks and sediments in future fieldwork? What are the
implications of the recovered materials suite for future field collection strategies?
Archaeological theory and method through which these questions could be addressed is
reviewed below. The review is structured into four parts: (a) approaches to classification
16

in ceramic studies generally, with some reference to Oceania; (b) a review of quantification
methods and methods of sample evaluation, (c) a review of seriation method and theory;
and (d) review of the links between form and function. Other topics, survey and sherd
taphonomy, are reviewed in detail in their own chapters.
Fundamental to the construction of temporal variability are: survey method and
extent, ceramic description, classification, quantification of class membership, exploratory
analysis of variability, the disentanglement of style and function, and seriation theory and
method. This is especially the case where stratigraphy and superposition do not supply
clear temporal information. In Oceania we tend to dutifully search out “undisturbed”
deposits and avoid “disturbed deposits” within what could be called in its more extreme
manifestations a “bedded strata” theory of archaeological formation, searching for stratified
sequences of (especially ceramic) change. These sorts of temporal constructions are tested
against other formation theories in Chapter 2, and in the main found to be insecure when
viewed from the perspectives of contemporaneous spatial diversity, settlement pattern
instability, turbation, erosion, re-deposition, postdepositional stratification and sherd
taphonomy, these factors acting in various combinations.
Conceptions of change and classificatory systematics of Lapita studies tend to
favour, or result in, coarse periodization, the temporal resolution of which is ill-matched
to some of the suggested social explanations of Lapita, leading to a proliferation of
speculative and untested social theory. Thus some of the temporal constructions reviewed
in Chapter 2 are phrased in terms of Lapita and post-Lapita; early middle and late Lapita;
Early Eastern Lapita and Late Eastern Lapita, Early motifs and Late motifs, utilitarian
vessel forms and non-utilitarian. Similarly, coarse approaches to time often involve
conception of complex variables like site occupation span in terms of atemporal states,
where site assemblages are “slices in time”, readable as Lapita in its early state, or Lapita
in its late state.
17

Two things we do know are that a style of pottery we call Lapita was widespread
in Oceania about three thousand years ago and that it is not around in the same style today.
Lapita has been defined in various ways, as a culture (Bellwood 1978:244), as a cultural
complex and as a ceramic series (Golson 1971, Green 1992). Regardless of which
definition one chooses, the weight of evidence available to us that there was a marked
similarity in material culture across occupied Oceania around the first millennium BC is
overwhelming to the point where this can be considered to be a fact. We also know that
many non-Lapita styles of pottery have been in use in the intervening millennia, (not
worrying too much about where the cut-off point between Lapita and non-Lapita is as this
is a matter of definition: non-Lapita because these do not fit the various definitions, not
because they lack heritable continuity with Lapita), and can expect, except under an
extreme bedded-strata formation theory, that the archaeological record will present us with
a vast array of varying mixtures of various temporal and spatial components of these styles
in various stratification and superposition arrangements. To suggest on the basis of these
facts that we know the direction of ceramic change is to treat the record as though the
direction of change is constant. Such a doctrine of gradualism seems implicit in the
conclusions of some of the studies reviewed in Chapter 2, and encourages complacence
regarding the resolution with which ceramic change has thus far been defined.
It seems unlikely for example, if aims are pitched at these coarse levels of
resolution, that the challenges posed by post-processual critiques or by evolutionary
archaeology to produce historical accounts and/or to model individual and political agency
can be met. We condemn ourselves to explanation only at the broadest timescale, in which
the material effects of individuals and smaller groups over short timescales are invisible.
For Lapita, the corpus of short-term events is more limited that in many other regions and
time periods around the world: Lapita burials, for example are rare or unknown; but the
stuff of this thesis, the manufacture and discard of pots, comprises short term events
18

structured by the actions of individuals, and rich in stylistic data which can be used to tell
time, and thus there must be information to hand in the patterning of these events across
the landscape that pertains to evolutionary historicist as well as humanist orientations.
This information cannot be accessed though, without a fine-grained ceramic
chronology. Social explanations require a measure of social time or risk reading temporal
or other sorts of variability as social, or vice versa. Ceramic descriptive and classificatory
units must be formulated to capture variability that is salient to the desired level of
analytical resolution, and yet the fineness of one’s classification is constrained in reality by
the properties of that sample in terms of historical continuity and sample size. Samples
without large time-gaps or space-gaps are needed if research questions are to be pitched
at a humanist or social-evolutionary level. Also, the more complete the sample, the more
finely it can be split into classificatory units, which also affects the resolution of seriation
chronologies. Sample evaluation is thus a crucial element in establishing the type of
questions that can be asked of sample variability.
Seriation continues to be one of the backbones of archaeological dating in the
radiocarbon era, to a greater extent than was initially expected (O'Brien & Lyman
2000:226), although this is not always apparent in Pacific archaeology. Over the decades,
in the work of Green, Parker, Sheppard and Green, Irwin, Specht, Wickler, Egloff,
Wahome and Best reviewed in Chapter 2, surface scatters of ceramics are valued, and
attempts are made to seriate these in some cases. Seriation method and theory have
undergone many reincarnations through the Culture Historical era and the period of the
“New Archaeology” and behavioural archaeology, and over the last decade, in which new
analytical techniques have become widely available.
An emerging concern with formation theory is apparent in recent decades in world
archaeology, which has complicated the interpretation of archaeological variability,
particularly for ceramics, and challenged our methods of identifying temporal, stylistic and
19

functional variability in the archaeological record. Regional historical examples are given
of Oceanic seriations in Chapter 2, which are compared methodologically with current
techniques, methods and theory.
A seductive notion for those using many of the newer techniques of multivariate
exploratory analysis for seriation is that the theoretical dimensions of sample-size, space,
time, style and function equate to summary dimensions of variability in the data
(Summerhayes 2000a, Wahome 1999). Seriation is a temporal construct, and is largely a
search for temporal variability, regardless of the seriation technique used. A theme of the
review in Chapter 2 is to assess the degree to which researchers have sought out temporal,
as opposed to other sorts of variability, with which to construct their seriations.
Furthermore, form and function are not equivalent, nor are style and decoration. There is
a tendency also for archaeologists, when faced with multiple summary dimensions of
variability, to ascribe each of these uncritically to one of the above. Thus in a Principle
Components Analysis there is a tendency to ascribe one summary dimension to function,
one to sample size, and so on, often with little or no explicit justification. If purely
temporal variability is input into such a multivariate analysis, we will still get multiple
summary dimensions of variability. Some attributes may serve to differentiate some
temporal units, other stylistic variants will serve to discriminate others, as has been
recognized for many decades from simpler methods of seriation.
Techniques like Correspondence Analysis, which provided diagnostic statistics for
the purpose, allow a clear understanding of the origins of summary components of
variability, to a greater degree than some of the early computer multivariate methods that
used similarity matrices. The era of similarity matrix clustering approaches is largely past,
as such analyses are difficult to evaluate, but it will be seen that we have a substantial
residue of Lapita studies of this sort, which require careful reading.
The conclusions reached in the review in Chapter 2 regarding the significance of
20

the main constructions of Lapita and post-Lapita variability, and regarding formation of
the Lapita record in general terms, can be summarized in the suggestion that Lapita
chronology in Near-Oceania is not yet well understood at any but the coarsest level of
temporal resolution, that of the culture historical horizon, and that a major methodological
and theoretical re-evaluation is required in order to achieve better resolution than this.
More of the same sort of data is not enough, the type of data and what we do with it must
change. I thus take issue with Spriggs’ characterization of Near-Oceanic archaeology, and
Lapita archaeology in particular, as being in a data-led pioneering phase (Spriggs 1993).
It will remain in that phase only if we continue to approach the record as though we are
in that phase. Gosden’s call for “concerted frameworks” for analysis (Gosden 1991a), is
extended here to a call for more emphasis on sampling and formation frameworks, applied
not just to individual site samples, but to analysis of regional archaeological landscapes.
Approaches to Ceramic Classification and Analysis:
Introduction:
Typological approaches to the archaeological record and the construction of chronology
have their origins in nineteenth-century Europe, with developments like the Montelian
synthesis of European prehistory in the 1880s, with full-fledged development of a culture
historical approach and the notion of “archaeological cultures” apparent by the publication
of Childe’s “Dawn of European Civilization” in 1925 (Trigger 1989:163-169). Culture-
historical approaches in the Americas involved typological method that attained its
archaeological zenith in the work of Willey and Phillips (1958).
Rice advocates what has more recently been called a reflexive approach to
classification and typology (Rice 1982:49), citing Brew’s 1949 statement “We need more
rather than fewer classifications, different classifications, always new classifications, to
21

meet new needs”. Rice identifies the perennial issues as
• Are types real or created, (essentialist vs. materialist approaches, see Dunnell
1986)
• Is it better to lump or split,
• What and how many attributes should be used in classifications?
• What kinds of types are there (historical, analytical, cultural)
• What is the difference between a grouping, a classification, and a taxonomy?
Beginning with the last question, Rice suggests grouping is phenomenological, groups
consist of members, while classes are defined by criteria. Groups are historical or localized
events, classes are ahistorical. Where attributes are of equal weight, they create unordered
classes or groups, whereas where attributes or features are differentially weighted they
create hierarchical classifications, or taxonomic classifications, like the type-variety system
or the Linnaean system.
In the type-variety system of hierarchical taxonomy, “ware” was a broad, high-level
unit of synthesis and comparison, defined, for example, by such attributes as commonalities
of paste and surface treatment (Rice 1982:50). Wares were composed of groups and
groups were composed of types. Types were the basic American unit of description,
frequently ascribed a cultural meaning as a ceramic idea of a society. Types were formed
by combining varieties, the initial sorting units.
Types vs Attributes:
Shepard pointed out the principal limitations of types is that these are time-sampling and
space-sampling-dependent, decrying the 1950s paradigm that types are in some measure
cultural entities (Shepard 1963:308). She proposed the use of technical features, such as
tempering material, as providing simple criteria for delimiting types, and this influence
22

can be seen strongly in Oceania in Specht’s analysis of Buka ceramic variation (Specht
1969). Where materials exhibit continuous gradation, these do not form separate types, but
where one class of tempering material does not grade into another, this provides a useful
criterion for classification.
She regarded 1950s types as excessively coarse-grained and proposed “design
styles” as a finer classification, defined in terms of elements and motifs, and falling within
a technological tradition, this being a cluster of associated or interdependent techniques
(Shepard 1963:320). Again the influence on Specht’s approach is clear. She noted Colton’s
proposition from the 1950s that a “series” consisted of a temporal sequence of types, as
opposed to a “ceramic group” which was a contemporary grouping of pottery types within
a site of short duration and limited area, and an “index ware” which was a group of
functional types peculiar to a particular prehistoric tribe.
Classification and the New Archaeology:
In critique of the type-variety system Dunnell wrote
“If classifications of any kind are to be devices useful in constructing
explanations, if they are done for something other than amusement, they
must be capable of evaluation, susceptible to change. In short, they must
be hypotheses about the ordering of data for a specific problem....To
expect that the same set of classes defined by criteria relevant to use will
prove the most useful for chronology is foolish....Brew’s admonitions for
more classes, not fewer, is today just as true as when it was
written....Classifications need not be taken for granted. They must suit their
problem or they are useless (Dunnell 1971).”
Groupings were not seen by Dunnell as an appropriate means of constructing classes, and
were regarded rather as a means of generating and testing hypotheses about classes.
The “New Archaeology” was a radiocarbon-dating-era development (Trigger
1989:294-295) heralded by Binford as a dramatic break from the methodologies of the
past (Trigger 1989:295). The New Archaeology was marked by a shift in objectives
beyond chronology of archaeological “cultures” (the latter strongly identified with ethnic
23

groups by the later end of the culture-historical period). Grouping rather than classification
was the preferred method of analysis of variability (Graves 1998), and variation was
increasingly interpreted using anthropological models. Chronology was no longer the major
goal of ceramic analysis, and there was frequently an assumption of contemporaneity for
pottery variety within a settlement. Ceramic design variability could therefore inform on
small-scale prehistoric organizational properties.
Graves’ historical account of ceramic method and theory in the American
Southwest makes some points which are relevant to the Pacific and Lapita studies during
the period of the New Archaeology. Graves describes this general situation in the
American southwest as atemporality to a fault. Where previously culture history had
employed relatively coarse-grained temporal-typological analysis of stylistic and to some
extent functional variation across regions, for the early “New Archaeologists”,
understanding prehistoric society through patterning in the site came to the fore. Graves
identifies this phase of archaeology in the southwest with a lack of explicit consideration
of site formation processes, although substantial progress was made in this regard through
the 1980s.
I argue in Chapter 2 that there is an “atemporality” of this sort in some analyses
of Lapita variability of that era (e.g. Green 1978), which is ultimately founded on a C14-
based slice in time view of site assemblages, and that this approach to chronology has
become widespread in Oceanic archaeology in subsequent decades. Unlike Specht’s
approach to definition of a cultural sequence for Buka, chronology was no longer the
primary objective of ceramic analysis, as chronology could be derived from other,
ostensibly more scientific and objective information (radiocarbon dating). Instead, ceramic
variability was used to identify activity patterning or to measure the similarity of sites to
infer interaction (Green 1978) rather than to tell time. Although Green more recently
returned to ceramic variability for a chronology when the accumulating 14 C evidence
24

egan to look less clear-cut (Green 1991c) the extent to which the understanding of
ceramic variability was initially predicated on a “slice-in-time” view of sites is difficult to
appreciate without careful reading of Donovan’s thesis (Donovan 1973) and Green’s
working paper (Green 1978).
Seriation continued in widespread use around the world through the period of the
New Archaeology, but with a difference. Typological classification approaches had been
largely swept away by attribute-based grouping analyses (e.g. Le Blanc 1975). Orton
identifies the work of Shepard as a nodal point in this transition, where in addition to
typological construction of chronology, a preoccupation with the geological origins of
ceramic materials emerged as an indicator of trade/exchange, and where physical properties
of vessels were used to infer technological character (Orton et al. 1993:13). Shepard’s
design attributes were however, far removed from the decorative attribute combination
approach common in Pacific archaeology since the 1970s (Best 1984, Frost 1974, Green
1978, Irwin 1972, 1985), and more akin to Mead’s structural approach (Mead 1975) in
that she felt we should “insist that the meaning of design should be sought wherever
possible (Shepard 1963:259)”, so Orton’s review, while pertinent to Pacific analyses of
technological attributes, glosses over some of the details of the transition to attribute-based
analyses of design systems.
Shepard attributed the notion of elements being the basic, irreducible units of
design to Chapman (Shepard 1963:266-267), while motifs were more varied, complex and
distinctive than elements, and more appropriate units of design analysis than elements for
some pottery. She noted that element/motif analysis was convenient for “strictly geometric
designs”. She pointed out that element analysis might be helpful in the circumstance
where such designs were in a state of extreme fragmentation, in which only a small part of
the design appears on the fragment, but cautioned regarding higher-level decisions
about which sort of analysis was appropriate to a given set of designs, that “It would be
25

unfortunate if dependence on sherds should dictate methods of analysis”
This is the essence of two critiques by whole-design proponents levelled at the
Mead system (Mead 1975), and also the Anson system (Anson 1983, 1986, 1987) to a
lesser extent (Best 2002, Spriggs 1990), that is, that these seek to analyse the complex
Lapita face designs at the level of the sherd by using systematics suited to strictly
geometric designs. Shepard’s comments are particularly relevant to the transition from
Lapita to non-Lapita, as this is widely thought by the whole-design proponents to involve
a transition from complex pictorial representations (ill adapted to element-motif
systematics) to strictly geometric patterns (to which such systematics are well adapted).
The historical process by which this mismatch has come about is that Mead’s initial
influential formulation was for a sample of mainly geometric decoration from Yanuca in
Fiji (Mead 1975:37), while subsequent sampling of Lapita in Near-Oceania has recovered
complex designs with greater frequency. The Mead system of analysis has in many cases
been extended to cover new complex motifs without much change to the underlying
element-motif-based systematics.
Shepard’s propositions regarding ceramic attributes became mainstream by the
1970s, by which time some were using the term “micro-seriation”. Le Blanc, for example,
considered that types in any form limited the temporal resolution of seriation, as one could
only define a limited number of types, beyond which declining sample size limited their
utility (Le Blanc 1975:24). Too few types, it was argued, limit temporal resolution of the
seriation. This is compounded by the difficulty of assigning many sherds to type, even
though such sherds may display attributes.
Element/motif attribute analyses possibly make better use of the variability present
in small samples, but are subject to at least as much need for either independent or
intuitive selection of attributes of chronological saliency, if analytical noise or
contemporaneous functional variation is not to overwhelm the chronological signal. For
26

attribute-based analyses, there has been very little general discussion in recent years of how
this step of the seriation process might be achieved.
In spite of the dominance of attribute-based analyses, arguments for a structural
approach to attribute patterning continued to be put forward during the processual era:
“in pottery studies, it is not the attributes (whether technological or
decorative) that carry information, but the way these attributes are
patterned on a particular vessel shape. By focusing on a potsherd rather
than a whole vessel, the basic unit of behaviour (the vessel shape) and the
structure of the other cultural units on it are missed. Ascertaining
behavioural structure from potsherds, rather the structure on the vessel
shape, and attempting to derive culturally meaningful information from
them is like shredding a dictionary and trying to reconstruct its organization
without reconstructing the pages.” (Arnold 1985:5).”
Ceramic Attribute-combination Grouping Approaches in Oceania:
The first example of a computer-aided multivariate attribute grouping approach to ceramic
analysis in Oceania, marking the impact of the New Archaeology on Pacific ceramic
classification was the work of Everett Frost in Fiji (Frost 1974). It is important to note that
the aims in these early computer studies were largely chronological, rather than to do with
interaction or within-site activity patterning. Frost used Sokal and Sneath’s numerical
taxonomy approach, developed for taxonomy in biology (Sokal & Sneath 1963) to classify
pottery assemblages. This has historical origins (in archaeology) in non-computer
numerical grouping approaches (Rouse 1960, Spaulding 1953), and marks a fundamental
departure (in terms of systematics) from previous approaches to ceramic chronological
analysis in Oceanic archaeology (e.g. Specht 1969). A plethora of innovations in the use
of computer algorithms to sort similarity matrices had emerged in archaeology in the 1960s
and early ‘70s, (Marquardt 1978:270-273), of which Frost’s work is a fairly typical
application for the times.
Aspects of Frost’s approach, particularly Frost’s choice and coding for the
computer of ceramic sherd attributes, rapidly became semi-standardized for non-Lapita
27

ceramics. This standardizing effect can be seen in a series of analyses in the years
subsequent to Frost’s Fijian thesis research. Frost’s analytical scheme was implemented by
Irwin for his study of the Shortland Islands sequence, with some lumping of attributes,
while Irwin’s PhD research on Mailu used a similar scheme (Irwin 1972:78-79, 1985:115-
116), and others followed suit (Best 1984, Egloff 1971).
While similarity grouping methods have varied more in recent studies, the method
of coding ceramic attributes and attribute combinations has in some studies maintained this
phylogenetic similarity (see for example Bickler 1998, Clark 1999:54, Wahome 1999:36-
46). Irwin provides a detailed explanation of how this was generally done. Each
combination of decorative characteristics of multiple parts of the vessel was counted, and
the frequency of these combinations by site-assemblage was counted (omitting measured
form variables, which were analyzed separately as an “independent test” of the other
results) (Irwin 1972:77-88). In Irwin’s case this initially resulted in 300 combinations,
which were further reduced by combining some rim shape classes and excluding some data
from sherds too small to yield anything other than “nugatory” data. Some decorative
attributes were disregarded because they were seen to co-occur with others which were
retained. The number of attribute combinations was thus reduced to 103.
Variation in brokenness between assemblages is potentially problematic when
using this method of ceramic description, although perhaps this is what Irwin meant by
“nugatory”. Two identical sherds could potentially count as three entirely different
combinations if one of the sherds is broken apart. In this sort of quantification tight
control for the vessel parts represented by sherds would need to be maintained to negate
this. Best achieved such control by recording different parts of the vessel as separate
records in his data table (Best 1984), confining his attribute list to technological attributes
rather than design motifs, but here the disadvantage is the loss of information from large
sherds concerning the structure of decoration across the vessel (see also Wahome
28

1999:100). Thus depending on the level of control for part representation, this approach
can be a structural approach to whole vessels where attribute combinations are reminiscent
of types, or a part-attribute approach, akin to Arnold’s “shredded dictionary”.
Other Systems of Decorative Classification:
A less subjective, but potentially less behaviourally salient approach is to avoid selection
of attribute combinations altogether, and simply count the occurrence of all individual
attributes (e.g. Anson 1983: motif inventory appendix) (see also Wahome 1999:101-103),
rather than trying to capture patterned covariance-by-part in the descriptive scheme. This
approach has the flaw that the types of attributes used are unknown. Which attributes are
chronologically sensitive and which vary with other factors than time? This general
approach has been applied by Wickler and Summerhayes, but in both cases some form of
control for vessel location was maintained: in Wickler’s case this was at the level of motif
counts by location, while in Summerhayes’ case broader decorative techniques, rather than
individual motifs, were coded by decorative location. The difficulty here is that the larger
view of patterning of decorative attributes across the vessel is lost, but one advantage of
such an approach is that data can be obtained from quite small sherds (in Summerhayes’
case patterning of decoration across the vessel was recorded by illustrating “minimum
number of vessels” (MNV) sherd sets or families).
Another potential problem with the Anson/Summerhayes motif analysis is that
some motifs are identifiable from very small sherds, while others are larger. This provides
an additional mechanism by which motif frequencies or presence-absence may be biased
by differences in brokenness between assemblages, particularly for large incised motifs, a
problem noted in relation to Reef-Santa Cruz incised Lapita motifs (Donovan 1973:15).
Other approaches to descriptive classification of Lapita decoration include Mead’s
formulaic system of design classification (Mead 1975) (see Specht 1977 for a critique),
29

and Siorat’s structurist system, whereby variants of a design are classified by reference to
clues as to their sequence of execution (Siorat 1990). There are some differences between
these broadly similar approaches in how the distribution of decoration across the vessel is
coded. Mead’s “design zones” are not very specific in some cases, are defined in relation
to a limited range of vessel forms and are not primary to the classification of design
variability into motifs. Parker, using the Mead system, noted which locations on the
vessels each motif occurred on (Parker 1981), but these records were not reported on a
sherd-by-sherd basis, and frequency data of decorative locations were not given. Siorat
was more specific than Mead on the sequence of application of decoration, and the
location of decoration, and noted the difficulties of applying such a system to very
fragmented collections.
Green’s synthesis of Lapita variability (Green 1978) serves as a study in contrast
to the Frost/Irwin approach to description of decorative variability, although analytical
methods were very much in the Frost/Irwin attribute multivariate grouping mode, and also
included Irwin-style application of Renfrew and Sterud’s close-proximity analysis. One of
Green’s aims was to show that there was a Lapita ceramic series divisible into a western
adaptation and a derivative, but isolated, Eastern Lapita. He thus wished to identify spatial
and temporal components to the Lapita horizon. His construction of a Lapita Ceramic
series using the Mead system of decorative classification was made to test whether ceramic
variability was consistent with his theory of an isolated Eastern Lapita region.
Green’s motifs were those identified by contributors to the Mead system of design
classification (Donovan 1973, Mead 1975), and were a fundamental departure in ceramic
decorative description from the Irwin/Frost attribute combination approach. A
consequence of the schism between the Frost/Irwin approach to ceramic description and
the Mead/Donovan approach was that different periods within the Oceania regional
30

chronology began to be analyzed using these separate systems of ceramic design
description. Frost/Irwin attribute descriptions (with modifications) were generally applied
to post-Lapita ceramics (Best 1984, Clark 1999, Egloff 1971, Wahome 1999), while
Lapita-phase ceramics were analyzed using the Mead-Donovan system, or its “splitting”
derivative, the Anson system of motif inventory (Anson 1983). Another Mead/Donovan
derivative, (having an extra hierarchical level of motif classification), was added by Sharp
(Sharp 1988, Sharp 1991), allowing a bit more of a lumping approach, always beneficial
for sample sizes, but this work was never completed.
Another structurist approach to ceramic attribute description, in this case to post-
Lapita decoration, for which variations in brokenness between contexts can be controlled,
is seen in a transformation of the Irwin/Frost data structure (Clark 1999:65-66, Appendix
2). For the purpose of his multidimensional scaling analysis (MDS) of variation, Clark re-
organized the attributes “surface modification position” and “surface modification” (which
had followed the Irwin/Frost/Best structure in initial recording) into three attributes,
“Decoration- Lip”, “Decoration-Rim” and “Decoration-Body” (a similar data structure
can be found in Shennan 1997:6). In this scheme control for brokenness can be maintained
more easily than in the Frost/Irwin scheme, by coding absence of a part from a sherd into
this data structure as a filter attribute, or by limiting the analysis to a subset of sherds for
which part representation is either complete or equal (as is also possible, but more difficult,
in the Frost/Irwin system by excluding some “nugatory” combinations). The tradeoff is
proliferation of data fields or attributes, as each vessel part needs its own set of attributes,
which can lead to an unwieldly table unless a relational database is used for recording.
A potential problem with such an approach is the saliency of the vessel part
classification used. If the conception by past potters of design zones/vessel parts is of a
different level of precision to that of the analyst, some analytical “noise” can result.
31

Examples of this will be given in the Roviana ceramic analysis in this thesis (see Chapter
9). The solution to this problem lies in conducting exploratory data analysis, seeking to
understand patterning in the data, covariance between decorative attributes and location
on the vessel. If the analyst’s definition of vessel parts is too finely split, there should be
a tendency for some decorative attributes to occur on adjacent vessel parts, and these part-
attribute combinations can be lumped to boost sample size. Iterative re-classification of
both decoration and vessel parts is needed to capture structure in the data, of the layout
of decoration across the vessel.
It can be seen above that decorative classification for different periods of the
Oceanic sequence has developed along different lines. Lapita and post-Lapita/non-Lapita
studies have diverged in their units of description/classification, more so than in methods
of analysis. One reason for this is the reduction in complexity of decorative technique and
decorative patterning from Lapita to non-Lapita. Lapita technique comprises a set of
decorative tools; roulettes, dentate-stamps, circular stamps and carving blades. Post-
Lapita, (or perhaps it would be better to say non-Lapita,) these specialist tools are absent,
leaving widely available general tools such as fingernails, shell edges, sticks or leaf midribs,
and simple blades or fingernails for incision (comb incision and carved paddles are
exceptions to this trend). The potters’ construction tools may not have changed much
(paddle and anvil technique seems to be ubiquitous in all periods, if predominant post
Lapita), but the decorative toolkit became less specialized. In addition to this change to the
decorative toolkit, the complex pictorial and geometric designs of the Lapita phase give
way to simpler designs of a more geometric type, some of which may be found anywhere
from the European Neolithic (e.g. Beaker pottery) to the late-prehistoric pottery of the
Amazon Basin (e.g. Arqueologicas 1970). It is also suggested above that many analytical
schemes are sensitive to differences in the level of brokenness or fragmentation of pottery
assemblages.
32

How then might we classify Lapita and post-Lapita pottery from Oceania to avoid
making a leap in systematics in mid-analysis? What is needed is a system of design
classification sensitive to temporal variability, but insensitive to variation in the level of
fragmentation or brokenness, and salient to the Lapita/post-Lapita transition. An example
of such a design-classification approach can be seen in Wickler’s analytical classification
of incised decoration into bounded incised versus unbounded incised (using a modification
of Mead’s general zone marker concept), although Wickler does not state the advantages
of his classification. By making a decision that this simple distinction was salient to his
chronological objectives, he subsumed a range of designs within a simple binary
categorization, where Mead, faced with the same sherds, would, in attempting to describe
how each design was put together, have come up with a whole list of motifs and their
alloforms (elaborations of the same underlying motif structure), and Anson would have
been in danger of splitting even further, classifying at the level of these variants.
Latitudinal bands of decoration (general zone markers in Mead’s terminology) are
usually located near corner points, like carinations, necks and lips. Typically carinations
and necks preserve well due to their form strength, and even in very broken assemblages
one can gain a good idea of the relative frequencies of these categories. Latitudinal general
zone markers are therefore identifiable to type and location on the vessel over a range of
brokenness in some cases, since lips, necks and carinations are identifiable even when
sherds are quite small. Where larger sherds suggest these bands are continuous around the
latitude of the vessel and comprising repeats of closely spaced elements, there is no danger
of their identity changing with changes in brokenness, provided the spacing of elements
is substantially closer than the minimum dimension of sherds. For more widely spaced
bands of elements, such as when an element is only repeated three or four times around
the band of latitude, a high level of brokenness (small sherd size) would result in a
mixture of plain and decorated sherds from what was originally a single decorated vessel,
33

so the spacing of elements in relation to sherd size must be borne in mind when using
banded decoration in this manner, and the abundance of plain lip sherds would be
misleading.
Quantification:
If similarity between pottery samples is quantified using counts of the abundance of
attributes, what measures of quantity are appropriate to one’s particular samples? This
subject has been thoroughly reviewed elsewhere (Orton 1982, 1993, Orton & Tyers 1991),
but the last word on the subject has yet to be written, and this fundamental problem can
only be partially avoided by resorting to occurrence seriation (discussed below), as sample
richness is related to sample size for unsaturated samples, and sample size is a
quantification problem.
Units of Quantification and Sample Brokenness:
Orton (1993) identifies one of the principal issues in pottery quantification as whether the
measure used will bias the relative abundance of types/attributes. A complicating factor
is that the attributes chosen for study themselves affect which measures will bias the
relative abundance. A good example of this is the relative abundance of plain versus
decorated pottery. If we have a hypothetical assemblage of ten globular pots, of which five
vessels each have a different band of decoration somewhere on the vessel, while five pots
are plain, quantification of the attributes is simple, until we smash the vessels. The
complete vessels can be counted if unbroken, yielding 50% decorated. Unfortunately, if
using sherd count as a measure of quantity of the attributes decorated and plain, the more
we smash them, the plainer the assemblage gets, because the undecorated areas of the
decorated pots are counted as plain observations. Clearly simple sherd count is the wrong
34

measure of quantity for these particular attributes.
If the smashed assemblage were quantified as the logical minimum number of
vessels (MNV) represented, the five different bands of decoration would result in a count
from the pile of sherds of at least five decorated vessels, while sherds from the five
undecorated vessels would be counted as a single vessel, generating an assemblage of 83%
decorated pots. Clearly this would be a complete mismatch between decorative attributes
and unit of quantification. There are measures that would work for this hypothetical
example, for example, the sherds could be quantified using sherd area, if we had some prior
knowledge of the extent of decoration on vessels and the surface area of vessels, we could
work out that each of the summed areas of the five decorative types corresponded to a
single vessel, and that there was enough sherdage in total for ten vessels, thus by
subtraction we might arrive at the correct answer, 50% of vessels decorated. Assuming,
that is, that the recovered sample includes all sherds from each pot.
The difficulty and need for prior knowledge could be substantially alleviated
though, by forgetting about the relative abundance of undecorated sherds, and
concentrating on the relative abundances of the five different types of decoration. If the
hypothetical situation is complicated slightly though, by having one of the decorative types
extend over a greater portion of the vessel than another decorative type, then the MNV
measure of relative abundance will be better than the sherd count. Unfortunately, if there
were five pots of one decorative type and one of each of the other four types, this would
not work. In this circumstance sherd count might be giving a better approximation of the
relative abundance of the vessel decorative types. The point is that quantification is a tricky
business, even if one is not so naive as to wish to seriate using the relative abundance of
plain and decorated pottery, quantified by sherd count.
Units of Quantification and Vessel Completeness:
35

In the critical review of the Lapita temporal constructs in Chapter 2, quantification is an
issue that arises continually, both in relation to relative abundances of decorative attributes,
and in relation to quantification of sample size. Sample size(n) in the statistical sense refers
to the number of observations comprising the sample. For pottery we must ask,
observations of what? Of pots or of pieces of pots? The difficulty with pieces of pots is
that the observations are clearly not independent if decoration occurs in repeated elements
or motifs across the pot. This is why sherd count is a biased measure if the extent of
decoration across the pot is uneven, and why MNV is not if there is only one pot per
decorative type. Pots themselves may not be independent observations of behavioural
variability if production is highly standardized, thus if one potter turns out fifty identical
pots of one type, while another turns out five identical pots of another, and we are
interested in the question of how variable pottery production was between producers, the
number of observations in the sample is two, not fifty-five.
In the review that follows I try to make the point that occurrence seriations of
broken pottery assemblages are mis-applications of a technique, because sample size is so
difficult to quantify, and yet sample richness is so often related to sample size. How can
we tell which decorative attributes are common in some sites, and thus robust indicators
of presence or absence, and which are not? Can this be done by counting the pieces or by
some other measure such as MNV? I have shown above that both of these measures can
be wildly misleading in some circumstances. A more fundamental question that might in
many circumstances allow an assessment of sample size, is to ask “what sort of sample is
this?” Can the sherds in the sample be refitted to form complete vessels, or do none of
them seem to match? If they don’t match, is this because edges are rounded from some
taphonomic process, or is it because the adjacent sherds are absent from the sample? Even
if none of the sherds has matching edges, the question of whether the sherds are likely to
be from the same pot or not can still be asked. Thus one of the primary questions in
36

ceramic sample evaluation is “what is the vessel completeness of the sample?”
Analysis of vessel completeness can give an indication of which measures of sample
size and/or relative abundance are appropriate to the specific sample. At one extreme, if
there are indications that most sherds are singletons, i.e. other sherds from the same pot
are not present, then sherd count and vessel count are the same, sherd count observations
are likely to be independent in most cases, and can be used as a unit of measurement for
relative abundances and sample size. At the other extreme, where vessel completeness is
high for all vessels in the sample, a vessel-based count might be the most appropriate
measure of relative abundances, and would certainly be a much better indication of sample
size for occurrence seriation than sherd count. Vessel completeness information is
therefore an important tool in choosing an appropriate method of sample size
quantification, although vessel completeness is not always easy or even possible to
measure. Where it is impossible to measure vessel completeness, sample size is difficult to
quantify other than by resort to sum of some measure of estimated vessel equivalent, to
calculate a minimum number of vessels.
Sherd Assemblages as Samples:
The development of relatively labour-intensive sherds-as-vessels approaches to
quantification is in part a recognition of the difficulty of comparing assemblage
compositions, unless they happen to have the same levels of completeness and brokenness
(Orton 1993: 169-170), and vessel completeness cannot be measured other than by the
grouping of fragments into vessel classes. If the measured or estimated levels of
brokenness and/or completeness differ between assemblages, we must conclude that the
assemblages are not equivalent samples of the material discarded by people in the past.
How numerically large is our sample of potting behaviour of the past? Can the
sample be read directly as representing people's potting behaviour, or is it a sample of a
37

sample of a sample, subject to a number of transformations in composition? In either case,
what sort of sample (representative or biased)? What is the sampling fraction? (Again,
implicit in this question is the problem of how and what we are counting).
Several Pacific archaeologists have touched on these subjects over the years.
Specht provided a cogent discussion (Specht 1969: 70-73), Egloff published on the subject
(Egloff 1973) independently inventing a new measure of quantity that has subsequently
been widely used. Parker was careful to provide at least two different measures of sample
size in her analysis of Reef-Santa-Cruz pottery (Parker 1981). Summerhayes has given the
matter some thought, particularly in regard to the effect of variation in the extent of
decoration across the pot on the ratio of plain to decorated sherd count (Summerhayes
2000c, 2002), and Clark has noted some of the biases inherent in the MNV count (Clark
1999). On the world stage, the subject has been thoroughly reviewed for pottery in general
(Orton 1993), amid some pessimism that we will ever posses absolute measures or
estimates of vessel populations (see for example Orton 2000).
The analysis of brokenness and completeness in Chapter 5 derives from an
unpublished manuscript (Felgate & Bickler n.d), a methodological paper on estimating
parent vessel populations from sherd samples. This approach is a fundamental departure
(enabled in part by properties of samples obtained from large-area surface collections) from
the approaches usually taken in Pacific archaeology, which implicitly regard the recovered
archaeological sample as the target population, by regarding the measured properties of
the sample as representative in an unspecified way of the properties of people’s lives in the
past. We need to be more specific about the nature of that representation if our behavioural
inferences based on sherd assemblages are to be robust.
An estimate of the total quantity of pottery represented by the recovered sample
from the Roviana intertidal-zone sites is useful firstly as it gives some idea of the potential
for bias in the sample. Seriation using decorative attribute abundances will be affected if
38

the relative abundance of an attribute is linked to the degree and type of postdepositional
sampling. Comparison of estimated breakage population and sample assemblage size will
allow some idea of sampling fraction (both in terms of the bulk of sherdage and the
number of vessels), and the potential for bias. This subject will be explored further in
Chapter 7, where differences in part-representation for different vessel styles will be
investigated.
Accumulations research (estimating the intensity/duration of occupation through
the quantity of pottery in the breakage population) is possible where an estimate of the
breakage population is available (see Varien & Mills 1997, Varien & Potter 1997). If an
independent chronology is available that allows total duration and historical continuity of
occupation to be modeled, then intensity of occupation can be the focus of study. The
results of the analysis in this chapter will be examined from this perspective in Chapter
13.
Knowing how much is missing from the sample informs indirectly on
postdepositional formation processes. In this thesis evidence is presented that suggests the
vast majority of the sherdage discarded at the Roviana intertidal-zone sites is no longer
extant or recognizable as potsherds. The extent of this loss of material is estimated and
explanations for this loss are explored in Chapter 7.
Interpretation of survey results requires an assessment of the probability of
detection of evidence: when is absence of evidence acceptable evidence of absence? The
sherd sampling fraction and the vessel sampling fraction are also ways of characterizing
the state of preservation of the extant deposit, which says something about site
detectability. To what extent are they preserved in particular circumstances? Under what
general conditions of wave exposure etc. do we see these sites preserved? Developing
answers to these questions is crucial to interpreting the archaeological record of the wider
region. If almost all of the breakage population has vanished even in the sheltered setting
39

of the Roviana Lagoon, this has fundamental implications for the interpretation of the
archaeological record elsewhere in Near Oceania. Just how fragile is this site type?
Sample Formation Theory : the Breakage Population:
The central theoretical concept put forward (following Felgate & Bickler n.d) to provide
a definition of a target population is the breakage population, a theoretical stage in the
formation of an archaeological sample where vessels are broken in some unspecified
manner, but are also complete, i.e. all the fragments are present (Felgate & Bickler n.d).
While it is clear that an archaeological sample of vessel sherds is unlikely to have ever been
all in such a state at the same time, this does not negate the utility of the concept as a
target population for estimation procedures (Figure 5) (see also De Boer 1983). Points
to note are that the “behavioural assemblage” is regarded as equivalent or closely related
to Schiffer's “systemic context” (Schiffer 1995a) rather than being a life assemblage in the
palaeoecological sense of a fossilized tableau of organisms in life position (Brenchley &
Harper 1998:69), such as might be seen in parts of Pompeii.
Felgate and Bickler introduced the term “breakage population” as roughly
analogous to DeBoer's use of the palaeoecological term “death assemblage” and draw a
distinction between the breakage population and DeBoer's “discard assemblage” or
Schiffer’s “discard flow”. The breakage population is defined as a theoretical time-
accumulated set of the broken, but complete, pottery vessels that form the parent vessel
population of a sample of potsherds from an archaeological context. This definition
assumes that the use-life of a vessel ends upon breakage, but is unaffected by re-use of
some portions of the vessel as a secondary artefact or recycling into grog temper. DeBoer
(1983) sometimes refers to a “total discard assemblage” which would seem to be
40

effectively equivalent to the breakage population.
Use-life seems to be the most important factor linking composition of behavioural
assemblages with breakage populations (David & Wilson 1999, de Barros 1982, Mills
1989, Shott 1989b, 1996). The relationship between behavioural assemblages and discard
assemblages is more complex than that between behavioural assemblages and breakage
populations, involving a number of factors in addition to use life, such as discard practices
and re-use of fragments. By separating breakage and discard, a parent population is
modeled in the formation of the archaeological record in which artefacts can exist beyond
their use-life, but in a complete state.
The composition of this breakage population has been structured by use-life, but
with vessel completeness at 100%, unaffected by any re-use of sherds, etc. No statement
regarding the degree of brokenness of this population is implied here beyond the
expectation that the vessel is cracked beyond use or broken into any number of sherds.
Comparison of average vessel completeness in the archaeologist’s sample with average
vessel completeness in the breakage population (the latter is by definition 100%) is the key
to estimating the breakage population.
It is worth reiterating that the target population in the estimation approach of
Felgate and Bickler is the breakage population, rather than any of the other “stages” in the
model of formation given in Figure 5. The estimate obtained does not inform directly on
the extent of the archaeologist's sampling of the extant deposit, or on taphonomic
transforms, nor does such an estimate provide information on use-life factors as these
transform a behavioural assemblage into a breakage population.
To show how vessel completeness can be used to estimate a breakage population,
41

Felgate and Bickler turn to an explanation of similar estimation in faunal analysis. Orton
explains:
“The formula was based on the idea of matching pairs of bone
elements; in a population, there are equal numbers of left and right
elements. As the population is reduced by sampling effects, some
left and some right elements are removed, leaving unmatched left
elements, unmatched right elements, and matching pairs. The
smaller the sampling fraction, the smaller the proportion of
matching pairs is likely to be. The numbers of unmatched left and
right elements, and of matching pairs, can thus be used to estimate
the original size of the population, given assumptions about the
nature of the taphonomic process (Krantz 1968:215). (Orton
2000:54).
Pots do not break naturally into matching pairs, unlike skeletons, but the model of sample
formation described by Orton can be expanded to any archaeological fragmented materials
in which essential classes exist, like collections of potsherds, in which essential vessel
classes must exist (even though these may be difficult or impossible to identify in many
cases). In general terms, the model of sample formation described by Orton can be
rephrased as the problem of estimating the number of species or classes in a population,
based on the properties of the sample.
This is not just an archaeological problem; nonparametric statistical methods exist
for solutions applied to a wide range of such estimation problems. Felgate and Bickler refer
to the vast literature on capture-recapture studies where capturing and recapturing of
tagged animals can be used to estimate populations of animal species in a landscape (see
Seber 1986 for a review). A review of the related problem of the estimation of number of
species/classes in a population (Bunge & Fitzpatrick 1993) noted a range of working
solutions to the “number of species” problem, and noted prospects for improvements in
this regard in the future (see also Solow 1994 for a Bayesian approach).
The question of the types of bias likely to be present in samples becomes important
in the context of samples comprising mostly singleton sherds. This is one of the questions
42

investigated in Chapter 7. The other major question raised by the analysis in Chapter 5,
also addressed in Chapter 7, is, “What happened to the pottery that didn’t make it into our
sample?”
For seriation, if it can be shown through an analysis of vessel completeness that
most decorated sherds are likely to be singletons, or independent observations of the nature
of vessels in the breakage population, then sherd count is a useful measure of sample size
and relative abundance of attributes. An argument will be put forward in Chapter 5 to
suggest that this is the case for the Roviana early ceramic samples from the sea. The same
argument is applicable to the calculation of sample sizes. If decorated sherds are mostly
independent observation of vessel properties, then sherd counts of decorative and form
attributes are good indications of the number of vessels represented in the samples. This
is important for seriation analysis, in which, as will be detailed in the following section,
attribute sample sizes weight the analysis and affect the outcome.
Seriation Theory : a Review:
Seriation is a term in use principally in the fields of psychology and archaeology. In
psychology it refers to the cognitive ability to order objects based on similarity, but in
archaeology the meaning is more specific. When archaeologists talk about seriation they
usually mean (or should mean) the creation of a temporal series based on similarity and
sometimes evolutionary hypotheses.
Types of Seriation:
A recent taxonomy of seriation techniques divides seriation into similarity approaches, and
evolutionary approaches, the latter based on a rule of evolutionary development.
43

Figure 5: Processes of formation of the Archaeological record (After
Felgate and Bickler n.d., adapted from De Boer 1983): inferring a
breakage population from an archaeological sample.
Similarity approaches are further subdivided into frequency seriation, occurrence seriation
and phyletic seriation (O'Brien & Lyman 2000:64). All of these approaches have been
applied to Lapita, in various mixtures, with little explicit awareness or discussion of the
differences between them.
Evolutionary seriation is one in which a constant direction of change or an
44

evolutionary rule of development is specified. Summerhayes’ statement, for example, that
his stratigraphic data identifies the direction of change (Summerhayes 2000a:151) has an
implicit evolutionary component.
Seriation involves firstly description of ceramic assemblages in terms of criteria
(types or attributes) that are thought to vary over time (Cowgill 1972, Duff 1996, Dunnell
1970, Lyman et al. 1998, Marquardt 1978, O'Brien & Lyman 2000, Shennan 1997:342).
If a randomly selected set of descriptive attributes is used rather than a set selected as
representing temporal change, an “...inseparable hodgepodge...” of geographic, functional
and temporal variation might result (Dunnell 1970:310). (See also Braun 1983:113 on the
confusion of functional and decorative variation, and the additional factor of the
relationship between vessel size and decoration). While form/function/technological
attributes may vary over time (Rice 1982:48), these should only be used in seriation if
independent evidence is available to show that this was the case. Also, pottery production
styles can be expected to vary across space. Braun also notes the need, given different use
and breakage rates in different contexts, to seriate comparable contexts of deposition, and
the need to control for vessel part in treating decorative variability (Braun 1983).
Approaches to the identification of chronological variability range from intuitive
examination in combination sometimes with the direct historical method of working back
from the recent past as in Kroeber’s original formulation (O'Brien & Lyman 2000:111-
114), to sophisticated computer techniques of data exploration. A simple method,
sometimes used by archaeologists in the culture historical era, was to control for functional
variation by performing seriation on a single vessel form, and to control for spatial
variation by limiting the seriation to sites from a small region, regarding the remaining
variation as temporal variation in style.
Various people have pointed out that seriation is best carried out on sites that
represent a short occupation, or on sites with similar occupation spans (O'Brien & Lyman
2000:117). Marquardt (1978) takes a more relaxed stance on this issue than Dunnell
45

(1970). Ideally one would like large samples from short, well-preserved occupations, and
these may well be conditions which are met in the archaeological record in some parts of
the world. This has yet to be securely demonstrated in the Oceanic Lapita examples
reviewed in Chapter 2.
In traditional graphical frequency seriation, assemblages are graphed as relative
frequencies summing to 100%. Bars representing the relative proportions of selected types
or attributes are sorted until unimodal “battleship curves” are produced, usually interpreted
as representing the waxing and waning popularity through time of items meeting the
classification criteria. (Marquardt 1978:261). This “popularity principle” is often
incorrectly stated as an assumption of the technique. Other interpretations are possible,
relating to deposition similarity rather than directly to production similarity and popularity
(Goldmann 1971). Le Blanc notes in regard to this “popularity principle”:
“Even if... changes were for all practical purposes instantaneous, the
archaeological record of these changes will almost invariably show them as
gradual with periods of mixed technology or style (Le Blanc 1975:23).”
McNutt shows that while temporal ordering of sites using graphical type frequency
seriation may or may not produce a correct ordering, to interpret battleship curves as
popularity curves is to misinterpret the seriation schema as reflecting absolute abundance
of types over time. (McNutt 1973:51, 58). McNutt considers frequent errors of this sort
to be due to casual use of the word “frequency”(relative abundance might be a better
term).
Dunnell points out that the extent to which types or attributes form such battleship
curves can be used in dialogue with reclassification to develop a classificatory scheme
(Dunnell 1970:309). Selecting attributes which form battleship curves and produce a neat
seriation is key in Dunell’s view to identifying those aspects of variability which are
stylistic.
The most widespread seriation technique in the 70s and 80s was to construct an
assemblage similarity matrix using attribute/type counts (converted to relative frequencies
46

summing to 100% for all units), and the Robinson coefficient of similarity. It is important
to note that there is no control or weighting for assemblage sample size built into the
resulting similarity matrix, and that, as for graphical frequency seriation, the resulting
pattern of site similarity may contain spurious observations from mixed contexts or small
samples. These may be less obvious than in graphical frequency seriation (where the
offending units would be unlikely to fit well anywhere in the series). Fortunately methods
exist for evaluating the goodness of fit of a matrix sort using this sort of data, and seriation
using Robinson coefficient of similarity, followed by matrix sorting or cluster analysis has
been successful in many instances.
Lack of fit in the sort order can be as a result of inadequate sample size, or
representation of an anomalous occupation span, possibly of material from temporally
separate occupations, but not necessarily so (de Barros 1982, Dunnell 1970: 312). The
trouble in Oceania is that usually, due to poor sampling, one must put up with a certain
jaggedness in the seriation or wind up with no seriation at all, the corpus of sites for the
Lapita period at least being too small to allow fussiness in most cases, when dealing with
Lapita sites from a small area.
McNutt considered that both graphical type seriation and similarity matrix
approaches such as Robinson matrices should be regarded as techniques that work
sometimes rather than as having any methodological validity, and bemoaned the lack of
caution in their application by the 1970s in contrast to a commended caution displayed by
earlier practitioners (McNutt 1973:58-60). These comments, although not directed at
Oceania similarity matrix-based seriations, seem pertinent here, as will be argued with
detailed reference to specific examples in Chapter 2.
The predominant seriation technique in archaeology today, especially in Europe,
is Correspondence Analysis (CA), a multivariate data-exploratory technique suitable for
presence absence data or counts. When used for counts, some control for sample size is
present, as both attribute/type total sample size and the total count for each unit are used
47

to weight data. Attribute/type covariation is controlled for by calculating summary
variables, which are easily plotted in two or three dimensions. Diagnostic information
showing the contribution of various attributes/types and units to the summary variables is
available (unlike cluster analyses of similarity matrices). As for graphical frequency
seriation and Robinson seriation, if non-chronological variation is input (e.g. spatial and
functional variation) the summary variables obtained may be strongly influenced by these.
Control for spatial variation is possible as input files can include geographic coordinates,
and diagnostic statistics make it clear to what extent variation across space is structuring
the seriation.
Multivariate ordination approaches such as principle components analysis (PCA),
correspondence analysis(CA), principal coordinates analysis and multidimensional
scaling(MDS) have percolated into Oceanian ceramic analysis over the last decade. PCA
and CA are generally more easily evaluated than cluster analyses (Shennan 1997:253, 297-
298). Ordination techniques in general give an indication of the relationships between
ceramic variables and units of analysis (e.g. site assemblages). They can suggest whether
there are any summary trends in the data. PCA is generally more suited to numeric data,
while CA is better for counted data, although PCA is sometimes used for this purpose.
Seriation is nowadays almost always done using CA (Shennan, 1997 342), on counts of
types/attributes in the case of ceramic refuse, and on presence-absence of features in the
case of grave-goods or architecture. CA as available in the Bonn Archaeological Software
Package (BASP or WinBASP on the internet) can output comprehensive diagnostic
statistics to identify the relative contributions of the various variables (including geographic
coordinates of the samples) and units to the results.
Principal coordinates analysis and MDS attempt to find the structure in a set of
proximity measures between objects. Results from these analyses are not as easy to
evaluate as those from PCA or CA, as they operate on a similarity matrix rather than on
raw data or data that has been transformed or standardized in some way (Shennan
48

1997:348-349). A key part of the MDS method is that it has the advantage of providing
an objective indication of the number of major summary trends that are significant, through
the measure stress. One advantage of retaining the use of a similarity matrix is that a mix
of metric and nominal variables can be used to construct the matrix, particularly useful if
one has metric data on vessel form or the scale of decoration.
Occurrence Seriation and Potsherd Samples:
Occurrence seriation in its graphical form arranges samples so that the occurrence of
criteria is continuous or as near so as possible. The usual interpretation is that this
represents an ordering of assemblages in time, assuming that any particular attribute was
continuously discarded through time. To be valid this requires, like frequency seriation,
that the attributes used in the seriation do not have an intermittent occurrence, and assumes
control for contemporaneous variation across space and in relation to vessel/site function.
Green’s seriation of Lapita sites (Green 1978) and Graves’ and Cachola-Abad’s
seriation of architectural features in Hawaii (Graves & Cachola-Abad 1996) are of this
type (O'Brien & Lyman 2000:120-121), but there are fundamental differences between the
data used in these two examples which have serious consequences for ceramic seriation.
A tradition of occurrence-based assemblage grouping approaches to ceramic seriation has
been a persistent thread in Oceanic pottery analysis, usually using the Jaccard coefficient
or simple-matching coefficient to create similarity matrices (Anson 1983, Best 1984,
Egloff 1971, Frost 1974, Green 1978, Irwin 1972, 1985, Summerhayes 2000a, Wickler
1995, 2001). In the wider archaeological world occurrence seriation has been largely
avoided by ceramicists since Dempsey and Baumhoffs initial proposal (O'Brien & Lyman
2000:119), except as a technique for ordering grave-lots. The reasons for this avoidance
are firstly that “chronologies based on the presence-absence of types or attributes will
never be as accurate as those based on their relative frequencies (Le Blanc 1975).” Le
49

Blanc noted the applicability of presence-absence seriation to grave-lots, where
assemblages represent small intentional sets of attributes rather than time-accumulated
discard from settlements. Secondly, and perhaps more importantly, the notion that
occurrence seriation is less sensitive to variation in sample size than frequency seriation is
questionable. Sparse data (a lot of types/attributes, each with few members) is where
occurrence seriation shines, hence use with grave lots and architectural features. This
assumes though that the sparse data is a complete representation of the unit to be
seriated. This point will be more fully discussed below in relation to some archaeological
seriations in Oceania.
In Oceania, for the Lapita period, we have yet to discover a landscape dotted with
generous numbers of Pompeii-like yet easily detectable and accessible ceramic vessel
samples, and must be satisfied with a less stringent set of limits within which seriation can
be considered to work. Regional site samples can be expected to be on the small side of
adequate, with assemblage vessel numbers lower than might be ideal, and durations of
occupation largely guessed at rather than tightly defined. Compounding these preservation
and mixing problems is the problem of sparse data (discussed above), where in some
classificatory schemes, a large number of types/attributes exist, each with few occurrences,
in only a subset of sites. This sparseness may arise through splitting approaches to
classification/grouping, in combination with the nature of ceramic production (lack of
standardization), and, also, as a result of the sampling processes acting to make the
archaeologist’s sample a fraction of the pottery discarded at the location in the past in
many cases.
In response to the problem of sparse data, an approach commonly taken in
Oceania has been to apply a form of occurrence seriation, by recording the
presence/absence of attributes in units/sites, and by creating a matrix of similarities between
units with a similarity coefficient suited to this type of data, and then performing either a
computerized sort or a cluster analysis on the similarity matrix. The two similarity
50

coefficients commonly used on this sort of data are the simple matching coefficient and the
Jaccard coefficient. Of these, the simple matching coefficient regards negative matches as
significant, and is thus less suited to sparse data, while the Jaccard coefficient disregards
negative matches entirely, and is therefore better for this type of data (Shennan 1997:228).
The widespread application of these techniques in Oceanian archaeology pays
insufficient attention to the sample-richness problem for this type of sparse occurrence data
from sherd samples (but see Anson 1987, Green 1978, Kirch 1987a). Whether or not a rare
attribute occurs in an assemblage of sherds may well be a sampling effect, even for
presence-absence data. Where sites/units vary greatly in sample size, spurious dissimilarity
may result through size-related differences in sample richness.
Consider a hypothetical case (Table 1), of two assemblages are drawn from
identical populations, differing only in sample size. All types are represented in the large
sample, but many are missing from the small. Calculation of a Jaccard coefficient of
similarity would result in the following data (where a=positive matches, b= present in the
small assemblage but not the large, c=present in the large assemblage but not the small):
a=3; b=7; c=0. Jaccard coefficient is calculated as a÷(a+b+c) or 3÷(3+7+0)=0.3, when
ideally these being samples from the same thing should show high similarity (s
approaching 1.0). In this example sample size is structuring the similarity coefficient to a
greater degree than behavioural similarity. It can be seen that although the Jaccard
coefficient is suited to sparse data, it assumes a complete representation of behaviour in
the sparse data, and is not suited to assemblages which comprise incomplete samples of
a behavioural unit. Using a technique adapted to sparse data does not correct for
inadequate sample of sites or inadequate (or unsaturated cf. Kintigh 1984) within-site /
unit assemblage size (see also Anson 1983:54, Kirch 1987a, Specht 1977). This example
illustrates a major potential problem with the application of the Jaccard coefficient to
ocurrence seriation of Oceanic ceramic data, which would not arise with data from grave
51

lots or architectural features.
Phyletic Seriation:
Phyletic seriation assumes heritable continuity, without specifying a rule of developmental
direction, by emphasising evolutionary change of a chosen attribute over time. Seriation
of Lapita face designs for example (Spriggs 1990) involves an assumption of the evolution
of a single design over time, although Spriggs tended more towards evolutionary seriation
in that instance as the direction of change was assumed to be from initial representation
of faces to a later stylized geometric remnant. A more recent example of phyletic seriation
of Lapita designs is more sophisticated, in that it constructs multi-lineal evolutionary series
for three different aspects of Lapita design (Ishimura 2002). Phyletic seriation is about how
change occurs, how one attribute becomes another, where frequency or occurrence
seriation use arbitrarily defined attribute classes to freeze-frame change for measurement
purposes. Mead also discussed phyletics of motif relationships by speculating that some
motifs were developments of or elaborations of others (Mead 1975:37)
Table 1: Sampling, assemblage richness, and the Jaccard coefficient (p=present, a=absent).
Attribute 1 2 3 4 5 6 7 8 9 10
Small
sample
Large
Parent
Popn.
p a a p a a p a a a
p p p p p p p p p p
52

Seriation in Oceania:
The best examples of local regional sequence building in Oceania have at their cores a large
sample of collection sites, representing a good percentage of the total ceramic diversity
within a small survey region. Additionally, they all use graphical attribute seriations and
stratigraphic superposition to check the computer matrix results. Most, but not all, use
ordered similarity matrices for their seriations, based on attribute frequencies and a
Robinson coefficient of similarity.
The Frost/Irwin numerical taxonomy approach was typified by a similarity grouping
analytical mode. In analysing his Shortland data, for example, Irwin adopted Clarke’s
proposition (Clarke 1968:512-549) that there is a functional convergence between
occupation sites that imposes some control on functional variation, and that relative
similarity between archaeological assemblages therefore indicates closeness in time (Irwin
1972:83-85) (This assumes invariant site function and complete representation of the
properties of a breakage population). Thus if a similarity matrix is ordered so that an ideal
pattern emerges, the sites are arranged in order of chronological distance. Irwin noted the
possibility of contemporaneous spatial variability creating such a pattern in the similarity
matrix, but made no specific mention of functional variability, which, as vessel form data
were included, could be seen as problematic. Irwin felt that because site surface collections
were quite large and of “not especially long duration” (he had culled them of material
thought to be intrusive from different periods prior to analysis) there was little likelihood
of anything other than temporal variation emerging in his patterning.
Irwin, citing Frost’s Fiji work, used the simple matching, Jaccard and Robinson
coefficients of similarity to perform sorted similarity matrix seriations. The first two
coefficients were used with occurrence data, while the Robinson coefficient is a
frequency-based measure of similarity. Green (citing Irwin) suggested that occurrence
seriation using the Jaccard coefficient was beneficial in reducing sample size differences
(Green 1978), but this is incorrect: the technique is suitable for the analysis of sparse data
53

(discussed in more detail below),which is not the same thing. The use of these two
different types of seriation on the same samples has been widely followed in Pacific
archaeology (Anson 1983, Best 1984, Green 1978, Wickler 2001). These techniques
commonly, and not unexpectedly, produce conflicting ordering of sites in the resulting
similarity matrices, requiring a decision as to the “correct” ordering, and encouraging a
variety of matrix sorting or clustering techniques, all of which can be decidedly divorced
from the original archaeological data, and difficult to evaluate.
Irwin’s study of a large number of late-prehistoric sites within the local Mailu
region analyzed 28 contextual assemblages, some from excavation zones, some from
surface collections, and constructed a sequence using sorted similarity matrices, graphical
attribute frequency seriation and superposition chronological inference (Irwin 1985:118-
162). Similarly, Egloff made use of attribute-frequency sorted similarity matrices of a large
number of surface sites and some mound excavations in the Collingwood Bay region, and
also made use of graphical attribute seriations (Egloff 1979:47, Figs 17-19). Egloff’s
sequence ran from the undated Lapita-like “Group P” ceramics to modern. Best applied
Irwin’s method to construct a ceramic sequence for Lakeba in Fiji running from Lapita to
modern. Both Best and Irwin controlled for sample size in their seriations by omitting
smaller assemblages (under 100 sherds in Irwin’s case), but neither discuss quantification
of sample size in any depth, or methods by which mixed or long-occupation assemblages
can be identified, an area in which both Specht and Bedford have pointed out the value of
stylistically homogenous sampling units (Bedford 2000, Specht 1969). In Bedford’s case
it is arguable that some of the lack of diversity is related to small sample sizes in the lower
levels of rockshelter test pits on Malekula, while Specht was referring to larger samples
surface-collected from abandoned village sites on Buka.
Best, like Irwin, used attribute frequencies to calculate a Robinson matrix of
similarity, and cluster analysis to create a seriation of excavation and surface collection
units (Best 2002:18-20). A particular feature worthy of note is the inclusion in his
54

seriation diagram of a “dead” column tracking the percent decorated, which shows six
peaks of decoration through his series, rather than a unimodal history, illustrating the
dangers of using a recurring or fluctuating attribute for seriation. Best shows sixty units
in his Robinson matrix, and thirty in his graphical seriation, with twenty-three attributes
shown, illustrating the desirability of having a reasonably large matrix of types/attributes
and units for seriation, to lessen the possibility of random repeats of stylistic attribute
combinations over time. It is important to note that in Best’s case and in Irwin’s, that
stratified sequences were available, which aided the identification of temporally sensitive
attributes used in seriations.
Selection of Variables in the Absence of Stratified Sequences:
How best to structure classificatory systems to meet chronological aims is seldom
discussed in archaeology generally. Some attributes/variables may fluctuate over time and
space (the relative percentages of decorated vessels versus undecorated vessels for
example), or may vary in relation to function (vessel form for example) which may in turn
affect decoration (Shepard 1963:260-261). The danger in analyses that utilize a large
number of variables/attributes in an a-theoretical manner is that the signal they are after
(temporal variation for example) will be obscured by other types of variation, such as
functional, spatial or sample-size variation. These problems are lessened in CA and PCA,
where it is relatively easy to discover if spatial or sample-size variables are contributing to
the results, but the golden rule in seriation analysis according to Dunnell is careful selection
of types or attributes that vary with time (Dunnell 1970). Some sorts of attributes carry
higher risk than others, and can usefully be omitted from analysis when the aim is
chronological. It may be possible as a result of such an omission to show, eventually, that
such variables do vary over time, but the converse approach, including all variables, may
result in a mishmash of variability that might be difficult to disentangle, particularly for
similarity matrix / clustering approaches in which it may be difficult to ascertain which
55

variables are principally responsible for the series obtained.
Just what is it that one’s units of classification are measuring? This is one among
a number of reasons why validation of cluster analyses of similarity scores is problematic.
The Mead/Donovan and Siorat approaches have such theory, and seek ideas of the
decorators regarding the structure (and sequence in the case of Siorat) of decoration
(Mead 1975:37, Siorat 1990). In regard to the Mead/Donovan and Siorat approaches, if
past ideas of structure is their objective, one must question whether there is likely to be a
strong shift in such structures over time alone, rather than over space or in relation to
functional variability. Clark also provides some discussion of the theoretical basis of his
attribute selection (Clark 1999:65), and Wickler and Anson are both forthright on the
difficulties of separating functional and temporal variability. Summerhayes advances a
specific theory of ceramic temporal-functional variability (Summerhayes 2000c),
concluding that function varied over time in the West New Britain data.
The review of seriation method and theory above sets the stage for a critical review
of constructions of Lapita temporal variability in Chapter 2, and for the analysis of
variability and seriation of Roviana data in this thesis. Frequency seriation, using attribute
abundances and correspondence analysis, has the potential to yield a high-resolution
chronology if samples are fit for the task and time-sensitive attributes are chosen.
Correspondence analysis negates the argument that occurrence seriation is better for
unsaturated samples, since correspondence analysis is weighted by site/attribute sample
size. The quantification issue raises its head here though, as sample size must be assessed
using a measure which is suited to the completeness characteristics of the sample.
Occurrence seriation is unlikely to match the chronological resolution of frequency
seriation for potsherd samples, being more suited to sparse data such as grave goods or
architectural features, where the number of attributes of a sampling unit are large and the
membership of each type/attribute is small, and where the sparse data is a complete
representation of each unit such as a set of excavated grave goods, rather than a sample.
56

I have argued above for the importance of sample evaluation as a first step in
seriation: sample sizes for sites and units are one side of this issue (including the difficult
problem of the number of independent observations that comprises the sample), but
another difficult area is the assumption of heritable and historical continuity in the overall
sample. It is important to try to identify sampling gaps or lack of heritable continuity
between units or types. Phyletic analysis has a role to play here. Phyletic seriation is a
mode of analysis that can usefully be combined with frequency seriation. When a series is
constructed, phyletic analysis allows the investigation of the nature of the transformations
that occur to make the series. How does one attribute transform into another? Phyletic
analysis in turn leads on to “why” questions: if the nature of change can be understood, the
explanation of change can be investigated. In Chapter 12 such a combination of frequency
seriation (using CA) and phyletic seriation is attempted.
This review has noted the importance of identifying the sorts of attributes in the
data (whether functional or stylistic, within a Dunnellian theoretical framework- explained
in the following section). The link between site/artefact properties and function is by no
means simple, and further review of this topic is warranted.
Form and Function: a Review :
Introduction:
Dunnell turned to evolutionary theory to explain why seriation worked, suggesting that
style and function were fundamentally dichotomous (Dunnell 1978). In this view style is
by definition that which is subject to evolutionary drift, while function is that which is
controlled by selection (Dunnell 2001). Functional variation is thus a poor tool for
seriation, as function may be stable or may fluctuate over time with the appearance of
analogous similarities (adaptations or evolutionary convergences), where style, by
contrast, varies stochastically over time by definition. In Dunnell’s definition of style and
57

function, form can have functional and/or stylistic components, as can decoration.
Separating stylistic variation from functional variation is simple in Dunnell’s view: stylistic
variation will seriate well (in battleship curves, in the traditional graphical form of
frequency seriation), while functional variation will not. Dunnell felt this explained why
proponents of seriation in the culture-historical era had been able to create successful
seriations through trial and error, trialling various classifications in a materialist manner
until the desired battleship pattern (a seriation) was obtained, and successfully isolating
chronological variation by this means (in conjunction with testing by other forms of
dating).
Dunnell’s “method” (my term) assumes that sample sizes from sites, the number
of sites in the sample, and the level of comparability of site occupation spans will be
sufficient to create an even battleship pattern (with exclusion of sites that don’t fit due to
small sample sizes or anomalous occupation span). As stated above, this is seldom the case
in Oceania: neither can the Roviana data support this assumption, as will be seen in
Chapter 12. Given a patchy archaeological sample, other methods of distinguishing
functional variation from stylistic, prior to seriation, are needed.
Techno-function refers to utilitarian aspects of an artefact’s use, as opposed to
sociofunction or ideofunction (Skibo 1992:33-34). Skibo makes a further division of
function into intended function and actual function, with intended function being reflected
in vessel design, and actual function reflected in use alteration. Theories of vessel
form/function correlation (Rice 1987: 211, Smith 1985) thus offer an opportunity to survey
the total Roviana assemblage from the perspective of intended technofunction. If intended
vessel technofunction can be attributed to ceramic forms, then function can be controlled
for to some extent by constructing seriations only within functional classes. The situation
is complicated in archaeology by the fragmentary state of recovered vessels, from which
forms are inferred.
The evolutionary concept of exaptation (Gould & Vbra 1982) could usefully be
58

applied to the relationship between pottery form and function, allowing vessels of
multipurpose forms, and allowing uses to which the form is adaptable, but not adapted. It
might be argued that there is no fixed vessel function, that artefact function is diverse and
unpredictable from form and therefore determination of vessel function is an impossibility,
but there are a number of factors working in the archaeologist’s favour that generate the
expectation of a correlation at least between form and function. Firstly, a substantial
ethnoarchaeological literature has appeared in the last twenty years on vessel use life, and
suggests that frequency of use and type of use correlate strongly with use life (David 1972,
De Boer 1983, Jensen et al. 1999, Longacre 1991, Mills 1989, Shott 1989b, 1996, Tani
& Longacre 1999, Varien & Mills 1997, Varien & Potter 1997). The more frequently a
vessel is used the more likely it is to be broken. Cooking and serving are a frequent
activities, and representative settlement ceramic assemblages are likely to be dominated by
utilitarian cooking and serving vessels (Varien & Mills 1997:147). Serving vessels, used
frequently but not subjected to thermal stress are likely to have longer use life, and be less
common in assemblages than would be the case if they were subjected to thermal stress,
while storage and ritual vessels are likely to have even longer use lives, and be less
common in assemblages, even if present in abundance during settlement occupation in the
past.
This formational bias towards cooking and serving vessels in assemblages (Varien
& Mills 1997:148), is good news for seriation, as this bias creates for the archaeologist a
measure of control for function, provided the sample obtained is representative of the
settlement site as a whole (likely in the case of surface scatters exposed over a wide area,
but more difficult to support in the event that one small excavation square is taken as
representative of the whole, or if “site” function varies between sites and not all sites are
“occupations”).
This does not negate the need for ascribing function to sherds. One needs to know
what sorts of vessel forms one has in the assemblage to be confident that such a bias does
59

in fact exist in any particular sample, and further, it would be useful to be able to exclude
some observations relating to non-cooking/serving vessels from consideration where these
are identifiable.
A use alteration perspective suggests that cooking pots are arguably more easily
distinguished from other vessel classes in archaeological assemblages based on cracking,
sooting and abrasion resulting directly from their different use environments (Skibo 1992).
This is clearly not the case for the Roviana assemblages, where sooting was noted only on
one vessel, (parts of which were found buried in mud), while for the majority of sherds the
vessel surface was either clean or ablated, probably a consequence of the intertidal, surface
context of deposition. For this reason vessel form seemed to be the best available
information about vessel function. Nor has a use alteration perspective been applied to
Lapita pottery in any of the studies reviewed in this chapter. Sooting, at least, in Oceania,
seems hard to find.
In the absence of use-alteration information, how might a classification of
sherd/vessel form according to vessel function be achieved? A number of measures of form
properties of ceramic vessels relating to vessel function have been suggested or
investigated by archaeologists, and a range of technical issues have surfaced in relation to
the measurement of vessel form, examples of which are given below.
The Culture History Approach of the 1940s and Onward:
In southwestern American archaeology, often the distinction is merely between jars, for
cooking and storage, and bowls, being serving and eating vessels (Skibo 1992:36), an
unsatisfactory generalization without much theoretical/methodological rigour.
“The notion that use strongly influences pot morphology is venerable, but
few archaeologists or ethnographers have examined it in a uniformitarian
theoretical framework (Smith 1985:257).“
Rice reviews several “inferred use” classificatory schemes, including height-to-width
vessel proportions, vessel contour classifications, and geometric/volume classifications
60

(Rice 1987: 215-222). All of these schemes place a heavy burden on sherd size, which
needs to be large to reconstruct forms, although Rice’s own approach, a vessel contour
approach, is less restricted by this factor than either the vessel proportion approach or a
geometric/volume classification.
The Methodological Uniformitarians:
As archaeologists we would like to be able to specify a deterministic link between form and
function, but this is not possible. Rye for example states,
“If it can be established that specific materials are usually correlated with
specific functions, the presence of these materials can be used to infer
function (Rye 1981)”
Rye goes on to detail ceramic properties appropriate to different functions. He seems to
be suggesting that he is providing a manual or method for the identification/determination
of vessel function, and yet the information cannot be used in this way in the absence of a
clear causal relationship. For example, the proposition that rounded forms improve thermal
shock resistance is not matched by any statement that rounded forms indicate cooking as
an intended function. There are examples in the ethnoarchaeological literature, too, that
suggest flat-bottom vessels are suited to cooking functions also (e.g De Boer 1983: 109,
Rice 1987:224-226).
A number of studies have attempted to develop methodological uniformitarian
predictors of use based on ethnographic data (Henrickson 1990, Henrickson & Mcdonald
1985, Smith 1983, 1985). These provide cautionary documentation of a complex situation,
where some classes of vessel use are relatively easily identified for whole vessels (e.g. dry
storage: Smith) but where identification of other functions based on morphology is
problematic, even at the level of whole vessels, and more so when dealing with broken
sherd assemblages.
Smith’s concern was solely with morphological indicators of use, in contrast to the
61

more common technical and use-alteration approaches. Smith used Binford’s distinction
between technomic, sociotechnic and ideotechnic aspects of culture (Binford 1962) to
argue that a narrow focus on technomic aspects of ceramic function was possible without
being diverted by this wider debate on function and style. Smith proposed no
comprehensive theory of ceramic vessel form: it was noted that many factors besides use
determine form. Smith allowed pots could have multiple uses, and specified also a
distinction between morphological variables of whole vessels (e.g. vessel volume, height
to width ratio, etc) and sherd variables, the latter being properties describing vessel size
and shape at a single point, using an assumption of vessel symmetry.
Smith’s sample of 39 ethnographic pots, for which eight ethnographic use classes
were listed, was drawn from six ethnoarchaeological/ethnographic studies of the American
southwest, which raises obvious sampling issues; her subsequent dissertation incorporated
an enlarged and geographically/culturally more diverse sample (Smith 1983:116-140). The
small sample did not allow detailed correlations of different factors of use with specific
morphological features. Her use classes were cooking; dry storage; wet storage;
processing; serving of food; individual consumption; transport of liquid; and washing
within the pot. Morphological data were reconstructed from illustrations.
Smith digitized pot profiles with the aid of a shadow device, and encoded these as
polynomial expressions of curves using the GLM regression procedure of SAS.
Discriminant analysis was used to evaluate which variables were of use in defining
function, and which sherd statistics (sherds statistics used experimentally were derived
from the form data of the same vessel sample by a computer simulation of breakage).
Predictive success was best when broader functional categories was used, which
cut out instances of multiple-use (her short list was cooking, dry storage, wet storage or
transport of liquids, and other). A clear distinction could be made between dry-storage
vessels (low exposed surface per volume unit and small mouths, with large volumes
characteristic to a lesser extent) and other vessels of all types. Consumption vessels
62

(bowls) were well discriminated (large orifice relative to capacity, large orifice in absolute
terms, and small volume). Processing, serving and washing vessels (other) had a
moderately large ratio of exposed surface to capacity and moderately large orifices, with
a slight tendency towards small volumes. Cookpots showed more variation than any other
category, suggesting a category ill-defined or too broad, or that crucial morphological
constraints were not monitored by the variables of the study. Cookpots overlapped
significantly with liquid storage and liquid transport samples, with the liquid storage and
transport vessels tending to have small ratios of orifice to volume, with these vessels not
displaying the expected provisions for closure (Smith 1985: correlate 9 on p265). This last
finding will be shown to be significant to the Roviana case in Chapter 8.
Smith concluded that food storage and individual eating should be activities with
reasonable archaeological visibility (without considering breakage rates). She noted the
need to develop this method in the direction of archaeological practice, where some of the
orientation variables are difficult or impossible to measure on broken sherds. Her efforts
in this direction are promising, but preliminary, and are a long way from establishing a
uniformitarian theory. Her results do suggest caution is required in the hasty assignment
of globular jars and globular jars only to cooking functions and broader, shallower vessels
to serving functions. Along with her finding that liquid storage/transport vessels had larger
orifices than expected, this suggests that the cooking/serving/water jar distinction often
seen in ethnographic terminology (e.g. Kirch/Green and Ross, discussed below) need not
have a clear morphological correlation (see Rice 1987: figure 7.14 for a similar point in
relation to cooking pot morphology).
Henricksen and McDonald (Henrickson 1990, Henrickson & Mcdonald 1985)
discriminated between function in a general sense (undefined) and primary function (also
undefined). No distinction was drawn between intended function and actual function. The
approach to the discrimination of function was less systematic than Smith's study:
discriminant analysis was not used. As in Smith's study, samples were small.
63

Morphological correlates of function were found to vary according to factors not available
to the archaeologist, for example, the mode of transport of water transport vessels had a
profound influence on form.
Some of their statements verge on “ just-so stories” masquerading as explanation
for morphology, for example cooking pots usually had: “A somewhat restricted mouth to
prevent rapid evaporation from boiling foods (Henrickson & Mcdonald 1985:631)”. The
degree of mouth restriction seems unlikely to have any bearing on the rate of evaporation
of the contents, as given a constant energy supply via the vessel body, evaporation can only
be substantially reduced by pressurizing the contents, (as in a pressure-cooker), with orifice
diameter potentially having only a slight effect on the condensation rate of vapour leaving
the vessel. Restricted orifice is more likely to reduce spillage and to increase form strength
of the vessel.
The case studies presented in the article made heavy use of use-alteration to assign
function; problematic if a distinction is drawn between intended and actual function.
Measurement of Vessel Form:
On the subject of vessel form measurement (orifice measurement and other curvature
measurement on ceramic vessels), Plog noted a number of studies suggesting that
measurement error was possibly a significant source of ceramic variability in some
instances, and might obscure significant variation in others (Plog 1985). Plog suggested
a dial-gauge technique for curvature measurement rather than curve-fitting using curvature
templates, although he noted a number of problems with this, most notably the need to
average a number of measurements along a curve to establish curve variability. Curvature
measurements in the Roviana study were made using a custom-made set of brass curve
templates, rather than the usual diameter chart (Chapter 4), which had some advantages,
not least of which was using curve fitting to determine measurement points at profile
corner points such as necks, not easily done using Plog’s dial-gauge method.
64

In the review of the Lapita ceramic series in Chapter 2, vessel form is an element
of some of the chronological propositions critiqued there, and uncertainties in the
measurement of vessel form are pointed out. Those criticisms are self-explanatory, and
therefore methods of vessel form measurement will not be reviewed in detail here.
Similarly, details of measurement methods used in analysis of the Roviana ceramics are
given in detail in Chapter 4.
Spatial Variation in Form:
There is some evidence from ethnoarchaeology that vessel form can vary spatially across
quite short distances, with no functional difference observable. Ethnoarchaeology of
Kalinga pottery identified spatial variation in the form of pottery over about 10km within
a single functional class (rice-cooking pot) (Longacre 1991:106-108). The variation is
similar to the distinction between soft-shouldered globular vessels and hard-shouldered
globular vessels in the Roviana assemblages detailed in Chapters 4 and 8, so it seems that
there can be substantial production variability for the same functional class within a
smallish region. This brings to mind Smith’s caution that vessel form/function categories
be kept broad.
Oceanic Historical Anthropology and Vessel Function:
Green’s substantive uniformitarian theory of Polynesian plainware vessel function, based
on ethnographic observation of material culture of descendant cultures, suggests the
smaller open post-Lapita bowls from Samoa were used for drinking, serving fluid foods,
kava serving, and holding dyes for decorating bark cloth (Green 1974a:129). Medium-
sized bowls could have been used for preparation of barkcloth materials, cooking, and food
preparation. Medium and large bowls were equated with Kava receptacles.
Linguistic terms for small cups and boiling have been reconstructed for Proto-
Polynesian (Kirch & Green 2001:136). Similarly, Ross used lexical reconstruction of
65

Proto-Oceanic terms to infer a Proto Oceanic taxonomy of four functional classes of vessel
(cooking pots, water jars, bowls and frying “pans”, but noted that identification of the
forms of these was a task for archaeologists rather than linguists (Ross 1996). The
simplicity of this scheme conflicts in some respects with expectations for middle-level
societies around the world. The various storage, processing, ritual and transport functions
observed ethnographically in middle-level societies in various parts of the world
(Henrickson & Mcdonald 1985, Smith 1985) suggest Lapita pottery would have had a
more complex range of functional classes of vessel.
Ross’s tentative assignment of Western Lapita form to function has biconical
rounded-base vessels as cooking vessels, with water jars having a narrower orifice
altogether (Ross 1996:figure 2). These are similar to Summerhayes type V vessels
(Summerhayes 2000a:34), Sigatoka type E (Mead et al. 1975:15, figure 1.5) and
Niuatoputapu vessel form 7 (Kirch 1988a:162 and figure 101d, 101e). This is at odds with
Smith's uniformitarian findings, where there was substantial overlap between water jars
and cookpots, suggesting multipurpose forms were possible in this category. Perhaps the
solution here is to regard small-orifice flasks as liquid transport/storage vessels, and larger
biconical forms as potentially multipurpose forms, which could easily fulfil either a
cooking or liquid-transport function.
It has been suggested that highly decorated Lapita ware might be a high-status
exchange item (Kirch 1988b:161, 1997:148), and that the amount of decoration in
assemblages might reflect contemporaneous inter-community social relations (Kirch
1997:147, 143-144). There is something of a conflict between this model of decorated
Lapita as exchange items and another proposed by Kirch of decorated Lapita as a
depiction of an ancestor of an Austronesian house (Kirch 1997: 188-191). While both
models are consistent with the notion of pots being tattooed for display/presentation, pots
as ancestors and pots as exchange items seem mutually exclusive, as house-ancestor
evokes in my mind manufacture by a locally-resident member of the descent-group, rather
66

than an exchange item (but this is a subjective interpretation) and these could be seen as
competing models, or models specific to particular vessel types. An ethnoarchaeology of
the manufacture and exchange of material representations of Austronesian house-ancestors
might clarify this point. At present my objection is not founded on any good data.
Neither of these models of the function of Lapita pottery have been discussed in
relation to formation theory regarding vessel function and breakage rates. If decorated
Lapita is high-status exchange ware with an occasional serving function, breakage rates
could be expected to be low, whereas if the decorated bowls and dishes were in frequent
use breakage rates would be higher, and frequency in assemblages would reflect this.
Similarly, if pots symbolising house-ancestors were curated in a quiet part of the house,
breakage rates could be expected to be extremely low, unless some specific regular
breakage ritual is invoked, and such vessels could be expected to occur only rarely in
assemblages. Use life and discard rated thus provide a test of functional interpretations of
decorated Lapita. High percentages of decorated Lapita pottery in assemblages imply a
high relative breakage rate, or perhaps a historically-particular mechanism of deposition,
such as abandonment, followed by subsequent breakage of a curated behavioural
assemblage of pottery (a life-assemblage) (Brenchley & Harper 1998, De Boer 1983). If
we are to model vessel function for Oceanic pottery, we cannot ignore the link between
vessel function, size or frequency of use and breakage-rates.
Summerhayes summarized widespread evidence for the loss of complex Lapita
forms with dentate decoration, and relatively unchanging persistence of simple
undecorated forms in many areas, seeking explanation for these decoration changes in his
Arawe Islands ceramic sequence (Summerhayes 2000c:293-303). Summerhayes saw the
slow rate of change of the plainwares as related to an “ongoing domestic/utilitarian role”,
with rapid change over time in the level of representation of dentate-ware in assemblages,
and the shape, motif type and production of dentate pottery. I critique Summerhayes'
chronological conclusions in detail in Chapter 2 as being a too-direct reading of
67

chronology from superposition, with insufficient attention to cultural or natural formation
processes.
The high labour-input into dentate decoration was seen by Summerhayes as
suggestive of a high-value or prestige “goods”, with the loss of this characteristic over time
“equating” to a lessening of the social importance of these pots. Summerhayes endorsed
Rathje’s contention that the dentate motifs were Austronesian social/ideological signifiers
(Rathje 2000), extending this by suggesting these functioned in an environment of
continuing interaction between widespread communities, and that loss of these social
signifiers indicates changes in the nature of interaction between communities. This
explanation is similar to Kirch’s prestige goods explanation, and unlike Kirch’s “house-
ancestor” model, in that the “house-ancestor” pattern is based on modern ethnographic
observation and persists until the present across most of the Austronesian world, making
it a poor candidate for a vanished system. Like Kirch, Summerhayes paid no attention to
the link between function, breakage/discard rates, and assemblage composition.
Theories of the Functional Correlates of Rim Form Variation:
Rim form, is a property of vessels relatively easily identified from sherds, and worth
considering in more detail from a functional perspective. Rim form incorporates the
attributes of the neck (in the case of restricted vessels with everted rims), rim and lip,
including the height of rim, degree of eversion/inversion, rim and neck profile curvature,
changes in thickness across the profile and neck orifice size. When tested
ethnoarchaeologically, observable rim form variability may bear only a loose relationship
to use (as suggested by Miller 1985:51-74). Nevertheless, if one wishes to avoid forcing
contemporaneous functional rim form variation into a spurious time series, consideration
of the possible meanings of form variance and some consequent control for rim form
variance in seriation is desirable.
Everted rims can be regarded as multi-functional components of vessels. They may
68

add form strength, reduce spillage (Henrickson & Mcdonald 1985:634), allow a tied cover
for storage or fermentation (Henrickson & Mcdonald 1985:632, Smith 1985:263), aid
pouring (Smith 1985:263) and filling, aid stacking (Miller 1985:63, 70, Skibo 1992:77) and
aid lifting when hot by use of a torque or tongs (Skibo 1992:66, Smith 1985:263).
Reduction of spillage might be best achieved by having a tall vertical or tall inverted rim,
but filling, pouring, stacking and lifting might encourage a more everted form. Heavily
everted and very short rims are probably not the best design for filling, lifting, or
preventing spillage, but may improve functionality for tying covers, pouring and access for
ladling or stirring, and may improve form strength, extending use-life.
In some methods of pot manufacture, rim form may be constrained by the plastic
limits of the clay-temper mix. Slab-built rims or coil-built rims make little demand on the
plasticity of clay during construction. A tall thick, heavily everted rim can be formed from
an arc-shaped slab or by coil building. In contrast, if the entire vessel including the rim is
formed from a single lump of clay by the paddle-and-anvil method, there are potential
difficulties in forming a tall and strongly everted rim, as the taller and more everted the rim
the more the clay must be thinned towards the lip, relative to the neck, potentially resulting
in an overly convergent, and thus weak, rim profile (thick at the neck, thin at the rim). If
a heavily everted rim is desired for some reason from a pot built by the one-piece method,
the rim may have to be short rather than tall to prevent excessive thinning towards the lip,
resulting in a rolled rim form in the extreme case. If a tall rim is desired in a one-piece pot
(to prevent spillage for example) it may not be practical to have a heavily everted form, as
the lip might split during manufacture or become overly fragile with reduced thickness.
Form and Function: Conclusions:
The detailed identification of intended use of ceramic vessels from archaeological
contexts, relying solely on vessel morphology, is not currently well supported by any
69

established body of theory. There are indications from methodological uniformitarian
studies that some broad categories of intended use might be identifiable from sherd
morphology in the future, but there are also strong indications from these studies that form
is determined by other contextual factors in addition to intended use, including pottery-
making tradition. While these rare uniformitarian studies are widely cited as providing a
theory of the morphological correlates of function (for example Sinopoli 1991:84) few
would claim that these correlates are deterministic laws. Some of the morphological
correlates of function suggested by earlier studies (Ericson et al. 1972, Linton 1944) have
not been borne out by uniformitarian research other than in the most general sense.
This does not, however, negate the virtue of controlling for vessel form variation
in seriations, using broad form-categories. While form may vary systematically over time
as a result of change in function, in which case a form-seriation may be justified, this
requires chronology to be established by other, independent means, particularly from
Dunnell’s position, that seriation of functional variation is a contradiction in terms. Where
the chronology is not known, and seriation is to be the primary means of constructing
chronology, control for form is essential, to reduce the possibility of incorrectly seriating
contemporaneous or fluctuating vessel functional variation.
Chapter summary and conclusions:
The Roviana intertidal archaeological record, investigated as a means of assessing whether
Lapita was continuously distributed across Near Oceania or whether there was a huge gap
in the Near-Oceanic Solomon Islands, requires archaeological theory and method
developed for or adapted to the nature and location in the landscape of the material under
study. Understanding temporal variability, always a basic task in the archaeology of
poorly-understood regions, is complicated by the pottery being in the sea. How can this
record be interpreted in behavioural terms? How can a high-resolution chronology be
constructed? How much of a sample is needed? How should sample size be quantified?
70

What are the biases present in the samples? Pottery recovered as surface scatters from the
intertidal and reef flat raises a series of questions regarding cultural and natural/post-
deposition formation processes, the investigation of which can contribute both to
understanding variability in the samples, and to higher-level questions about sample
preservation and visibility, and the implications for the writing of Oceanic prehistory.
What is the state of preservation of the samples? In what taphonomic
circumstances? A summary of the physical geography/geology of the research region noted
two unusual properties of the Roviana landscape, which relate to the preservation of
surface sites in the sea:
• the Holocene marine transgression has approximately kept pace with tectonic
uplift, as Plio-Pleistocene uplifted barrier reefs are a few metres above sea level
across most of the survey region, which is inferred to mean that sea levels have
been relatively stable over recent millennia, falling at least1.5 metres from a high-
stand circa 4500bp, exposing ceramic deposits that were formerly subtidal to
littoral-zone processes, creating lag deposits of ceramics, favouring detection by
archaeologists.
• the chain of upraised barrier reefs that encircle New Georgia and form a series of
lagoons, of which Roviana is one, shelters the coastline within the lagoons from all
ocean waves, favouring the preservation of littoral-zone ceramics, also a factor in
the detection of these sites by archaeologists.
The extent of these effects is investigated in detail in Chapter 5.
Sample acquisition, through the systematic survey and sample collection of a
region, is the basis of all further analysis. How should the survey region or regions be
chosen, and what methods should be used to survey it/them? How extensive, and how
intensive, should the survey be? How much information is enough for the questions to be
asked of the material to be satisfactorarily answered? In Chapter 2 these questions are
investigated, with examples from surveys in Near Oceania reviewed. The methods and
71

esults of the Roviana surveys are discussed, and some conclusions are drawn regarding
research design.
A review of approaches to ceramic classification identified a need for a materialist
approach, where the units of classification used derive from an analysis of variability of the
samples in hand, in a dialectic with the research questions. Temporally sensitive attributes
are needed if a fine-grained chronology is to be constructed, and the identification of these
requires a careful analysis of the sorts of dimensions of variability in the data, whether
functional, stylistic, geographic, sample-size-related, or to do with the brokenness or
completeness of the sherd samples from which pottery characteristics are inferred. Units
of classification that are not sensitive to differences in vessel brokenness are needed, and
units of decorative classification need to be identifiable by location on the vessel, to capture
the structure of decoration across the vessel, even in broken sherd assemblages. With these
requirements in mind it was concluded that latitudinal bands of decoration should be
central in the classification, as these were identifiable by location even for very broken
assemblages, due to location near to the rugged and well-preserved corner points (lips,
necks, carinations) of vessels, which also had the advantage of boosting sample size.
Chapters 4 and 9 cover this in detail.
Evaluation of the recovered ceramic samples is an important step in the study of
formation processes. Estimating the brokenness and completeness of our samples allows
choice of appropriate units of quantification for evaluating sample size. Comparison of the
sample with an inferred breakage population is also possible in many situations, especially
where samples derive from surface sites, or large area excavations. The relationship
between the breakage population and the extant deposit can delineate the potential for
taphonomic bias in recovered samples, and provides a method of responding to Gosden’s
call for controlled comparisons. Analysis of brokenness and completeness is reported in
Chapter 5.
A comparative study of wave exposure of the Roviana ceramic sites is needed to
72

e able to assess the relative contributions of natural and cultural formation processes. In
Chapter 6 the oceanographic setting of the sites is described, and an analysis of relative
wave exposure is undertaken. These findings form the backdrop to the study of other
formation processes. A range of formation models are evaluated in Chapter 7 with
reference to the properties of sherd assemblages. Spatial analysis as an aid to the
identification of formation processes is deferred to Chapter 11, as it is necessary to
formalize units of classification of materials prior to undertaking spatial analysis.
An extended discussion of seriation theory and the need to control seriations for
vessel form variation was given above. In Chapter 8 the details of an analysis of form
variability are presented, preparatory to seriation analysis. It was also pointed out above
that units of classification need to be adapted to the place and period under study, and also
to the range of variation present in the sample. For the Roviana materials this means that
the units of classification used need to be designed to capture with sensitivity the
transition from a style of pottery that is recognizably Lapita to something that is not
Lapita. An analysis of variability is needed to translate the initial units of description into
units of classification appropriate to these aims. Furthermore, units of classification need
to be insensitive to differences in vessel brokenness if they are to be useful for
comparisons. Finally, it is beneficial to the analysis of relatively plain pottery, in terms of
the sensitivity to change, if units of classification make use of the information to be had
from the structure of decoration across the vessel. For all of these reasons a detailed
exploration of ceramic variability is desirable, reported in Chapter 9.
The collection sites included lithic items alongside ceramics. While there has been
little discussion of these in this introduction, everything written above regarding treating
ceramic sherds as sedimentary particles applies equally to lithic manuports and artefacts.
Analysis of lithics in Chapter 10 is made largely from the point of view of
geoarchaeological sourcing of raw materials, although a discussion of adze forms is
included also. The sourcing analysis, which seeks to identify procurement units, is the
73

asis of the lithic component of analysis of intrasite spatial structure in Chapter 11. Spatial
analysis yields information bearing on the identification of cultural and natural formation
processes. In Chapter 11 spatial units of classification are re-evaluated prior to seriation
analysis. One collection site (Zangana) is split into two spatial units for seriation as a result
of the spatial analysis.
The extended discussion of seriation given above is translated into practical analysis
of Roviana ceramic data in Chapter 12. Carbon 14 data and thermoluminescence data are
integrated with seriation data to produce a provisional chronology in need of further
independent corroboration and fleshed-out sampling.
Information emerging from these various analytical chapters is pulled together in
Chapter 13. Here the focus returns to the significance of the findings regarding Lapita
detectability in Near Oceania. Despite the difficulties of constructing a reliable chronology
with the sample in hand, there is also some discussion comparing the Roviana data with
ceramic data from other areas. There is a distinct possibility that some ceramic variants
mixed in with Lapita in other places are strongly separated spatially from Lapita in the
Roviana lagoon early ceramic sample. If this is so, then the approach taken to constructing
chronology, across space rather than by the bedded strata method, is amplifying temporal
resolution beyond what is obtainable from the test-pitting approach of the bedded-strata
school.
Ceramic petrographic sourcing has not been discussed in this introduction, and will
not be discussed in any detail in this thesis. Results of this sort of analysis were interesting
though, but for reasons of brevity discussion of the implications of work either published
elsewhere or appended (appended see Dickinson 2000a, Felgate & Dickinson 2001) is
confined to a short section in Chapter 13, integrating the analysis of ceramic tempers with
petrographic analysis of lithics.
Before embarking on these analyses of variability of the Roviana materials though,
it is necessary to review existing literature pertaining to Lapita and the archaeology of
74

Near Oceania, in the light of the methodological review above, to provide ceramic-
chronological context for the Roviana study. The particular foci of this review in the
following chapter will be the temporal resolution of existing constructs of Lapita temporal
variability, and the level of confidence with which these can be accepted as tested
syntheses.
75

CHAPTER 2:
A REVIEW OF CONSTRUCTIONS OF LAPITA
Introduction:
TEMPORAL VARIABILITY
The Lapita Ceramic Series (Golson 1971, Green 1978, 1979, 1990, see also Sand 2000)
while not a term in universal use, embodies a concept of Lapita ceramics as a time-
transgressive changing system of ceramic production/discard comprising regionally varied
sequences within the overall region in which Lapita has been found. Summerhayes
recently suggests a universal series, where there is geographic variation, but where in
some respects ceramics change in parallel through time across the Lapita distribution,
indicating continuing interaction (Summerhayes 2000a, b, 2001, 2002). In this review I
suggest that the security with which a Lapita ceramic series has been defined has been
over-stated in most cases examined, and that formulations of a Lapita ceramic series are
mostly working hypotheses rather than confirmed histories of ceramic change. This means
that higher-level inferences regarding social processes, such as colonization rates (e.g.
Anderson 2002) are not founded on secure high-resolution chronologies. Getting the
historical facts straight is vital if explanations for change are to be stable and sensible
(Spriggs 2001).
Golson (1971) noted technological continuities between Lapita and post-Lapita
plainware, and in view of this evidence for homologous similarity, and also in regard of
the increasingly varied corpus of radiocarbon dates accumulating from Lapita sites,
extended Gifford’s conception of Lapita as a widespread pottery time-horizon (Gifford &
Shutler 1956:93-95) to include the notion of temporal variability by use of the term
tradition, a situation for which Golson suggested the term series as appropriate (Golson
1971:75) (although Gifford did not use the term horizon, he clearly regarded the Watom,
77

Isle De Pins, Site 13 and Sigatoka materials as roughly of equal age). Golson’s
understanding of “series” was that: “The series is made up of a set of ceramic styles which
are similar and contiguous in either space or time or both (Golson 1971:75)”. Golson’s
“Lapitoid” ceramic series was a hypothetical rather than an “established” detailed
construct, as “...the individual styles that it includes have yet to be isolated, described and
named (Golson 1971:75).”
Green constructed a series for plainware in Samoa, running from circa 800BC
Lapita to Polynesian plainwares by 500BC, and saw parallels in the sequences of Fiji and
Tonga (Green 1974b:253), but objected to Golson’s use of the term Lapitoid, in spite of
the evidence for continuity with Lapita, in view of the absence of definitive (in Green's
definition of Lapita) Lapita design motifs on these plain vessels (Green 1974b: 250-251).
He preferred at that time to think of Lapita sensu strictu as a style-horizon, characterized
by a wide range of elaborate vessel forms, divisible at most into a series comprising Early
Western and Early and Late Eastern Lapita styles, on the basis of differences in vessel
shape and the style and frequency of decoration (Green 1974b:251). Green saw the demise
of Lapita, and then of Polynesian plainware, as a functional change, possibly simply
culinary, but also possibly indicating fundamental social changes consequent on settlement
of a previously unoccupied island world (Green 1974b:253). Green defined the Lapita
Ceramic Series as
“...assemblages in which various shouldered pots, jars and bowls, as well
as flat-bottomed dishes and plates, occur in association with widely varying
percentages of dentate-stamped, notched and incised decoration. The
decoration consists of a catalogue of elements and motifs whose
combinations can be listed and compared. In addition, there is a range of
infrequently decorated bowls of simple shapes and varying sizes, plus
several forms of rather more frequently decorated sub-globular pots. Site
assemblages of Early Western and Early and Late Eastern styles can be
recognized within the series by differences in vessel shape and by the style
and frequency of the decoration.” (Green 1974b:251)
78

Green’s 1978 Lapita Ceramic Series:
Green summarized the existing Mead-system analyses as indicating a division into
Western and Eastern Lapita, which both shared a core of widespread “early” motifs. The
Western division was characterized as having a diversity of complex motifs, while the
Eastern Lapita division had a relatively poor inventory of comparatively simple motifs.
Green defined these motif sets more specifically in his subsequent analysis of
Lapita motif occurrence (Green 1979), departing substantially from Golson’s and
Specht’s sequence of “styles” approach, taking instead an attribute/computer clustering
approach to group similar assemblages in an ordered similarity matrix, very much in the
Irwin mold (Green 1990:37), using Mead-system classificatory units. Although his
primary objective in highlighting differences between Eastern and Western Lapita in this
way was to test his hypothesis that the Fiji/Vanuatu water gap caused an early break in
a unified Lapita interaction system, his paper included a number of chronological
statements concerning agreement between his C14 chronology and decorative similarity,
which effectively constitute an occurrence seriation.
At the time Green conceived of the sites in his analysis (the Reef-Santa Cruz sites
at least) as slices in time over up to 1000 years of Lapita production/discard, and had
pooled C14 ages by site to reflect this view (Green 1976a). Occupation span was not the
focus of any extended discussion. These sites are some of the best-sampled and dated
Lapita sites on record, and therefore provide an illustration of the need for a general shift
in emphasis from the age of the “site” to the age of the artefact in seeking to construct
high-resolution chronology. This can be done by focusing on the Reef-Santa-Cruz case
where the sampling difficulties of a reconnaissance test-pitting approach have been
avoided by systematic surface collection and large area excavations.
Green regarded the closeness of sites SZ-8 and Vatcha in the matrix (Green
1978:figure 7) as confirmation of the early date of both sites, as suggested previously by
the particular archaeologists concerned with those sites (Green/Donovan for SZ-8 and
Frimigacci for Vatcha). He also saw this analysis as reconfirming the close relationship
between RL-2 (RF2) and Watom.
Green presented one further matrix, an analysis of the three Reef-Santa Cruz
assemblages (Green 1978:figure 8). He re-stated his 1976 position, that the relative ages
79

Table 2: Reef/ Santa Cruz motif counts as given in Anson 1983.
Site Total Motif count Motif Richness
RF2 841 examples 178 Motifs
RF6 252 examples 79 Motifs
SZ8 627 examples 133 motifs
of the sites were known, and saw a high degree of motif sharing between these sites as
indicative of a strong continuity over an extended period among those motifs restricted
to the Reef-Santa-Cruz area. He saw a different temporal trend in the Reef-Santa Cruz
sites to that noted by Birks and Shaw for the Eastern Lapita sequences, suggesting
instead a trend of initial motif efflorescence followed by impoverishment through time,
as evidenced by the motif-poor RF6 sample. Green had applied a correction for sample-
size-related richness using Donovan’s unpublished motif frequencies to assess whether
a motif was absent as a result of sampling error or absent-not-present. This assessment
is challenged below, on the basis that allowance for the effect of sample-size differences
was probably insufficient.
Despite subsequent revision of the C14 chronology (Kirch & Hunt 1988) which
had the three sites of indistinguishable age, Green considered that the sequence of sites
as outlined by in his (1978) seriation could still be supported on the basis of Donovan’s
chronological inferences from the ceramics (Green 1991c). Best has recently taken issue
with this view, suggesting that the diversity of motifs in sites is sample-area related (Best
2002:91). Best argued that the chronology of the sites should be reversed. Here I present
an alternative view, similar to Best’s in that the “temporal” similarity relations are
suggested to have more to do with sample size than initially thought. I look to Donovan’s
and Parker’s theses for additional information on sample sizes, assemblage brokenness,
and relative frequencies of decorative techniques (Table 2, Table 3, Table 4).
Donovan’s hypothesis was that Reef-Santa Cruz Lapita had a decorative
80

Table 3: Sherd counts and MNI assembled from various tables in Parker (1981).
Site Dentate
rims
“richness” that suggested a greater developmental time-depth than had previously been
considered (Donovan 1973:IV). This was consistent with the then C14 chronology, which
had the sites as slices in time over a period of many centuries. Assemblages from three
sites were analyzed, RL2 (RF2), thought at that time to date to approximately 900BC,
RL6 (RF6), approximately 600BC and SZ8, undated at that time.
All rims were removed from these assemblages prior to analysis for special study,
and are not included in Donovan’s counts (Donovan 1973:5). The rims were later
analyzed by Parker, providing a useful independent quantification of relative sample sizes
(Table 3) (Parker 1981). Counts (sherd counts?) of decorative techniques were given by
layer by Donovan, but these data cannot be read as sample sizes in the absence of any
information on vessel fragmentation, completeness, part representation, or sherd sizes.
The relative frequencies of decorative techniques were calculated with counts of plain
sherds included, which tends to muddle the picture, as plain vs decorated may easily
fluctuate over time rather than follow a steady trajectory, and is sensitive to differences
in brokenness. The percentage of dentate to incised sherds by site/layer was recalculated
with counts of plain sherds excluded (Table 4).
These data do not support any change in relative frequency of dentate-stamping
to incised decoration by level in any of the three sites (which is not surprising as the sites
had been subject to post-depositional gardening), but more significantly, differences
between
Incised
Rims
81
Impressed
rims
Plain
rims
Total Rim
Sherd
count/
MNI
RF6 43 1 15 2 61/55
SZ8 111 8 10 14 143/?
RF2 848 50 425 100 1423/666

Table 4: Relative proportions of dentate and incised, recalculated from data presented by
Donovan (1973).
Dentate
Count
sites are slight. (As presented by Donovan, there were substantial differences in the
relative frequency of techniques by site and by layer, but as the recalculation shows, these
have more to do with differences in the percentage of plain sherds, which can mean
differences in brokenness between contexts, than with differences in potting behaviour
in the past.)
Donovan thought that marked decorative similarities between the sites were due
to common origin within the same tradition, and strong regionalism (Donovan 1973: 36,
43). The possibility that the sites were similar because they were all of similar or
overlapping age and that differences arose for reasons other than chronology, was not
considered, because the chronology was thought to be largely known from the C14
evidence.
Dentate % Incised
count
Donovan’s motif counts as given by Anson (Anson counted multiple motifs on a
82
Incised % Total Count
(ds + inc)
Context
40 74 14 26 54 RL2 surface
1647 64 926 36 2573 RL2LayerA
758 66 384 34 1142 RL2LayerB
586 75 199 25 785 RL6LayerA
240 79 65 21 305 RL6LayerB
416 74 143 26 559 SZ8 Surface
789 70 344 30 1133 SZ8levelA
643 66 328 34 971 SZ8levelB
206 67 99 33 305 SZ8levelC
147 67 73 33 220 SZ8levelD
25 71 10 29 35 SZ8levelE

single sherd separately) are the source of the data in Table 2 (Anson 1983:175):
The low motif count at RF6 is most economically explained as sample-size related,
particularly if the units of measurement of sample size are reexamined. Decorative
technique abundances are given in Table 3, extracted from various figures in Parker’s
thesis.
Differences in sampling method by site and within sites (Green 1976a: 251-255)
are significant, as Best has pointed out: the RF2 site is the most intensively surface-
collected and Green excavated the largest area here. The recovered sample from RF2 is
thus a more complete representation of ceramic variability at the site. But even without
making any reference to these data, the ceramic quantities reported by Parker hint at the
same thing. Parker’s rim quantities throw sample sizes for RF6 into stark contrast with
the other two sites. Using sherd count, RF2 has around 12 times as many rims as RF6,
23 times as many using MNI, and 21 times as many counting only those sherds large
enough to measure mouth diameter (providing an unintended size filter). In view of this
MNI information on relative sample sizes it seems unlikely that Green’s Jaccard similarity
matrix of the three sites is being structured by behavioural variation, and highly likely that
motif presence/absence is strongly structured by sample size. Green’s selection of motifs
that were common in some sites to include in the Jaccard similarity matrix is made on the
basis of Donovan’s sherd count quantification of motif frequency, and is probably an
inadequate compensation for sample-size-related differences in sample richness.
Parker, for vessel forms, like Donovan for motifs, remarks on the similarity of the
three sites, also concluding this is “most probably indicating that all belong to a closely
related tradition” (Parker 1981:76). She notes the appearance, though, of vessel type 14
(heavily grooved walls) on as many as ten separate vessels at SZ8, absent from other
sites. (These are assumed to be Donovan’s “large, rung-like projections”).
Thus Parker's 1981 conclusion:
83

“It seems that over a period of more than a thousand years (1600BC to
400BC) there is in the Western Lapita area little change in the shape of
pottery vessels, and that we are looking at a very long-lasting and stable
tradition (Parker 1981:115).”
stems more from the earlier slice-in-time view of the C14 chronology and the consequent
interpretation of Green’s (1978) Jaccard occurrence seriation, than from temporal
behavioural variation in ceramic manufacturing/discard style. An alternative explanation,
where the sites are much closer to each other in age, and potentially contemporaneous or
with overlapping occupation/discard periods, seems a more cautious interpretation of the
data, that is, treating Reef-Santa Cruz Lapita as horizon rather than series. This
assessment largely ignores the C14 evidence, which has always tended to suggest that
those RF6 materials which have been dated are younger than dated materials from
SZ8/RF2, and the results of current dating research are awaited with interest.
The number of sites in the Reef-Santa Cruz sample, and their relative occupation
spans are also crucial factors in seriation. Seriation works best when there is a large corpus
of sites, and all sites included in the seriation have approximately equal occupation span.
Traditionally this was achieved by rejecting units comprising mixtures from widely different
or excessively long periods, as evidenced by diversity of styles. While temporal variation
undoubtedly exists in these three Reef-Santa-Cruz ceramic samples, detailed explication
of this variation has yet to be achieved. Best’s call for a re-examination of the ceramics
(Best 2002:93) cannot realistically be expected to support a reversal of site chronology
as it is unlikely to be possible to construct a robust seriation sequence for Reef-Santa-Cruz
Lapita from a sample of only three sites from the Lapita period, four if the Mdailu site is
included (Mccoy & Cleghorn 1988). The chances of all three being of similar duration are
slim, there is a sample-size problem with RF6 at least, and three is simply too few site-
samples to yield a robust seriation. While there are now a number of other sites of post-
Lapita age recorded in the Reef-Santa Cruz region, these have yet to be seriated, and do
not hold the potential to give a fine-grained chronology of Lapita the way
84

a larger sample of Lapita sites would. The nature of this critique needs to be borne in mind
in relation to the Roviana materials from which a seriation is constructed in this thesis: it
suffers from similar problems, with less radiocarbon evidence to test the conclusions.
Difficulties in assessing the relative occupation spans of sites are compounded by
differences in the representativeness of the ceramic samples between sites and within sites.
Comparison of RF2 and RF6 illustrates this. Even the RF2 sample, which must rate as the
most comprehensive and detailed surface collection and area excavation of a Lapita site
undertaken to date, is more complete from the southern half of the site than the north, due
to more intensive surface collection strategy, a greater degree of vertical disturbance to the
south, and concentration of excavation in the south. Green felt that the northern area of
RF 2 was of shorter duration than the southern, with deposition of ceramics beginning
earlier in the southern area, as evidence by more complex stratigraphy (Green 1976a :255).
This raises a question whether C14 dates from the north of RF2, had they been available,
would be similar in age to those from RF6? Certainly some of the pottery in the RF2
sample is very similar to some of the RF6 pottery, so a simple comparison of C14 data
from the RF2 excavations and from RF6 excavations may mislead regarding the stability
of the ceramic tradition over time, within the period of Lapita production.
Anson’s Early Far Western Lapita:
Anson’s splitting non-classificatory descriptive approach to motif description (Green 1990)
resulted in much sparser data than the Mead/Donovan alloform/motif classification,
potentially creating even greater sampling-related problems for those of his analyses that
were motif-occurrence-based. Anson identified nearly 500 different “motifs” in the
analyzed sherd samples, most of which occurred only in low frequency, or in a single
instance in many cases. Anson gave “sample sizes” (as motif relative abundances, not as
counts as suggested in Anson 1987) for his sixteen analyzed site assemblages in a lengthy
85

motif appendix, with total motif occurrences (sherd counts?) tabulated by sample in a
subsequent paper (Anson 1986:Table 1). While the overall sherd count of motif
occurrences is large for some sites, where sherds-as-vessels analysis has not been
performed we have no way of knowing whether these are independent observations. The
sherd counts and motif occurrences listed by Anson are not necessarily sample sizes in a
behavioural or discard sense.
Using a Robinson coefficient of similarity to construct a similarity matrix from
motif relative frequency data for twelve sites from across the Lapita distribution, Anson
found that motifs from Ambitle, Eloaua (Mussau) and Talasea formed a group separate to
the western Lapita sites, in which grouping the Reef-Santa Cruz sites, the Watom sites and
the New Caledonian sites formed separate, and looser sub-groups (Anson 1983:180). On
the basis of this grouping and the observations that a similar dichotomy could be seen in
the size and spacing of dentate-stamp “teeth”, Anson tentatively defined an “Early Far
Western Lapita”, questioning as he did so whether sample error or other (e.g. functional)
variation could provide a non-temporal explanation for the pattern.
In providing this caveat, Anson sowed the seeds of the current critique. The spatial
extent of the sampling procedure, or the spatial representativeness of his samples, is the
first issue. The size of samples, in terms of the number of independent behavioural
observations represented, is the second. The type of behavioural variation that caused
Ambitle, Eloaua and Talasea to cluster separately from the other sites (chrono/stylistic,
functional, geographic) is a third uncertainty.
To compare data from sites in Vanuatu and Fiji for which presence-absence of
motifs was the only information available, Anson used the MultBET hierarchical cluster
analysis program (he does not give details of the similarity coefficient used in this
programme, but it seems likely, like Jaccard’s, to be sensitive to sample size differences in
sparse data, whatever coefficient is used). Kirch has suggested that samples must show
86

diversity saturation (Kintigh 1984) to be comparable (Kirch 1987a:124), but this only
applies to comparing the presence or abundance of rare types, while Anson argues the
MultBET clustering results are explicable in terms of common types (Anson 1987). This
is supported by similar groupings using the Robinson coefficient of similarity with relative
abundance data (Anson 1983:180), and by his sample-size-corrected manually-calculated
motif sharing percentage data (Anson 1987).
As in my critique of Green’s 1978 analysis above, the basis on which Anson
discriminated rare motifs from common motifs is questionable, given the impossibility of
evaluating motif sample sizes in any behavioural sense from sherd counts alone. Similarly,
for the Robinson coefficient results, we cannot know, on the information supplied by
Anson, to what extent the relative motif frequencies as quantified by sherd count are
representative of ceramic variability of the breakage populations from which they derive.
The effects of variable occupation span may also be fundamental to the patterning
observed.
The implications of the spatial/geographic component of variation was never fully
discussed. With the exception of the three Watom samples, which grouped with the
Western/Eastern Lapita sites, particularly the New Caledonian sites, Anson’s grouping of
sites largely correlated with geography (Anson 1986:figure 5). The similarity between
Watom and western Lapita may be temporal, i.e. Watom and Western/Eastern Lapita is
later than the “far Western” sites; or phyletic ( i.e. either Watom was the origin locality of
eastwards expansion into the Solomons and Remote Oceania; or Watom represents a back-
migration from the East); or may be reticulate, resulting from some form of more
continuous far-west-West-East interaction involving Watom (although the dissimilarity of
the nearby Ambitle samples to Watom argues against this). Anson in the end plumped for
a temporal explanation, at that time regarding this as a hypothesis in need of further
testing.
87

Mussau:
The Mussau sites, characterized by Gosden as yielding one of the better Lapita ceramic
data-sets of the Lapita Homeland Project (Gosden 1991a), cannot be regarded as providing
secure and detailed evidence of ceramic change over time until the details of the ceramic
analysis are published. Some general review is attempted here, based mostly on preliminary
analyses of ceramics (Kirch 1987b, 1988c, 2001). The Mussau Lapita Homelands Project
research under Kirch’s leadership located what are claimed to be bedded Lapita deposits
at site ECA area B, documenting a declining frequency of dentate-stamping and an
increasing frequency of incision with superposition in a stratigraphic/superposition series,
with rim notching reaching a peak in the earlier units (Kirch et al. 1991: Figure 3). There
has recently been a substantial amendment to this preliminary sequence based on C14
evidence, with an early plainware site now understood to predate the dentate Lapita (Kirch
2001:206, 219).
It would be interesting to know whether or not there were sherds from the same
vessels in different zones of ECA area B (Summerhayes provides useful information of
this sort in his Arawe analyses). Zone B1 and B2 were described as highly disturbed by
sand-crab burrowing that was stated to have transported substantial amounts of pottery to
the surface (Kirch 2001:86,92), although an argument was made that this had not affected
the waterlogged portion of the deposits. Kirch does not theorize the nature and effects of
bioturbation and swash turbation when the environment of deposition was shallow water
in the past. A substantial level of burrowing and turbation by marine arthropods,
mammals, flora and fish might be expected to have occurred (Ferrari & Adams 1990);
also in some weather conditions wave processes may have affected the pottery while in
the sea. Regarding the superposition sequence from the upper zones of the square,
Gosden’s admonition that sequences need to built from comparable site types (Gosden
1991a) has clearly not been extended to include comparable depositional contexts within
88

an excavation square.
An initial claim for “stratified” evidence from Weisler’s excavation at the EKQ
rockshelter on Elaoua has since been revised. While Kirch now regards the lower
component as “including a late-Lapita component dominated by incised ceramics”(Kirch
2001:214), initial interpretation of the ceramic deposit was that it “provided the best
stratified sequence of ceramic materials from any Mussau site” (Kirch et al. 1991:151).
Summerhayes, in synthesising the Bismarcks Lapita chronology, incorporated the earlier
view, with a claim for a transition from dentate to incision over time in this deposit
(Summerhayes 2002:27). The numerically large sherd sample had an average sherd weight
of 1.54 g (Kirch et al. 1991, Weisler 2001), and assessment of sample significance is thus
problematic, if even possible. Weisler quantified decorative changes by level using summed
weight by decorative technique, but the total count of 26 dentate sherds looks to have been
spread largely at random though the 30 ceramic-bearing levels of units one and two (data
for which are illustrated in Weisler 2001:159), and motif analysis has not yet been
published.
The “Changing” Face of Lapita:
In a departure from either Anson’s Motif inventory approach or Mead’s structural
approach to sherd decoration, Spriggs, reminiscent of Specht’s (1968) call for a focus on
whole design, advocated a shift of focus to Lapita vessel design in total rather than
quantification of decorative motifs represented on sherds (Spriggs 1990). This brings to
mind Shepard’s caution that the brokenness of assemblages should not be allowed to
determine units of analysis, and that complex pictorial designs are not suited to
element/motif analysis. Spriggs noted that this approach has been severely limited by
generally highly fragmented samples of Lapita pottery, but suggested that the Mead
system might not be the most appropriate method of analysis for such complex designs,
89

in which he echoed Donovan’s expression of the difficulties of coding a description of
complex curvilinear Lapita designs (Donovan 1973:Vol 2: 64, 130-34).
Spriggs suggested that the curvilinear ‘face’ designs formed a coherent
evolutionary chronological series from complex to simple, and that this could be used to
date sites. Spriggs did not discuss the implication that where such a 'series’ was present in
widely-separated sites these would have to have a sufficient level of interaction and
convergence in ceramic design (not the same thing as interaction) to have caused parallel
changes in different regions. His analysis is thus founded on an assumed universal Lapita
series rather than derived from detailed regional sequences of ceramic change. Some of the
principal questions regarding the Lapita ceramic series are to do with whether ceramic
change was synchronous or not across broader regions (Spriggs 2000:355, Summerhayes
2000a:235).
While Mead considered that his structural approach established a phyletic similarity
between all Lapita decoration (as did Green), this is an overstatement for some of the
simpler motifs, which could usefully be omitted from analysis as potentially being
analogous similarity rather than homologous. An advantage of Spriggs’ approach,
concentrating on the more complex aspects of the designs, not stated by Spriggs, is that
potentially analogous similarity of some simple designs, widespread throughout the world,
is excluded.
Spriggs’ phyletic approach has recently been elaborated and revised (Ishimura
2002), who, like Spriggs, adopts a view that the overall trend in Lapita decoration, “...the
dynamics of chronological change in the design structure....” is one of simplification over
time. Mussau is cited as one of the regional sequences thought to support this doctrine of
simplification, which is not wholly congruent with more recent publication of Mussau
results which have an early plainware predating Lapita (Kirch 2001:206, 219). While
Ishimura makes no reference to sample-size differences in discussing the presence or
90

absence of design variants in site assemblages, his analysis of the Santa Cruz sequence,
for example, the occurrence of a late variant of his “spade” design only in the small RF6
sample but not in the larger RF2 or SZ8 samples is noteworthy. This supports an
interpretation of RF6 as including a later date in its occupation span than RF2, unless
absence from RF2 is sampling error.
While Ishimura feels his typological analysis supports a scenario of gradual
dispersal, it could also be argued that it supports a scenario of rapid dispersal, as some
early variants (Type 3 spade designs for example)are found from Watom to Fiji, showing
no gradual cline, and Type 2 spade designs are only found in the Bismarcks sites, bringing
us back to the uncertainty faced by Anson in trying to interpret essentially the same
dichotomy: is the Type1-Type 2-Type3 evolution of spade designs really temporal, or is
this simply geographic variation? Are the Bismarcks designs different because they are
geographically separate or because they are earlier? The radiocarbon evidence suggests
the latter, but Type 10 faces occur at Eloaua, RF2 and Site 13 on New Caledonia,
supporting a relatively rapid spread of Lapita from West to South within the occupation
span of Eloaua, rather than the gradual spread Ishimura suggests. The similarity of Watom
to New Caledonian Lapita brings to mind processes of reticulaton rather than gradual West
to East spread, where potentially Watom is stylistically anomalous in terms of its face
designs as a result of its geographic location on the stepping-stone route to and from
Remote Oceania. It is not difficult to entertain the possibility of a group of migrants from
further East becoming established at Watom, for example.
Ishimura demonstrates that a phyletic approach to design seriation is valuable, in
that ideas about the relative occupation spans of sites begin to emerge from the lacuna in
which they have languished through the period of attribute-based similarity grouping
seriation and heavy reliance on radiocarbon as a primary chronological tool rather than a
confirmatory technique. Ishimura has done us a great favour by bringing focus back to
91

occupation span as a crucial aspect of chronology, breaking free from the slice-in-time
view of sites so prevalent through the period of the New Archaeology. If his phyletic
classification does turn out to be valid, it poses a problem for some of the slice-in-time
assumptions, for example a short occupation at RF2 (Sheppard & Green 1991), a site
which along with Eloaua appears on the basis of Ishimura’s construct to have a longer
occupation than any other Lapita site from the perspective of face designs, while in terms
of spade designs it is exceeded in occupation span only by Watom.
While the sampling strategy employed by Green at RF2 is clearly more
comprehensive than the other Lapita samples in Ishimura’s study, which might account for
the richness of the RF2 sample, Ishimura’s organization of this RF2 material into a phyletic
series spanning Type 3 to Type 14 “face designs” and Type 3 to Type 12 “spade designs”
seems to invoke an alternate explanation, at odds with a slice-in-time view of RF2 as a
short occupation.
Whether Ishimura’s Phase 1 (1500-1200BC) is really three centuries in duration,
or whether it is a geographic variant or a slightly earlier variant (perhaps beginning fifty
years earlier?) of the more widespread Phase 2 Lapita is not yet known with any certainty.
Ishimura sees his results as supporting Summerhayes’ Early, Middle and Late Lapita, but
one is left wondering how this phyletic analysis gels with Kirch’s recent swing to an early
plainware predating the elaborate designs at Mussau. If his phyletic evolutionary rule from
complex to simple is valid, perhaps the early plainware at Mussau is a functional variant
of an even more elaborate set of Asiatic designs as yet undiscovered?
A problem with Ishimura’s scheme is that the C14 chronology against which the
phyletic series is tested is of greatly different temporal resolution than this phyletic
seriation. The C14 data is mostly site-based, while the phyletic data is vessel-based. If ages
of “Types” rather than sites in Ishimura’s scheme were available, the prospects for testing
competing theories of variability in the ceramic data would be improved.
92

Summerhayes, West New Britain, Anir, and a Three-stage Lapita Ceramic Series:
Summerhayes regarded his west New Britain data as support for Anson’s Early Far-
Western Lapita hypothesis. Like Anson, he saw a temporal dichotomy in the Bismarck
Archipelago data, and extended this to the wider Lapita distribution. Unlike Anson, he
closely identified this dichotomy with form/functional variability within ceramic
assemblages, suggesting that the difference between early and late sites was the loss of
certain special-function forms, on which particular dentate motifs were found.
His construction of a three-phase series for all of the Lapita distribution requires
that the New Britain superposition evidence (on which the suggestion that Anson’s theory
was “confirmed” was based) be reviewed in detail. Summerhayes regarded some West
New Britain sites as having lengthy ceramic superposition chronological sequences which
allowed examination of the changing nature of ceramic production (Summerhayes 2000a:
3-4). He wished to know whether the West New Britain sequence paralleled changes
elsewhere in the Pacific (Summerhayes 2000a:4). Based on his analysis of the ceramic
results of several West-New Britain excavation conducted during the Lapita Homeland
Project, followed by wider comparisons, he concluded that the stylistic province models
of Green (Green 1978, 1979) and Kirch (Kirch 1997) could be replaced with a simple three
stage overall sequence of ceramic change, where the far-western, western and eastern
styles become early, middle and late Lapita (Summerhayes 2000a:235). Sites FNY and FOJ
were regarded by Summerhayes as having long sequences, with heavy use made also of site
FOH stratigraphy (squares D, E, and F) also, in defining a temporal directional change in
pottery style. Summerhayes’ evidence from these sites will be reviewed in order to evaluate
his Lapita Ceramic Series.
Site FOH:
FOH was located on a sand spit, with the dense pottery concentration below the water
93

table (as at site ECA in the Mussau group). Excavation over four field seasons included
geomorphological investigation (Gosden & Webb 1994, Summerhayes 2000a:21-22). The
lower levels of the site were thought to have originated as the discard from stilt villages
into shallow water, and these lower 45-50cm of deposit were excavated in spits forming
Summerhayes’ analytical units A-E for squares D, E and F, with A being the basal spit and
E the uppermost, the latter in the partly-concreted sand layer between the brackish
groundwater and the salt tidal incursion.
Summerhayes reported that sherd counts decreased dramatically up through these
five levels (Summerhayes 2000a:43), and vessel refitting found a high degree of vessel
completeness (see illustrations Summerhayes 2000a:67-70) with many joins between these
levels. Summerhayes considered that as only three co-joins from level E were with
underlying units there was little evidence for vertical disturbance (Summerhayes 2000a:22-
23) but I would argue that 100% of the joins in level E were to other layers, and that
Summerhayes' conclusion that little disturbance had occurred since deposition is not well
supported by his data. The number of co-joins to other levels for each level (Summerhayes
2000a: Table 3.1 on p23) seem consistent with the analyzed sample size for each layer
(Summerhayes 2000a: p44 Table 5.1) and a reasonable explanation for the overall pattern
of co-joining is that turbation of some description has occurred across the four lower
excavation levels. Summerhayes’ claim that pottery assemblages from these levels can be
regarded with confidence as superposed temporal sets is open to question on this basis
alone.
Are there perhaps taphonomic or other sampling explanations for the changes in
decorative technique and motif observed through the levels? There is clearly a different
chemical weathering regime at work in Layer E as evidenced by the presence of the
concretion here, and the decorative and stylistic temporal changes suggested by
Summerhayes are furthermore evidenced by a sample of only 49 analyzed sherds from this
94

upper concretion level, including18 decorated sherds; thus sample size effects might also
easily account for the “temporal” patterning noted for this part of site FOH. This is
particularly so where vessel completeness is this high, as evidenced by Summerhayes'
excellent illustrations of vessel families, with many sherds potentially from single vessels.
The vessel sample may consequently be much smaller than the sherd count, which may not
provide independent observations in these circumstances.
FOH squares G1 and G2 were situated 15m south of squares D, E and F discussed
above, and yielded 2883 sherds, of which 156 were analyzed. Stratigraphy seems almost
identical to squares D, E, and F, with layer three and layer four seeming likely to equate
to the previous units A-D of squares D, E and F (Summerhayes 2000a:23). Layer 3 yielded
a total of eight decorated sherds, while Layer 4 yielded 89 (Summerhayes 2000a:91-92).
Summerhayes argues in Chapter 10 that the assemblage from this square is later than the
assemblage from Squares D, E and F, units A-D, and contemporaneous with unit E (the
sample of 18 decorated sherds). As stated above, a chrono-stratigraphic distinction
between any of these lower units seems unwarranted on the laudably detailed evidence
presented by Summerhayes, and a stylistic distinction faces the serious and fundamental
questions of whether one is seeing activity/functional patterning across a single occupation
unit or chronological change between areas, or whether this is simply sampling error. The
differences in the frequency of dentate-stamp decoration may also relate to
(contemporaneous?) functional differences in vessel form (Summerhayes 2000a:101) rather
than differences in time.
Site FOJ:
Summerhayes’ evidence for stratified stylistic change at FOJ (one line of evidence for his
contention that the direction of ceramic change is known) is examined below. Units B, C
and A seem broadly uniform in their characteristics, the major difference being that unit
95

A has about six dentate-stamped sherds less than would be expected from the sample size.
In view of the difference between these depositional contexts (B/C, D 1n shallow water
and A on land) attributing such differences to temporal change in ceramic production is
problematic.
Gosden and Webb describe unit A as follows:
“The bulk of sediment accumulation probably occurred during storm
activity, when sea-level is probably elevated.... (Gosden & Webb
1994:41)”, and comprising: “...a unit of light coloured sand...roughly 1m
thick...contains relatively small amounts of pottery and obsidian, but
moderate amounts of shell....Occasional mumu (oven) stones are also
present....The reduced percentage of artefactual material in this deposit
indicates that it is unlikely that there was still a village located directly on
the site (Gosden & Webb 1994:40-41).”
The presence of the oven stones was for unspecified reasons considered evidence for
terrestrial use of this location (presumably because these were unlikely to remain in
suspension in the water column, but transport by rolling onto the beach ridge as a result
of storm-wave action was apparently not considered). Summerhayes states that: “It is
within this layer that occupation occurred on dry land.... (Summerhayes 2000a:24)”,
implying the ceramics from this layer were from a settlement on dry land. Many of the
cultural materials in this layer, including pottery, might derive from material originally
deposited in the sea at an earlier time (contemporaneous with the lower units?) and
redeposited by storm waves to form a beach ridge.
Site FNY:
Summerhayes’ analysis of the FNY ceramics concluded that dentate-stamp decoration
drops in frequency in the upper of three units, and noted trends in vessel form from lower
to top unit (Summerhayes 2000a:125-126). In Chapter 10, this is re-stated as:
“incised and fingernail impressed decoration increasing in the upper units,
although here dentate decoration is always dominant. (151)"
Gosden and Webb interpret the lower two units, both brown clay separated by a white
96

sand lens, as back-swamp deposits during a period of Lapita occupation and beach-ridge
formation (Gosden & Webb 1994:46-47, Summerhayes 2000a:25). The top unit comprised
black shell midden with Lapita and Late-prehistoric (my emphasis added) pottery mixed
(Gosden & Webb 1994:47). Summerhayes describes this top unit as comprising
“A top black midden layer (1.2m) incorporating post Lapita (my emphasis
added) and Lapita remains (Summerhayes 2000a:25).”
Summerhayes’ use of the term “post Lapita”, suggests episodes of ceramic
deposition contributing to the upper unit are not widely separated in time. The zone
between the upper mixed unit and the lower, purely Lapita units yielded several
radiocarbon determinations which varied between 1290 and 690 cal BP, with a Lapita-age
estimate from an oyster shell lower down the brown clay. (Summerhayes 2000:25). It
seems well attested from the C14 evidence and from the Lapita and late-prehistoric pottery
styles represented that the upper unit represents a mixture of materials from widely
different periods, while at least the upper portions of the brown clay also comprise mixed
deposits. The absence of recent styles in the brown clay lower units, and a single Lapita-
age date, do not provide evidence that the lower clay is temporally un-mixed, merely that
the clay formed during a period in which only Lapita age materials were available as either
a natural or cultural sediment supply. If this site comprises variable mixtures of styles from
widely separated periods (c.1000 BC Lapita and c.1000 AD late-prehistoric) it should not
be regarded as a superposition sequence of change in the Lapita ceramic complex over
time (see for example Summerhayes 2000a:151), and the status of this unit in subsequent
seriations (for example Summerhayes 2000a:152 Figure 10.1) should be interpreted with
this in mind.
Summerhayes’ construction of the direction of ceramic change does not
convincingly link superposition assemblage variability to time, by under-utilizing Gosden
97

and Webb’s study of site formation processes, and by comparisons between samples of
dissimilar depositional type (see Phelan 1997:36-38). Summerhayes’ conclusion that
“the Arawe assemblages largely agree with other Lapita assemblages which
show both a decrease in dentate and an increase in linear incision over
time....”
and that the Arawe assemblages “confirm” the trends noted by others in this regard
(Summerhayes 2000a:151) is not well supported by the evidence presented. Summerhayes
went on to use this directional theory in a series of evolutionary seriations, using
decorative technique frequency and motif occurrence.
Summerhayes’ Seriation Using Decorative Technique/ Vessel Form:
Having concluded that he had confirmed the temporal direction of ceramic change,
Summerhayes constructed a seriation of West New Britain assemblages based on
decorative technique frequency (Summerhayes 2000a:152) Here excavation squares are
treated as single-period samples rather than time-transgressive units as in the previous
stratigraphic analysis, with all layers being included as the site sample. There is no
discussion in relation to the seriation of the relative occupation spans of these units, or the
degree of temporal mixing of separate periods, which are both important aspects of
seriation method as outlined in Chapter 1.
Decorative technique and vessel form were found to be highly correlated. Sites
failed to seriate when controlled for vessel form in his decorative technique seriation
(Summerhayes 2000a:p153 Figure 10.5), leading Summerhayes to conclude that what
changed over time was that some vessel forms dropped out. Alternative explanations for
the failure of these sites to seriate when controlled for vessel form is that the classificatory
units of decorative technique used were too coarse, or not salient to stylistic variation, or,
more serious, that stylistic change did not occur across his “sequence”. This last
possibility contrasts with Dunnell’s definition of style as evolutionary drift. Some form
98

of stylistic change must occur within a single vessel form, even in the most conservative
milieu, unless extraordinary measures are taken to prevent change. The difficulty is in
identifying that change. It is possible therefore that the “seriation” is simply a similarity
grouping of functional variability, that has little or nothing to do with time. The most
striking example of this is that his seriation separates FOH squares DE&F from FOH
square G, despite convincing stratigraphic similarities between the lower levels of these
excavation units and a spatial separation of only 15m.
Summerhayes’ Lapita Motif Analysis and Early/Middle/Late Lapita:
Summerhayes went on to construct a motif presence/absence/sharing similarity grouping
for a range of Lapita sites from across the Lapita region, using Anson’s motif inventory
as units of classification (Summerhayes 2000a:table 10.5). It is important to note that
presence/absence of motifs rather than motif frequencies were the base data for his
analyses. Anson’s splitting classification creates a large inventory of motifs for which
there is sparse data. Motif sharing may be quite arbitrary in these circumstances (Kirch
1987a).
Summerhayes performed a cluster analysis of motif presence/absence, and a
Principal Components Analysis of the number of shared motifs between sites. He used
Ward’s clustering method (which tends to force outliers into groups and produce tight
clusters in the data or different clusters compared to some other methods such as average
linkage), but does not state what sort of similarity information was input, whether number
of shared motifs, or some similarity matrix such as Jaccard. This clustering method has
been widely used in archaeology for numeric data such as trace-element concentrations
(Shennan 1997:244) and produced what I see as anomalous groupings with Anson’s motif
presence-absence data. The Reef-Santa Cruz sites formed a cluster separate from all other
sites, and Summerhayes states this is a sample-size effect, without discussion of the
99

independence of observations and how sample size is measured. As shown above, the
Reef/Santa Cruz Lapita sites include quite large samples from RF2 and SZ8, and also a
very small sample, RF6, so sample size should not cause these to cluster separately.
A PCA plot of the first and third PCA components agreed with the cluster analysis,
in separating the Bismarcks sites into similar groupings to those obtained in his vessel-
form/decorative technique analysis. As in his decorative technique frequency seriation,
FOH squares fall into separate clusters, Square G being more similar to Yanuca in Fiji
than to squares D, E and F 15 metres away. Summerhayes is not clear about the type and
structure of his primary data: he refers to performing “multivariate analysis on the
presence/absence of motifs in all sites” (Summerhayes 2000a: 160) but also captions
Figure 10.2 (his PCA plot) “Motif Sharing”, and does not provide the raw data input to
his analyses, rendering his multivariate computer analysis something of a black box. PCA
is most appropriate for numeric data (Shennan 1997:298), and Summerhayes may have
done better to use sherd counts or MNV to identify a selection of common motifs with
Correspondence Analysis used as a seriation technique on this selection.
Comparisons are made (Summerhayes 2000a: Table 10.5, Fig 10.12, Table 10.7)
without discussion of the occupation spans of sites or potential functional variability
within sites. If site occupation spans are lengthy or variable, and function varies
substantially across space within sites, assemblages may comprise an inextricable mix of
temporal and functional information, without any clear temporal patterning identifiable by
PCA or cluster analysis.
Summerhayes’s statement that “...differences in (sample) size are seen in the first
PCA component...(Summerhayes 2000a:160, 161)” suggests that it is indeed motif-
sharing counts being input to the PCA. Eigenvalues of the PCA components are not given,
and he chooses to plot the third component against the first, without discussion of the
significance of either the second component or the third. What is the significance of the
100

second hidden component, which must account for more variation in the data than the
third? Furthermore, if the first component is primarily a reflection of sample size, why use
it in the plot, or ascribe temporal significance to the clusters thus formed?
The results of one further analysis, using Anson’s manual motif-sharing similarity
method (Summerhayes 2000a :160-163 and table 10.7), contrast sharply with the previous
multivariate numerical taxonomy results in some respects. FOH square G, more similar to
Yanuca in Fiji than to the adjacent excavation of FOH squares D, E, F in his PCA and
cluster results, appears, using the Anson manual method, to be more similar to Squares D,
E and F than to anywhere else, as might be expected from the same layer of the same site,
separated by 15m of unexcavated ground.
Summerhayes goes on to use these motif analysis results to bolster his argument,
following Anson, that there is a dichotomy in the Lapita assemblages of west New Britain,
and that this is present also in the wider Lapita universe, ascribing a temporal dimension
to this dichotomy, crosscutting a regional (spatial) dimension of variation.
“Similar changes in the Lapita decorative system occur in the west and
east....the product of information exchange which necessitates the
movement of people...” (Summerhayes 2000a:233).
In his discussion of the motif results, he is clear that the similarities noted between
assemblages relate to the presence or absence of particular vessel forms in the recovered
samples, but still does not talk about the extent to which contemporaneous within-site
functional variation, rather than temporal variation, might account for his groupings of
West New Britain (and other) Lapita assemblages.
While the inventory of vessel functions in use in Lapita communities may have
shifted over time, it is also entirely possible that vessel forms/functions were unevenly
distributed across such communities, and that telephone-booth archaeology is hitting
varying mixes of the plain stuff and the fancy stuff. FOH in the Arawes may be the spatial
lesson in this regard, as Sheppard and Green have suggested also for RF2 in the Southeast
101

Solomon Islands (Sheppard & Green 1991).
My conclusion is that the Early/Middle/Late Lapita construct is an example of
Dunnell’s “...inseparable hodgepodge...” of geographic, functional and temporal variation
(Dunnell 1970:310), to which could be added uncertainties regarding taphonomic bias,
quantification of motifs and the degree to which samples are representative.
Early/Middle Lapita in the Anir Group:
Recent extension of the west New Britain research through a program of survey and
excavation in the Anir group has augmented the corpus of Bismarck Archipelago Lapita
samples, with the most notable assemblages excavated from a single productive testpit at
Malekolon and from a grid of 14 productive testpits at Kamgot (Summerhayes 2000b,
2001, 2002). Summerhayes suggests Malekolom TP4 (where the main cultural layer was
located) was deposited on land, near to the then shoreline, while the extensive deposit at
Kamgot was thought to have been deposited either in the water or at the water’s edge
(Summerhayes 2000b). Summerhayes slotted Malekolon and Kamgot sample assemblages
into his early-middle-late Lapita chronology, using the same methods of vessel form
classification, frequency of dentate-stamping (usefully quantified using both sherd count
and MNI this time) motif-based assemblage grouping, and obsidian source frequency).
Summerhayes' use of C14 evidence to test his Early/Middle temporal construct
(Summerhayes 2002) is not an independent confirmation of the temporality of variation.
While a plot of some of these dates appears to be convincing evidence of the temporal
primacy of Summerhayes’ “early Lapita”, his process of chronometric hygiene excludes
any dates which don’t agree with his temporal construction. Thus Beta 55323, a date of
2880 ±70BP from Adwe squares DEF (“early” Lapita) is rejected solely because it does
not fit his temporal hypothesis, and similarly, WK7564, a date of 2960-2760 cal BP for the
Kamgot “early” Lapita deposit is rejected “because it is statistically distinguishable
102

from the other four dates (from that context)”.
Summerhayes also rejects Kamgot dates that have a wide confidence interval
(ANU 11193, ANU 11190, ANU 11191). One of these dates(ANU 11190: 2750-1530 cal
BP), rather than limiting “the ability to make fine discriminations in the chronology”would
tend to undermine his chronological hypothesis (Kamgot as early Lapita) in spite of its
broad confidence interval. Summerhayes’ conclusion that “These new determinations
confirm the chronology for sites in that region that were based on regional comparisons
of pottery decoration, form and production.” is thus not unequivocally supported by
current data. Any evidence for occupation span (non-overlapping ages) is ruled out rather
than put to good use to understand what sort of temporal mixture the samples comprise.
The Watom Lapita Series:
Watom, SAC location, had Lapita deposits preserved with a minimum of recent
disturbance under a volcanic ash fall (Green & Anson 2000b:37). Anson concluded that
the stratified deposits containing pottery at SAC locality “enabled the chronological
position of unstratified material from SAD and museum collections to be inferred (Anson
2000)”. Best critiqued Green and Anson's temporal conclusions, that Watom Lapita dated
500BC-AD200, suggesting the younger C14 dates obtained postdated the Lapita pottery
recovered (Best 2002:87).
Green and Anson noted that initial Lapita occupation at Watom SAC zone C2
predates their earliest C14 date (Green & Anson 2000b:38), preferring to discard a
Trochus shell date of 3490±80BP (calibrated at 1-sigma to 1509-1350BC) in the surface
of Zone D as being unrelated to Zone C2 occupations (Green & Anson 2000b:39). Green
and Anson present ample and explicit evidence that the buried Zone C1 paleosol is a
loamy gardening soil incorporating cultural materials (Green & Anson 2000b:42), but this
103

stratigraphy is elsewhere described as “...two cultural occupation layers, C1 and C2....
(Green & Anson 2000b: 35).” Zone C1 was dated using several partly dissolved Tridacna
maxima shells from the base of the layer (Green & Anson 2000b:38). C2 was dated using
a stack of Tridacna shells in a pit feature unconformably overlain by zone C1. These were
calibrated and interpreted to yield a time span of 400BC to about AD100. As Green and
Anson are careful to point out, features associated with Zone C1 could only be identified
as intrusions into the surface of C2, confirming a turbation unconformity as the formation
process of Zone C1, as do the smaller and more fragmented bones and potsherds from
Zone C1 in comparison to Zone C2.
For Zone C2 it was possible to demonstrate sequences of superimposed features.
Fom the section drawings, it looks as though a burial ground was cut in from C1 level,
prior to the formation of the C1 unconformity, and which postdates some pottery and
obsidian from C2 (Green & Anson 2000b:45). In view of these formation processes,
Anson’s conclusion that “A fair degree of motif sharing between layers C1 and C2 at SAC
gives an overall impression of continuity, with no evidence of a dramatic break....(Anson
2000:133)” ignores the fact that zone C1 clearly incorporates materials mixed in from, and
originating as the upper levels of part of Zone C2, during the creation of the C1 turbation
unconformity. Furthermore, a complex history of excavated unconformities is evidenced
in Zone C2 stratigraphy (Green & Anson 2000b:44), which must have acted to bring
materials from the earliest facies of C2 into younger stratigraphic features, and into
stratigraphic association with Mopir obsidian, thought to be unavailable in Lapita times
(White & Harris 1997). The burial pits cannot have been “...sealed in by layer C1 (Green
& Anson 2000b:45)." given the formation processes so well documented for C1. Materials
constituting C1 must in part derive from the upper levels of what was originally C2 prior
to gardening.
Rather than there being no evidence for a dramatic break, an important question
104

is whether there is historical continuity of occupation beneath the ash from the time of
initial Lapita deposition to the time of the use of the site as a burial ground. As Best points
out, some of the zone C1 pottery is not necessarily from a late-Lapita context. There is no
evidence to rule out the possibility that some of this pottery could substantially post-date
the Lapita-horizon, while pre-dating the Zone B ash, with an intervening hiatus of several
centuries, while those sherds that are clearly Lapita may be incorporated by the well-
evidenced mixing which was the origin of C1. How does one tell whether Lapita pottery
in a deposit is mixed with post-Lapita, when post-Lapita sequences are so ill defined as in
the Bismarck Archipelago? Given the clear evidence for excavation into C2 after
deposition of Lapita pottery, which involved burial practices not known elsewhere in Near
Oceania for Lapita, it seems reasonable to worry that the stratified sequence of “change”
may be blurred by these formation processes. Anson’s analysis of ceramic style by layers
is critiqued below in view of these formation processes and uncertainties regarding
occupation spans and historical continuity of deposition.
Nail-impressed sherds were found in layer C1 with Lapita, but not in Layer C2.
This may simply be a sample size effect (Green & Anson 2000b:77), or a mixing of
different periods.
Anson found that Lapita motifs from SAC C2 were more similar to SAD and
Meyer’s collection than to SAC C1, although SAD and the Meyer collection are more
similar to each other than to C2 (Anson 2000:132). Anson concludes that SAC Zones C1
and C2 represent temporal stratigraphy, based on a relatively low degree of motif sharing.
Given that the 38 motif occurrences listed for “SAC L1" and “SAC L2" in Table 1 are
represented by sparse frequency data, in 29 instances by single sherds per layer, and never
by more than two sherds per layer for the remaining nine sherds, it is not surprising that
these layers have a low degree of motif sharing. Contrast this with the combined Meyer
sample where single motifs are represented by up to 16 sherds per layer, (motif 429) and
105

it seems clear that there are major unresolved sampling issues here which have not been
discussed.
Anson, in grouping collections using motif sharing, is attempting to produce an
occurrence seriation, without due regard to the effects of sample sizes on motif
occurrences, and also without consideration of what his counts mean behaviourally, and
how sample size might be quantified. Even assuming that his method of quantification is
salient to the completeness properties of the samples, if seven out of 16 motif occurrence
at SAC C2 are shared with SAD, and four out of 18 motif occurrences in SAC C1 are
shared with SAD, it cannot be concluded with confidence that there is a significant
difference. The difference of three motifs may simply be due to chance.
Anson’s analysis of the complete corpus of Watom pottery states that most of the
Mussau Lapita sequence is much earlier than Watom Lapita on radiocarbon evidence
(Anson 2000:120, footnotes), but Green and Anson noted that initial Lapita occupation
at Watom SAC zone C2 predates their earliest C14 date (Green & Anson 2000b:38). In
a similar vein as for the Reef-Santa Cruz sites, I suggest that insufficient consideration has
been given to the possibility that the SAC site and the Reber/Rakival area generally have
been the focus of Lapita occupation for an extended period, potentially beginning many
centuries earlier than 400BC, and that individual site/excavation unit assemblages may be
palimpsests of a variety of occupational events and pottery styles within the Lapita period,
with lengthy hiatuses in occupation not ruled out by the evidence. Ishimura’s phyletic
analysis would tend to support this alternate view, given the diversity of face and spade
designs from Watom (Ishimura 2002).
The pottery styles illustrated (e.g. Green & Anson 2000a: Figure 1) seem to me to
have equivalents in all three analyzed Reef-Santa Cruz assemblages and also in Mussau
and the Arawes. The possiblity that initial domestic occupation began substantially earlier
than 400BC, potentially as early as the dated Trochus shell from Zone D (which is
106

consistent with the earlier dates from house-posts at Mussau) is not precluded by Green
and Anson’s detailed excavation data and materials analyses. The large numbers of Lapita
sherds recovered from Zone C1 are from a chronostraphic unit containing ceramics
potentially manufactured between c. 1400BC and c. 100BC, whereas some of the earlier
stratigraphic units in Zone C2 incorporate materials manufactured over an earlier, shorter
period.
Wahome’s Seriation of Admiralties Pottery Assemblages:
Wahome’s comparison of incised/applied pottery site across Island Melanesia has been
called into question by new data from Vanuatu and by criticism of classificatory units used
as being too coarse to be salient to the research question (Bedford 1999:167-189). For the
Admiralties, Wahome came up with a local sequence of four periods, from Lapita, through
early post Lapita, late post-Lapita and late prehistoric (Wahome 1999:112-118). This
series conflicts with Ambrose’s assessment of historical continuity of the Admiralties
sample. According to Ambrose a gap of up to seven centuries separates Lapita pottery
from subsequent styles (Ambrose 1991, Bedford 2000:183).
Potential problems with historical continuity aside, Wahome’s approach is
interesting for his use of Correspondence Analysis. Wahome used Correspondence
Analysis to seriate fourteen Admiralty Islands pottery samples (Wahome 1999:33-34 and
96-99). Both Lapita-style and later/other pottery variants were included in his attribute and
attribute-combination analyses. Wahome was concerned to identify attributes and attribute
combinations that varied over time. Wahome saw the summary variables in his analysis as
separating functional and temporal variation. He looked also for “reliable sequences”
in these dimensions of variability by cross checking against stratigraphic trends in
excavated sites. He had difficulty seriating some units and concluded that some of the
division of excavated sites into stratigraphic/superposition units was unjustified, due to
107

lack of variability between units (Wahome 1999: section 5.3). His seriation became more
coherent when these units were lumped.
His Lapita unit remained distinct from all other units in his analysis (which is
consistent with Ambrose’s suggestion of a lack of historical continuity). The site of Mouk
yielded both Lapita and post-Lapita evidence (Wahome 1999:5.3, 5.5.1). Three “early
post-Lapita” sites dating to around 2000BP could be seriated using the frequency of shell
impression, with shell impression absent earlier than 1900BP (Wahome 1999:5.4.1,
5.5.2.1). A group of ten “Late post-Lapita” sites seriated in the first dimension (66.8%
inertia) on the frequency of cross-hatch incisions, rolled-rims, shell impressions, and
combinations of these. A second dimension of variability (17%) comprised the frequency
of deep notching, flat lips on rolled rims, and notching on the plain lips of incurving rims.
The first dimension was ascribed to time by comparison with stratigraphic relationships
between units, and the second was disregarded. Eight of the ten late post-Lapita units
included fingernail and/or fingertip impressions, with one site showing a greater diversity
of motifs in these decorative techniques. A “highly diversified range” of incised decorations
was recorded in all these units, and punctations were present in seven sites.
His final grouping of 19 “Late prehistoric” sites (post-800BP) was differentiated
primarily (inertia of 24%) on the frequency of brushing, in which respect one site in the
group differed significantly from the others. A number of attributes/combinations
contributed to the second dimension (inertia was eighteen percent). He considered that
most of this variability correlated with space rather than time or function. There was a
wider range of vessel forms in evidence than in the preceding period, with brushing and
perforation more common (Wahome 1999:5.4.3, 5.5.2.3).
Overall, taking into account units of classification used, and the use of seriation
in combination with stratigraphy and surface sites, Wahome’s approach is similar in
method to the Irwin/Frost school, but with the more up-to-date technique of
108

correspondence analysis substituted for similarity matrix assemblage grouping. Like Irwin,
Wahome put some effort into identifying temporal variation, primarily through comparison
with stratified superposition chronologies. Wahome’s Lapita and Early post Lapita phases
cannot be considered to form a historically continuous series though, as historical
continuity is poorly evidenced at best, and Wahome makes no argument in support of
heritable continuity. Thus Lapita remains a horizon rather than a series in the Admiralties.
The Buka sequence, to the north of Bougainville, was established in its current
incarnation primarily through the pioneering work of Specht (Specht 1969), and a further
indirect contribution came out of the Lapita Homeland Project on Nissan (Spriggs 1991).
More recently, the Buka area has been substantially researched by Wickler (Wickler
2001).
Specht on Buka:
Specht constructed a detailed sequence for the Buka area based on information from 73
sites (Specht 1969: 211-213), spanning the period from Lapita to the present (Specht
1969: Figure XII-52). Specht did not regard his sequence as continuous, being sure there
were temporal sampling gaps. Specht used the hierarchical construct of tradition, style,
substyle and attribute in his ceramic analysis (Gardin 1967, Willey & Phillips 1958),
regarding attributes as irreducible units of description, with essential properties as
discrete, mutually exclusive and immutable entities (Specht 1969: 64-65). Although he
implicitly rejected the concept of pottery type as hindering the study of phyletics,
regarding styles, in a materialist manner, by contrast, as involving time in their definition,
his use of phyletic stylistic analysis was limited, as discussed further below, with styles
being largely a finer-grained classification than types. Specht avoided multivariate
computer classification approaches in favour of an iterative visual approach to analytical
109

classification. He took a metric approach to vessel form variability, making use of vessel
restriction factor, a measure of vessel form popular in American archaeology at that time.
Specht included a succinct yet pertinent discussion of ceramic quantification, which
anticipated Egloff’s significant contribution in this area, suggesting that the subject had
received some consideration at the ANU generally at that time, perhaps more so then than
more recently (Egloff 1973, Specht 1968:70-71, 2002). He used sherd count as his primary
measure of quantity, but counted joining sherds as one. Where there were joins between
layers, the combined sherd was counted in the layer with the most sherds/largest sherd,
providing a rough correction for mixing in some instances.
The recent end of the sequence was defined using surface-collected material. Here,
Specht again provides a succinct and pertinent methodological discussion, this time of
seriation method, and suggests that unrestricted cliff-top site locations are more likely to
provide short duration sites than those on the narrow coastal flat constrained by cliffs, the
later being more likely to present mixed assemblages from extended or multiple periods of
occupation. He supported this theory with a number of stylistically homogenous collections
from the cliff tops thought to be “single-period” occupations (Specht 1968:160-161).
Specht placed heavy emphasis on paste classes in his analysis, an approach
followed more recently on materials from the same general region by Summerhayes, and
by Wickler and by Spriggs (Spriggs 1991, Summerhayes 1996, Wickler 2001) and in line
with Shepard’s emphasis on technological attribute analyses (Shepard 1963).This approach
has been partially borne out by Wickler’s subsequent analysis, in that the Lapita-age
pottery largely contrasted with all other groups in this respect, being calcite-tempered,
where all subsequent pottery had a significant terrigenous component, apart from
occasional post-Lapita Sohano-style sherds (Specht 1968:192-194).
Specht constructed a sequence of styles and sub-styles, beginning with slab-
110

constructed, shell tempered Buka style, followed by Sohano one-piece paddle and anvil
built style, (considered to have been strip-built, followed by paddle and anvil forming:
Specht 1969:196). Hangan, Malasang, Mararing and Recent styles were considered a
gradual evolution from Sohano, but Specht was equivocal on whether homologous
similarity/heritable continuity and historical continuity could be seen between Buka
(Lapita-age) and Sohano phases (Specht 1968 :195, 214, 216, 230). On the one hand some
early Sohano substyle vessels occurred in crushed shell paste, and use of the paddle and
anvil is a linking feature, but abandonment of slab construction, and the absence of a
Watom-Lapita-style of broad rim notching from the large Sohano and later assemblages
suggested a lack of heritable continuity.
Some of Specht’s complex motifs (e.g. M34) diagnostic of particular substyles, are
listed as occurring in other styles also, and this seems to indicate a too-direct reading of
stratigraphic superposition as recording changes in ceramic production. I think it more
likely that complex motifs like M34 have a relatively specific period of production, and that
these occur in other levels as a result of the vagaries of site formation. Just where this
leaves Specht’s construction of a gradual evolution from Sohano to Recent is difficult to
assess (see for example Specht 1969:258)
Wickler on Buka:
Wickler’s recent re-appraisal of the Buka/Nissan sequence, which largely reiterated
Specht’s and Spriggs’ ceramic sequences with minor refinement, added substantial
additional Lapita-phase data, resulting largely from inclusion of reef-flat locations in his
survey strategy (Wickler 2001:8).
Numerically large (high sherd count) yet quite broken assemblages of ceramics
were recovered from Lapita reef sites and terrestrial test excavation The average sherd
weight for a sample from site DAF was five grams, although sherds on the outer reef at
111

this site were larger. At site DJQ 7.6% of sherds were over 6cm maximum dimension
(Wickler 2001:76).
Attribute description of Lapita sherds was made with the aims of investigating
assemblage variability, and of seriating assemblages using a temporally sensitive subset of
attributes. Vessel forms were defined primarily on the basis of rim attributes (Wickler
2001:77), but the criteria for distinguishing between everted rims of restricted vessels and
open bowls are not clear. Wickler’s finding that the Lapita assemblage from site DJQ was
dominated by open bowls (Wickler 2001:90) is difficult to evaluate in the absence of
illustrated examples of sherds that were used to make such identifications. Similarly,
functional distinctions between assemblages based on vessel form (Wickler 2001:122) are
not easily verified. No raw attribute or decorative data at the level of the sherd were
presented, nor are examples of the sherds identifying some vessel forms illustrated. Some
vessel form categories are very broad (“Form 9" for example). Sherd size was not recorded
on a sherd-by-sherd basis, but for rim sherds an approximation of sherd size could be
generated from the attributes “estimated rim diameter” and “rim percentage. The lack of
sherd size information makes identification of taphonomic processes and motif difficult.
Lip form data was strikingly patterned by site, with the plain “Form 1" lip dominant
at site DES (on Nissan) (commonly found on “Form 1a bowls” and “Form 4 bowls”).
“Form 5" lip dominated at site DJQ (suggested to be diagnostic of “Form 2b” everted
bowls), and “Form 11" lips were dominant in the large DAF sample (diagnostic of “Form
9b” necked jars) (Wickler 2001:91). Continuous attribute interval class frequency data
(Wickler 2001:92, Table 4.8) provides useful information on the ceramic sample size and
vessel forms of the Lapita sites, with counts of neck angle classes, carination angle classes,
and rim percentage classes, but it is impossible to know whether these counts are
independent observations.
112

Macroscopic temper analysis used a system of relative abundance ranking (Wickler
2001:98) very similar to the final scheme adopted in Chapter 4 for the Roviana Data.
Ferromagnesian tempers do occur in the Buka Lapita sites in low frequencies, so Specht’s
finding that Lapita-age sherds were exclusively shell tempered was a sampling error. Seven
sherds out of the 6519 for which temper class was identified had quartzo-feldspathic
temper dominant, which may turn out to be more significant that initially thought by
Wickler, as sherds of this temper type produced arresting petrographic results in the
Roviana case (Dickinson 2000a, Felgate & Dickinson 2001). These sherds were thought
to be local by Wickler, but should be compared with the quartz-calcite hybrid tempers in
the present study.
His starting point in seriating the reef Lapita sites was to use frequency of
decoration, and frequency of dentate decoration, both of these characteristics thought to
be indicators of early assemblages, based on the work of Green, Donovan, Parker, also of
the Arawe assemblages from the Bismarcks as outlined by Gosden and others, and also as
in Anson’s analysis. Wickler’s seriation thus had an evolutionary component as the
direction of change was assumed to be known. This may turn out to be a self-fulfilling
prophesy, in light of Kirch’s recent suggestion of an early plain assemblage predating
Lapita from Mussau.
Wickler’s attribute frequency seriation of Lapita reef sites used an unusual and
potentially problematic set of relative frequencies: frequency of decorated sherds
(expressed as a percentage of the total sherd count), the frequency of dentate-stamping
(again expressed as a percentage of total decoration?), bounded incision, and unbounded
incision (these are also assumed to be percentages of the total number of decorated sherds
as the rows in the seriation do not sum to 100). He found a low overall frequency of
decoration on the beach at DAF, with a high frequency of unbounded incised decoration,
and characterized this a late site (300-100BC), while DJQ was regarded as falling within
113

the Lapita period envisaged by Sand, 1000-800BC, with the highest frequency of
decoration, lots of dentate-stamping, and bounded rather than unbounded incision (Wickler
2001:122). As discussed in the methodological reviews of quantification and seriation in
Chapter 1, the percentage of sherds with decoration is easily biased by differences in
brokenness (and brokenness is clearly different across his seriation units), while percentage
decorated is also an attribute that may not seriate well in the first place due to the
possibility of decorative extent and abundance over time.
Wickler was careful to note the difficulty of separating functional and
chronological variation, but felt that functional variation could only account in part for the
reef site decorative/form variation. Also, the total number of collected samples was five
(three areas from DAF and one each from DES and DJQ), which is on the light side for
assessing the degree of temporal mixing. The possibility of temporal mixing with other
later ceramic styles will be discussed further in Chapter 13.
Wickler’s Lapita design analysis was a hybrid of Mead’s and Anson’s analytical
schemes (Wickler 2001:122-127), with an additional feature of careful structure in relation
to decorative location and details of execution (Wickler 2001: Appendix 1). By using
Mead’s conception of design zone, and Mead’s concept of General/Restricted Zone
Markers (Mead 1975: 26, Fig 2.11, 2.12), some conflation of the categories of decorative
technique and decorative layout was introduced, creating proliferation of descriptive terms.
The careful control for decorative location in his motif frequency analysis (Wickler 2001:
Appendix A) is a welcome innovation to Lapita analysis (Mead’s design zones are
locationally vague by comparison), but structural covariation of decoration of various
vessel parts was not investigated in any detail or made use of in analysis. His motif
occurrence and motif frequency comparisons were very much in the mode of Anson.
Decorative classification did not take account of the structuring of decoration across the
vessel.
114

In reporting his most common Lapita stamped and incised motif frequencies as
Anson motifs 435, 2, 187/188 and 421 (Anson 1986: 189-256, Wickler 2001: 124-125),
the effect of sherd size on assignment of these very similar motifs is not discussed. In view
of the small sherd sizes reported, how does one differentiate between motifs 435, 297, 190,
188, 187 or 434 when these are all made up of parallel lines? Illustration of examples on
which such identifications were based would have clarified these issues, as would
presentation of raw analytical data, including sherd sizes (not recorded individually), vessel
parts represented on the sherd, vessel form ascription and motif ascription. This point is
not entirely critical for comparison with the Roviana materials, as, in his decorative
attribute analysis of Lapita, Wickler made a useful distinction between bounded and
unbounded incision, compatible with the present Roviana units of classification as
discussed in Chapters 1, 4 and 9.
Wickler’s Robinson Coefficient motif frequency comparison of Buka reef sites with
Lapita sites of the Reef/Santa Cruz area, Watom and New Caledonia found the reef sites
DJQ and DAF much more similar to each other than to any of the other sites, while DES
(Tarmon on Nissan) was about as similar to RF2 and RF6 as to the other sites. The Jaccard
coefficient motif sharing analysis using agreement scores gave different results, and
Wickler was inclined to favour the frequency data (Robinson similarity coefficient) over
the occurrence data. DES had a low motif count, so sample error was a possibility there
(Wickler 2001:128-129). This general finding contrasts with Summerhayes’ parallel
changes across the Lapita distribution from early Lapita to middle Lapita, unless the Buka
Lapita is all very late, and thus more regionalized (in which case the presence of flat bases,
stands, and some fine needle-like dentate conflicts with the Summerhayes/Anson
hypothesis).
Wickler’s analysis of post-Lapita excavated assemblages will not be reviewed, for
reasons of brevity, since his findings did not differ from those of Specht. Wickler’s
115

finding that there was evidence for heritable continuity between Lapita and Sohano phase
is significant in that it links Lapita and post-Lapita in a tenuous way (as discussed by
Specht, despite his eventual opting for no connection). Also Sohano-phase ceramics are
the only post-Lapita Buka ceramics that bear any resemblance to the Roviana intertidal
ceramics, although late-prehistoric Buka ceramics could be usefully compared to late
Roviana terrestrial plainware (some of the Roviana “post-Lapita” styles may be present in
the reef-flat collections of Wickler in low frequencies: this will be discussed further in
Chapter 13).
Regarding the nature of the transition from Buka (Lapita-Phase) to Sohano
ceramics, he concluded:
“Although handicapped by low sample size, disturbed deposits and a lack
of reliable radiocarbon dates, the available evidence indicates a temporal
overlap in the production of Buka (Lapita) and Sohano style ceramics, and
a gradual replacement of the former by the latter (Wickler 2001 :144). ”
While occasional compositional overlap and a few sherds of transitional style hinted at
heritable continuity (Wickler 2001:141) the interpretation of this as gradual change, by
default, in the absence of evidence to the contrary (Wickler 2001:168) cannot be regarded
as evidence delimiting rates of ceramic change, and I take Wickler’s statements to mean
that Buka and Sohano are related traditions, without necessarily accepting his suggestions
of a gradual transition or temporal overlap in production, for which there seems to be no
particular evidence in the detail of the excavations, where ample evidence for mixing of
ceramics across levels is presented.
Vanuatu:
This review will focus on the recent work by Bedford, in which earlier works are reviewed.
Although a number of surface exposures of Lapita pottery have been recorded on Malo,
116

Hedrick, who recorded many of these sites, focused on investigating settlement strategy
through site location (Hedrick n.d:225-243), and, despite working with large surface
samples, did not draw any detailed conclusions regarding temporal changes in Lapita
decoration. Galipaud has conducted extensive surveys and limited excavation on Malo,
Santo and Torres (Galipaud 1998), but has yet to advance any detailed ideas on the nature
of temporal changes within the Lapita phase.
Bedford in Vanuatu:
A recent extensive reworking of the Vanuatu sequence by Bedford, which included
reporting of substantial new excavation sequences, resolving a number of primary culture-
historical issues, has Lapita as the foundation pottery style (Bedford 2000: 1, 35-37, 49-50,
81, 153-165, 239, ) rather than Mangaasi (Bedford 2000:27). Subdivision or explication
of Lapita-phase temporal variation was not attempted by Bedford, due to the paucity of
Lapita-phase ceramics recovered from reconnaissance testpitting in landscapes often deeply
covered by tephra. Bedford considered that all Lapita samples were at the late end of a
Lapita “Horizon”, with initial Lapita settlement circa 3000BP, after consideration of the
radiocarbon evidence and the coarse dentate-stamping on the sherds recovered, which in
the Anson/Summerhayes model indicate late date in a Lapita series. It seems likely that
evidence for settlement dates similar to New Caledonia will turn up eventually.
On Erromango and Efate, Bedford concluded that local sequences were a product
of continuous ceramic evolution out of Lapita, while on Malakula, a post-Lapita Malua
phase also bore a homologous similarity to Lapita. On Malakula, however, in contrast to
Erromango and Efate, late Chachara bullet shaped pots were regarded as potentially non-
homologous having “no apparent antecedents” among the poorly understood surface
collected wares from the period between Malua and Chachara (Bedford 2000:147, 241).
Accordingly, four regional Vanuatu phyletic series can be constructed from Bedford’s
117

analysis, as follows:
• Erromango: Dentate Lapita >>Ponamla Plain >> Early Ifo excurvate vessels with
Fingernail impression >>Late Ifo incurving rims with fingernail impression and
other decoration>>heavy notched plainware
• Efate: Arapus Lapita-style plainware >> Early Erueti >> Late Erueti >>Early
Mangaasi >> Late Mangaasi (Mangaasi possibly appearing on Malakula in surface
collections also, although this is unconfirmed)
• Malakula 1: Dentate Lapita >> Malua >> ? (small surface collected sample)
• Malakula 2: Chachara >> Naamboi
It is important, in reviewing Bedford’s data and constructing these four phyletic series,
to question their historical completeness. While we know a lot more than we did
previously about ceramic change in Vanuatu, do we know how well the regional ceramic
changes within these four series are sampled? Bedford points out some blank spots in the
Malakula record. Elsewhere, mixing of materials across strata may be a factor obscuring
poorly-sampled periods. In Ponamla Area A, for example, age-depth correlation and
“evidence for stylistic change....confirmed the stratigraphic integrity of the site” (Bedford
2000:43), but stratigraphic unconformity (Bedford 2000: Figure 3.4) is best read as
indicating that some excavation and possibly erosion of Layer 2 and 3a at least has
occurred prior to deposition of Layer 1, incorporating materials originally deposited there
into Layer 1, and that “stratigraphic integrity “ should not be read as absence of mixing,
which Bedford confirms (Bedford 2002). Also, some of the dated Layer 1 material
overlaps at two standard deviations with material from Layer 5, which is not to say that
118

it is mixed to the point where stylistic change cannot be detected, but which should
encourage a cautious approach to the question of whether gradual change with height in
the column directly indicates gradual change in ceramic production style over time.
Phyletic blank spaces in the series could be obscured by mixtures from different periods,
especially when plainware phases exist, as on Erromango in the post-Lapita-phase
(Ponamla plain). A degree of mixing might be hard to distinguish from the plain component
of overlying Early Ifo, other than by vessel form (not possible in the Ponamla plain phase,
in which vessel form seems little different to the succeeding early-Ifo vessel form). Having
the two site sequences to compare has allowed a more complete sequence than would
otherwise be the case, but is there yet more variety to be discovered in additional sites in
the future? If so, how much more? Also, as mentioned above, it seems likely that Lapita
face designs as found on New Caledonia and in the Southeast Solomons and Fiji will turn
up eventually in most areas of Vanuatu.
New Caledonia:
The New Caledonian Lapita/Post-Lapita ceramic sequence has been elaborated recently
by the work of Sand and colleagues, in which dentate-stamped pottery dates favour the
range 1100BC to 800BC (Sand 1998:25-28) (Sand’s table of dates shows a broader
absolute spread for Lapita contexts but he appears to have compressed this at an
interpretive level). Sand sees these materials as allowing study of the decorative
characteristics of the local Lapita motif inventory, with the aim of testing Kirch’s
“Southern Lapita “ hypothesis. Sand sees a progressive replacement of Lapita ceramic
ware by other incised or applied decorated ceramic traditions, and development from
Lapita of the Podtanean paddle-impressed wares and Puen Mangaasi-like incised ware.
Sand considered that some of the non-dentate pottery assemblages and non-dentate
components of assemblages comprise a “Lapita-associated ceramic-series” within the
119

Kone period (Sand 1999:143). He notes that these pots are found in spatial association
with dentate-stamped Lapita pots, and in sites postdating Lapita. This category comprises
triangular incised “non Lapita”, a range of shell-impressed styles, and paddle-impressed
Podtanean pottery. While the presence of Podtanean in most Lapita sites, and in the
earliest stratigraphic context at the Vatcha site, are taken by Galipaud to indicate
contemporaneity with Lapita from first settlement, Sand suspects that this style of pottery
in many cases represents later habitation than Lapita (Sand 2000:146). Galipaud sees the
absence of Lapita in some Podtanean sites as a contemporaneous functional difference,
with Podtanean being a utilitarian Lapita-age vessel type (Galipaud 1996:303). Sand has
late/post Lapita impressed/incised wares in the north differentiating from incised Puen
wares in the south, within his early Kone period (Sand 2000:155).
Sand notes that ceramics excavated are still under analysis, but presents a
preliminary synthesis of Kirch’s proposed “Southern Lapita Province”, defined by Kirch
on the presence of the “Podtanean” paddle impressed wares in association with dentate-
stamped Lapita ceramics (Kirch 1997:73). Sand proposes a short chronology for Lapita
in this Southern province, beginning not earlier than 1100BC (Sand 1997a), and with
dentate-stamping out of use by 750BC, a similar chronology to that recently proposed for
the Eastern area of Fiji-western Polynesia. This is substantially shorter than the 1000 year
chronology proposed for Lapita in the Bismarcks (Kirch 1997), although that chronology
is in turn showing signs of contracting, with a figure of 500 to 600 years suggested more
recently (Spriggs 2001).
Sand saw “Southern Lapita” as having most direct links to the rest of the
Melanesian chain, and having few links to less diverse eastern Lapita, which suggests that
“southern” may be a misnomer for the more widespread characteristics, and “Central
Lapita” might be a better term. Sand identified the major presence of composite rims on
carinated pots as a Southern Lapita typological differentiation, but also noted a similar
120

presence in Buka sites (Wickler 2001) (see comments above on Wickler’s identification
of these as occurring on bowl vessel forms). These composite rims will be shown to be a
feature of one of the collection samples in the present study, so use of these as a defining
characteristic of the Southern Lapita Series is significant to external comparison of the
Roviana Lapita (see Chapter 13). Sand notes a varied set of tooth sizes in dentate-stamping
from single sites, which contrasts with the Anson/Summerhayes “Early Far Western” fine-
dentate hypothesis in the same way as Wickler’s Buka reef site data did. Sand notes that
some parts of sites have almost all vessels decorated, while others have a relatively low
percentage. He makes a distinction between geometric and anthropomorphic designs, and
includes the stylized evolutions of anthropomorphic designs in the latter category. Of a
range of complex geometric motifs, he describes one geometric motif as “characteristic of
the end of the Lapita period”. This motif is not illustrated, but is described as “the imprint
of successions of straight lines, mostly done with a u-shaped tool”.
Of the anthropomorphic motifs, he regards “double face motifs” as rare and found
only in the earliest levels, and simpler than those found in the Western Lapita province. He
notes that some face motifs (his “Type 3") are found in the Western and Far-Western
provinces also, but not on cylinder stands and open pedestal bowls as in those provinces,
rather on carinated pots, and occasionally plates with feet. In common with Spriggs
(1990), Sand sees an evolution to an eye-nose-eye simplification, with final simplification
being the loss of the eye, but does not support this with stratigraphic or other temporal
evidence. In contrast, he states that this “evolved” motif is common from the earliest
levels, indicating a foundation style or rapid local development, but regards these “stylized
faces” as characteristic of the “Southern Lapita” dentate inventory. The rapid evolution
from the “early” faces to their “later” derivatives in these New Caledonian sites puts the
evolutionary seriations of Spriggs, Best and Ishimura under some pressure. Are the
121

“earliest levels” in these sites representing substantial periods of time, of the order of a
century or so, allowing this evolution to occur, or are the supposedly phyletic variants of
the faces contemporaneous? The way to rule out contemporaneity is to either find a
sequence with the variants conveniently arranged in levels (Best does not seem to have
evidence of this sort in his recent exposition)(Best 2002) or to seriate a good number of
large-sample low-diversify assemblages to create a fine-grained chronology, to test the
chronology with some (preferably direct) dates, and to examine the relative positions of
the variants in the chronology. (Ishimura suggests that Frimiggacci has recorded stratified
evidence for such change, but I have not had the opportunity to examine this).
Non-dentate pots are also characteristic of the “earliest levels” (incised, shell-
impressed, plain and paddle impressed). Incised decoration occurs on carinated pots, often
with a notched rim. Rounded triangles in a frieze are common. These motifs drop out “at
the same time as the dentate-stamped motifs”, and do not occur later, supporting in my
opinion a view of Lapita as a stylistic horizon rather than series on current published
evidence. This opinion is given tentatively, in the absence of published detailed ceramic
analyses from the recent New Caledonian excavations, but it should be noted that this is
a major contrast to the series proposed for the Bismarcks by Summerhayes. Incised motifs
are said to be strikingly less diverse than those in Western/Far Western Lapita. Shell
impressions are uncommon but are present in all major Lapita sites. Notched lips or
entirely plain pots are mostly carinated with excurvate rim, but some are simpler
uncarinated forms.
Paddle impressions are restricted to carinated forms with outcurved rims (although
Sand’s “carination” as illustrated is a substantially wider-angle affair than that illustrated
for dentate carinations). Paddle-impressed decoration is said to occur on better quality
paste with thinner and harder pots. Paddle-impression occurs in low percentage in “early”
sites; “...their major development takes place after the demise of the dentate-stamped and
122

incised Lapita motifs....” Sand does not explicitly discuss the role of mixing and issues of
occupation span in the association between Podtanean and early dentate Lapita but, he
gives examples where the dentate and Podtanean are separate and superposed, e.g. in
rockshelter LWT008 of Hnajoisisi of Lifou island, while elsewhere intermediate forms are
found. There seems then to be still an open question of whether Podtenaean evolved
gradually out of Lapita or whether the transition, of whatever kind, was relatively sudden,
and blurred in general by mixed deposition and lengthy occupation spans.
The View from the East: Fiji and Tonga:
Best constructed an attribute-based Robinson coefficient similarity matrix seriation for the
entire Fijian sequence based on evidence from stratified excavations on Lakeba,
supplemented by other excavated and surface collected assemblages. For the Lapita period
his data comprised the Natunuku and Naigani excavations, and the stratified Lakeba Site
196 (an extensive and rich open site thought to have a stratigraphic sequence of extended
duration with some temporal gaps and some mixing). Within the Lapita “style”,
“...the number of vessel shapes reduce from 12 to six, carinated forms
vanish, and decoration declines from initial fairly complex designs to simple
arcs and zig-zags along the rims (Best 2002:17).”
The chronological trend within the Lapita period noted by Best, the loss of carinated
vessel forms, is not in evidence in Best’s data from site 196, where these occur in all levels
of the area excavation squares (Best 1984:ppA3-A4). Strictly speaking, some carinated
forms persist in the later levels of Lakeba site 196, and it is dentate-decorated sharp
carinations which drop out rapidly from the lowest levels (Best 1984: table of attribute
combinations for site 196).
While the Lakeba excavations provide convincing evidence for a shift to arc-
impressed or arc-stamped rims over time, Best’s data on Lapita motifs is not as clear-cut.
A complicating factor making the argument difficult to abstract from the data is the use
123

of two separate systems of motif description/classification: the Mead system and the
Frost-Irwin system. Arc rims are coded in the Frost-Irwin manner as attribute
combinations, while the Mead-system motif classes are omitted from the seriations.
The Mead-system data from excavations at Lakeba site 196 is quantified by sherd
count in an appendix, and does not show clear trends by level, beyond what might be
expected from the motif sample sizes (Best 1984:A21-A22). Best stresses that if the counts
of arc rims and plain sherds were added into these data a clearer trend would emerge
(Best, personal communication 2002), but there is no clear motif shift by level using the
Mead motif data alone.
In his recent synthesis which has roulette stamping and “face” motifs early in the
Lapita series, simpler Mead-type motifs incorporating design elements such as DE1 and
DE2 (Best 1984) persisting from early Eastern Lapita to some sort of mid-Lapita, and arc-
rims as the last stage of Lapita, the supporting data for the first two stages of this transition
are seen in the similarity on the one hand of Natunuku and Naigani and the Reef/Santa-
Cruz sites (with high frequency of “face” sherds and Mead Motif M.33), and on the other
hand the Lakeba open site and most Tongan Lapita sites, being “late” with the complex
anthropomorphic motifs not found or else explicable as heirloom effect.
The low frequency of “early” motifs at Lakeba site 196 may be a sampling error.
Although 2000 sherds were recovered from the site, from a scatter of 17 test pits and two
groups of excavation squares each totaling 12m 2 (Best 1984:figure 2.16), the 15000m 2
open site may hide considerable spatial diversity in the large unexcavated areas.
Best’s suggestion in “a View from the East” (2002) that motif M12.2 is common
at Lakeba site 196 and can be traced through a sequence of decreasing complexity if
examples are contrasted with sherds from the Reef Islands is a phyletic or evolutionary
seriation that is not supported stratigraphically, as the sherd from square 12 layer A1 of
site 196 has seven underlying occurrences of the same motif, mostly from the basal layers
124

C and B3, none of which are illustrated in his “series”. Surely these sherds are relevant to
the proposed series? Picking a phyletic series from the vast area which includes the Reef
Islands and Fiji loses control of sources of variability, and becomes a re-statement of the
definition of Lapita, that Lapita simplifies over time and becomes non-Lapita, rather than
being a demonstration or confirmation of the way that change happened.
While Best demonstrates convincingly that a specific carination form is present only
in the lowest levels of Lakeba, and Best’s recent evolutionary simplification seriation may
well be correct for Fiji, and even for the entire Lapita distribution, like Spriggs'
formulation, it suffers from a circularity of reasoning. The direction of change is known
and is the same everywhere for Lapita, therefore we can pick sherds out of sites and order
them in a series, therefore we can order sites in time, and by doing this we can show that
changes occur in concert across the Lapita distribution, therefore interaction is happening,
as similar changes occur everywhere.
Best’s proposal that particular Mead motifs can be used to date Lapita sites might,
like Summerhayes’ model, be attributing the utilitarian component of sites to a late date,
and the fancy component to early Lapita. Rounded carinations are present throughout the
Lakeba strata, yet an excavation which finds the rounded carination form but not the sharp
carination form with dentate is liable to be regarded as late (I am not referring here to the
arc-rim post-Lapita decoration, which clearly supersedes Lapita at site 196).
Lapita Temporal Variability -Conclusions:
For West New Britain and Anir I have argued in detail above that Summerhayes’ chrono-
stratigraphic decorative trends are problematic, and that higher-level inferences regarding
Lapita interaction and a universal Lapita series need to accommodate these uncertainties.
This critique was made possible by Summerhayes’ thorough sherds-as-vessels analysis,
125

combined with Gosden and Webb’s geomorphological data. These trends should not be
regarded as confirming a regional West-New Britain gradual transition from dentate
stamped decoration to incised, but should rather continue to have the status of hypothesis,
in that they tell us little more than that at first there was Lapita and then there was
something else. A correlation between vessel form and dentate-stamping has been
demonstrated, but whether this varies temporally is less secure.
Summerhayes’ motif-occurrence/motif sharing clustering/PCA comparisons with
the wider Lapita world were dogged by a fundamental mismatch between the systematics
of Anson's splitting classificatory approach (which results in sparse data of unknown
representativeness), and the motif occurrence grouping methods used. Even if the
dichotomy in the data is real, is the explanation temporal? Summerhayes’ recent
chronological hygiene exercise for the Arawe and Anir dates is not conducted independent
of the ceramic temporal hypothesis, and is not independent confirmation of the
Early/Middle Lapita chronology in its present form, and is better regarded as a preferred
explanation. Some late dates from “early” contexts were rejected because they did not fit
the percieved pattern of temporal ceramic variability..
Watom: the potentially broad age-span of materials from the upper C1 SAC
excavation zone means that the recovered ceramics from zone CI and C2 are not of an
accurately known age, Also, the spread of ages can be expected to differ by level at SAC,
to an extent that suggests Watom does not yet provide a secure well-tested construct of
Lapita temporal variability against which Roviana variability can be compared.
Mussau: Detailed results of ceramic analyses are as yet unpublished, and Kirch’s
claim for stratigraphic evidence for a progression from plain to dentate to incised
decoration is subject to similar C14 and formation-process related criticisms to those
directed at Summerhayes’ Arawe analyses. Formation-based criticisms are principally that
sub-tidal mixing of area B zone C may have occurred to some extent prior to burial, and
126

that superposed incised pottery in Zone B may ultimately be derived from functionally or
taphonomically differentiated areas of the sub-tidal deposit, cast up on the beach at a later
date. Until detailed information on vessel joins, appropriateness of units of quantification
etc is available, preliminary ceramic conclusions from Mussau do not provide a detailed
construct of temporal variability to compare the Roviana ceramics with.
Admiralties: Wahome’s seriation of a large number of Admiralties sites, including
excavated and surface-collected units, provides a four-period chronology that on the
surface avoids many of the pitfalls of seriation by attention to the statistical basics. The 700
year gap suggested by Ambrose limits the utility of Wahome’s sequence for comparisons
in the present context.
Buka: Specht’s chronology from Buka to later styles, amplified by Wickler to
include more Lapita variability as evidenced by three reef-flat sites (one from Nissan), and
to tenuously link Lapita to Sohano, provided the nearest and most comprehensive ceramic
chronology of relevance both temporally, geographically, and in terms of formation
process, to the Roviana materials. His seriation of a small number of ceramic samples
encompasses, in his view, functional variability between assemblages indicated by vessel
form, and also possibly taphonomic variability in that three of his assemblages, from
various parts of site DAF, represent taphonomic variability between reef edge, reef flat,
and coastal-erosion scatter, the latter from what was regarded as formerly a terrestrial
deposit. DES from Nissan falls into this latter category too, according to Spriggs (1991).
Wickler’s seriation also faces the “number of sites” sample size problem (as does the
current Roviana data) and suffers from the use of a ratio of plain:decorated, a measure
easily biased by differences in brokenness.
Wickler’s motif frequency grouping of Buka and other assemblages does not
accord well with Summerhayes’ universal Lapita ceramic series. Despite Wickler’s
hypothesis that the sites represented some time depth, those with reasonable sample sizes
127

were much more similar to each other than to either Watom, the RSZ sites or Ile des Pins,
suggesting a greater degree of regionalization than predicted by the Summerhayes model.
Reef / Santa Cruz: use of Jaccard coefficient on sparse data of unknown
representativeness/sample size (the latter resulting from choice of units of quantification)
is a flaw in Green’s (1978) seriations. Spriggs, in his temporal hypothesis of the changing
face of Lapita, suggested that a variety of face design types were present at both SZ-8 and
RF-2, as did Ishimura, and this corresponds with my own reading of Donovan’s and
Parker’s analyses, in which, in addition to the varied face designs, there seem to be
incised/notched and other geometric linear Lapita variants not discussed by Spriggs, but
potentially of chronological significance. Parker understood these sites as sequential slices
in time, based primarily on the then radiocarbon evidence, as did Green in 1978, since
revised (Green 1991c) and currently the subject of further dating research by Green and
others. I have suggested that an alternative interpretation of these sites as roughly
contemporaneous with variable, potentially overlapping occupation spans is consistent with
the evidence. While C14 dates obtained so far from RF6 are younger than the other sites,
I would think that at least some of the material in RF6 is of a similar age to SZ8/RF2,
unless it can be demonstrated by C14 that such an overlap in occupation span is unlikely
(the portion of the RF2 sample from the northern half of that site may be key here). Best’s
suggestion that the site order should be reversed (Best 2002:93) is in my view not
supported by current evidence, and also involves a sequential slice-in-time view of the
sites.
The evidence from Vanuatu suggests that rather than a Lapita ceramic series,
evidence for ceramic variability does not presently extend beyond the general features of
the transition from Lapita to post-Lapita, and in this latter period diverse regional
sequences are seen. The samples of Lapita recovered by recent testpitting reconnaissance
and excavation are too small to allow a fine-grained temporal construction within
128

Bedford’s Lapita phase. It is tempting to see parallels between some Erromango ceramics
and some of the Roviana materials but these may be unrelated co-incidence, as will be
discussed further in a Chapter 13 (see also Felgate 2002).
New Caledonia is emerging as a region in which a process of Lapita-phase
sequence building is made possible by a large sample of Lapita sites, and the only proviso
here is that the descriptive records published to date lack the detail of, for example,
Summerhayes’ Arawe sherd analyses. Comparisons can be usefully drawn between New
Caledonian Lapita and Roviana ceramics from Honiavasa, as predicted by Sand’s
comments regarding similarities of New Caledonian Lapita to Buka Lapita.
This review of the Lapita ceramic series concludes that the regional Lapita ceramic
series from New Britain, Mussau, Buka, and the Reef-Santa-Cruz sites should be
characterized as untested hypotheses rather than securely established chronological
frameworks, and that Lapita still constitutes a ceramic stylistic horizon rather than a series
in the Western/Far Western region at least, for practical purposes of external comparisons
of the Roviana materials. Publication of details of the Mussau ceramics would probably
revise this assessment. The best samples for external comparisons are the Buka reef-site
samples, since natural formation processes are so similar (wave processes are dominant).
The terrestrial samples and buried stilt house samples are problematic in this respect, but
also suffer from chronological uncertainties which make comparisons difficult. Aspects of
the Roviana ceramics bring to mind characteristics of Southern Lapita, and characteristics
of Summerhayes’ early Lapita, but decoration is “Late” if Roviana Lapita is slotted into
the Summerhayes model. These issues will be covered in some detail in making external
comparisons in Chapter 13; it is simply noted at this stage that some difficulties arise in
trying to slot the Roviana pottery into any existing “Lapita ceramic series”. These
difficulties may arise from errors in the current formulations of those series rather than in
idiosyncratic variability in the Roviana case.
The impression given by the authors of many of these regional sequences (with the
129

exception of Anson (1983) and Wickler (2001) is that these are all confirmation of an
expected Lapita trend from dentate to incised/applied, and of the temporal primacy of the
“Far Western” sites (Kirch mentions early plainware at Mussau recently though). There is
a circularity in this reasoning stemming from a tendency to recognize evidence for the null
hypothesis, that Lapita became non-Lapita, as evidence for the details of the transition.
I conclude specifically that some attributes of pottery samples are particularly likely
to mislead in this regard, for example, using ratio of plain to decorated pottery, quantified
by sherd count, in seriations. Smash the pottery up and it gets younger. More robust
properties of ceramic samples, such as the number of shared motifs, are sensitive to
sample-size differences, particularly when units of motif classification result in sparse data,
requiring careful attention to sample evaluation, selection of units of quantification and
method of analysis.
This assessment of the nature of our current regional samples and the stability of
existing temporal constructs is not an argument against stratigraphic excavation, nor is it
a condemnation of the value of ceramic chronology in general, or an attempt to denigrate
the work of others. It is an assessment principally of the level of temporal resolution of
secure regional chronologies. I argue that while outlines of a culture-history are stable now
for several island groups, we are running up against the low temporal resolution of
reconnaissance sampling and C14 dating-by-association. Research design of the Lapita
Homeland Project focused on the identification and excavation of stratified sequences, as
has recent work on Vanuatu. There is nothing wrong with that, as long as it is accepted
that it is difficult to assess within-site spatial structure, particularly for the Lapita period.
There is a dearth of large, representative samples of Lapita from Near Oceania,
which means seriation studies, which are key to increasing the temporal resolution of
chronologies, are not succeeding as well as they might were better samples available. The
difficulty of obtaining fine-resolution stratified sequences may well be compounded by
Lapita and post-Lapita settlement pattern instability, and low-deposition life ways
130

compared to the archaeology of the Near East or American Southwest (post and thatch
in an environment of rapid organic weathering rather than mud brick houses in a dry
climate) leading to a dearth of well-stratified records of change unless there is a regular
natural airfall tephra supply, as in much of Vanuatu.
Our archaeological methods have yet to adapt to low-deposition conditions by
focussing on the horizontal component of temporal patterning, rather than expecting a
temporal layercake in settlements. Beyond deconstructing Summerhayes’s current
universal Lapita Ceramic Series, this review identifies a need to search out and analyse
archaeological distributions in horizontal space to a greater extent than has been the norm
for Lapita in Near Oceania to date. The recovery and seriation of a large number of good
short-term samples is key to boosting ceramic time-resolution. C14 dating, in this view,
can be used as a confirmatory technique rather than the primary means by which ceramic
chronology is constructed, especially with the rise of direct-dating using AMS.
131

132

CHAPTER 3:
SCALE AND METHOD OF FIELD SURVEY
REQUIRED TO GENERATE A SAMPLE OF
LAPITA SITES FOR SERIATION IN THE NEW
Introduction:
GEORGIA REGION
It is traditional to preface the analytical portions of PhD theses with a chapter on the sites
and their setting. Here, I try rather to take a problem-oriented analytical approach to the
archaeological landscape of Near Oceania, in the course of which I hope I make clear the
location and nature of surface collections used in other analytical chapters. Detailed spatial
analyses of sites are provided in Chapter 11.
While the claim could be made that the Roviana intertidal survey data were the
result of careful and systematic research design from the outset of the project, the prosaic
version of events is that research design occurred as a series of ideas arising before and
during the fieldwork. This is particularly true of the survey design, which began life as a
search for pottery sites primarily through local knowledge (hereafter referred to a the
informant-prospection method), and evolved to a more systematic inspection of the
intertidal and subtidal zones once it became apparent that these were the locus of early
ceramic evidence. Initial aims were simply site discovery. A concern with site visibility,
survey coverage, site preservation and distributional patterning crystallised in the later
stages of fieldwork, during the 1997-8 field seasons, when survey was focused in the
Kaliquongu region, with the aim of characterization of the archaeological distribution in
the intertidal zone. These two distinct stages or types of survey will be referred to as the
133

Roviana survey and the Kaliquongu survey respectively. During writing up, I saw a need
to phrase the information in terms of an estimation approach and hypothesis testing, largely
in order to demonstrate the limitations of the data in this regard, and to generate
conclusions for future practice.
While the following chapter might well be characterized as more and more about
less and less, which in terms of the size of the region it is, this is not a bad thing; the results
of iterative survey at an increasing level of detail are informative, and raise fundamental
epistemological issues for the early ceramic period in Near-Oceania. The foremost of these
concerns sampling at the level of regional survey. The Roviana Lagoon surveys are
presented below within a theoretical framework of sample-surveying. The objectives of the
surveys are first to test the hypothesis that early-Lapita was present in the New Georgia
raised coral barrier-island/lagoon system in the past, second, to estimate site density for
Lapita and post-Lapita ceramic-lithic intertidal scatters (using a ceramic chronology which
will be justified in detail in succeeding chapters), and third, to draw methodological
conclusions about the scale of survey needed to obtain a sufficient site sample for seriation
of Lapita sites. The characteristic of the local archaeological record that allows this
approach is good site visibility on coralline lagoon shorelines in the intertidal zone. The
level of preservation evident in samples, implications for survey and analysis of these
results, and what the collection sites represent in terms of settlements, etc are the subjects
of the following three chapters.
Review of Landscape/settlement Pattern Studies in Oceania:
Through the 1970s to 1990s a number of settlement-pattern studies were published, which
involved coarse ceramic periodization and an implicit commitment to synonymy of “site”
and “settlement”(Anderson 2002, Best 1984, Burley 1994, Hedrick n.d, Irwin 1972, 1985,
Kirch 1988a, Lepofsky 1988, Sand 1997b, Swadling 1976). More recently, landscape
approaches have been advocated, in which
“structures of reference are made manifest not only through sites, but
134

through broadly distributed scatters of material of the sort picked up by socalled
“off-site” approaches....understanding the overall distribution of
materials is dependent in turn on understanding the changes to which the
landscape has been subject (Gosden & Head 1994)”.
Approaches of this sort to surface scatters, in which formation processes, archaeological
visibility and site preservation are controlled for have not been applied to the archaeology
of Lapita pottery. This aspatial tendency of Lapita archaeological method, where the site,
the layer assemblage, or even the test pit tends to be the primary unit of analysis, at the
expense of regional distributional analysis (see Clark 1999:455 for a recent example)
requires further discussion, as the Kaliquongu survey described below stands in contrast
to this tendency. In near-Oceanic Lapita studies, much effort has been directed at finding
what has been described in another context as,
“The optimal solution to a culture-historical puzzle...a single point where
all of a region’s typological complexes could be found on top of each
other, separated by sterile deposits.” (Wobst 1983:42)
In archaeology elsewhere prior to 1960 this emphasis worked together with a large-site
bias to divert attention from issues of sampling and estimation, and limited archaeological
survey to a prospecting method (Wobst 1983), and this characterization would seem to
hold in near-Oceania also, particularly for the Lapita Homeland Project, but also for some
more recent studies (Bedford 2000, Clark 1999).
Wobst noted that seriation studies, in contrast to culture-historical excavation
approaches, led to a search for short-term occupation sites within short distances of each
other, which necessarily increased concern for intrasite distributions. Examples of this
sort of approach to survey can be found in the works of Specht and also Irwin (Irwin 1972,
1985, Specht 1969). Wobst criticised culture historical approaches as lacking an explicit
rationale for excavation and survey, and for an inability to evaluate the accuracy,
precision, and consistency of their results, and Gosden expressed similar sentiments in
relation to Lapita in the Bismarck archipelago (Gosden 1991a). The recent use of site
density data by Anderson in the course of assessing Lapita mobility suffers a similar fatal
135

flaw, unmitigated by the argument that,
“...the broad pattern of current distributional data has remained much the
same for more than 20 years, suggesting that it is not just on the grounds
of relative site recording intensity that the density of sites on the coasts of
large islands is lower than on small islands throughout the Lapita range, or
that mainland New Guinea, and perhaps the main Solomon Islands, are
largely bereft of Lapita sites (especially early Lapita sites) (Anderson
2002).”
Evidence of absence does not accumulate by default, nor does absence of evidence
transform into evidence of absence with the passage of time (see Felgate 2002 for a
discussion of the epistemology of the Lapita distribution in Near Oceania). It remains to
be demonstrated that the recorded density of Lapita sites is a preserved settlement pattern.
I think this is highly unlikely, and showing why this is so is a major component of this
thesis.
Early approaches to archaeology as sampling have been described as “...something
of a programmatic reaction to culture history rather than well justified applications to
resolve important research questions.”(Wobst 1983:43) Between 1960 and 1975 Binford’s
research design/sampling approach was applied predominantly in regions where there was
excellent archaeological visibility on the surface (Wobst 1983:47). A feature of most such
studies was stratified sampling using existing knowledge, and multi-stage survey strategies,
although objectives seldom extended to specifying the wider ramifications for the
distribution beyond the sample region.(Wobst 1983: 47-48). The introduction of
hypothesis-testing survey objectives in the mid 70s (for example Cowgill 1975:261)
marked a further advance in sampling research design, but such developments were still
confined to regions with good archaeological visibility. Wobst notes a trend since 1976 to
extend the new methods to areas with poor surface visibility and a consequent emphasis
on subsurface prospection techniques for such regions.
Seen in the context of Wobst’s analysis of the development of sampling
approaches to regional archaeology, a likely explanation for aspatial approaches to the
study of the regional distribution of Lapita is the environmental context of the Lapita
136

Homeland Project, the major contributor to the Lapita knowledge-base in near-Oceania.
Holocene rhyolitic volcanism and tectonic instability in the Bismarck archipelago has
created a landscape in which few archaeologists would have the temerity to claim the
pristine preservation and visibility of Lapita social landscapes. Lapita coastal site
distributions in this dynamic landscape are expected to be subject to a high degree of
obliteration or burial. Anderson points out that most Lapita sites are coastal, without any
significant discussion of the impact of coastal geomorphological processes on site
preservation and visibility. It cannot be merely
“...accepted either that the archaeological patterning is approximately
representative, or that proposition may be taken simply as a working
hypothesis (Anderson 2002:16).”
If the record is simply accepted as representative of a phenomenon as a working
hypothesis, there can be no assessment of the confidence we can have in higher-level
hypotheses built on that data. Resort to a recursive array of contingent untested hypotheses
renders archaeology a continual tabula rasa for the idiosyncratic rewriting of higher level
theory.
The sampling sophistication necessitated by areas with poor preservation and/or
poor surface visibility has been slow to develop in Lapita archaeology, and while it has
been theorized to some extent (Gosden 1991b), the requisite sampling sophistication that
would allow distributional analyses of Lapita is largely absent from existing treatments of
the data and approaches to the archaeology of the region and period. Thus the Binfordian
exhortation that archaeology should promote the region to the position of the primary unit
of archaeological research (Binford 1964) with explicit research design and explicit use of
sampling theory, has largely fallen on stony (tephra covered?) ground in near-Oceania,
where the site has remained the primary unit of analysis in most cases.
A notable exception to this generalization is the landscape approach taken on
Garua Island in the Bismarck Archipelago (Torrence & Stevenson 2000). Here the
approach to the landscape sought explicitly to abandon the coastal bias and occupation-
137

site-prospecting objectives of previous work in the Bismarcks. In spite of a comprehensive
series of C14 and obsidian hydration dates on Garua materials of roughly Lapita age, the
study faces difficulties associating ages of dated materials with the pottery styles found
with them, and the resulting chronological inferences are potentially spreading the span of
Lapita production/discard rather than refining a ceramic chronology. The Garua ceramic
sample would seem to hold considerable potential for a program of direct dating of local
pottery production styles. The focus on inland locations also initially missed significant
Lapita pottery deposits in the sea nearby, but these project-specific problems do not
detract from the value of such landscape-based approaches. In Remote Oceania there has
been some effort put into intensive systematic survey of landscapes (e.g. Best 1984, Kirch
1988a, Rogers 1973). In the case of Niuatoputapu this approach was favoured by
concentrated effort over a number of years by two research teams on an island of only
15km 2 .
The current study seeks method by which the regional distribution of Lapita-age
archaeological materials can be established and interpreted in the Roviana
geomorphological setting. While explicit research design in terms of sampling theory was
not a feature of the 1996 Roviana survey (Sheppard 1996, Sheppard et al. 1999), and data
collected reflect this, both the Roviana and Kaliquongu surveys can profitably be phrased
in terms of sampling theory when assessing the significance of results, since they clearly
are samples of some sort.
Intensity of Survey:
Wobst found in a study of archaeologists that they tended to sample a larger landscape less
intensively regardless of available resources when asked to design a survey (Wobst
1983:61). A similar tendency is proposed for Near-Oceanian archaeology, although in this
case resources have clearly been limited. Faced with large terrestrial regions, surveys have
tended to try to cover wide regions (see for example Allen et al. 1984) in an opportunistic
manner, relying largely on informant prospection as a site-discovery technique, although
138

this is not universal (as the example of Torrence & Stevenson 2000 shows). The following
section reviews the regional survey literature in the light of Wobst’s observation.
Review of Near-oceanic Survey Methods and Results:
Objectives of the Review:
One approach to archaeological sampling is to ask how much work needs to be done to
find out what we want to know (Orton 2000:5). By taking a sampling approach to the
Roviana Lagoon early-ceramic survey, the interest is not so much how much survey is
needed, but, given the amount of survey that was done, what is the significance of the
results towards answering the major research questions? The following review of Near-
Oceanic survey objectives, methods and results had these aims:
• to critically assess existing knowledge (pertaining to Lapita site survey method,
recorded site density and site preservation/detectability)
• to arrive at a characterization of existing knowledge in terms of values
(observations) of a variable (site density) measured on objects (survey regions)
To do this requires the reworking of existing information to arrive at the values of this
variable (site density). This is necessarily imprecise in some instances due to the lack of
specific detail in reports, the need for a definition of what constitutes a site, and the unit
of measurement (sites per unit area or per unit coastline length).
Method of this Review:
The literature on archaeological surveys in near-Oceania was examined for information on
survey objectives, methods and results. The coastal extent of surveys was reconstructed
as far as possible from the literature available. The unit of measurement used was site
density per km of coastline, which poses a problem of scale, since coastlines become more
indented at larger map scales. Distances were calculated using published maps of survey
139

egions in conjunction with descriptions of surveys, using dividers at a scale of either 1cm
=1km or 1cm=5km to “walk” the coastline. This method at the latter scale can be expected
to under represent the length of indented coastlines relative to the former. The effect of this
on the conclusions is discussed. Since almost none of the studies supply a definition of
what constitutes a site, structure in the resulting data needs to be discussed in these terms
too. Ideally, we should seek measures of intensity of use in behavioural measurement
devices such as estimates of discarded ceramic vessel quantities rather than site counts, but
there is no data of this sort currently available.
At the broader regional scale of the Lapita era in near-Oceania, existing knowledge
includes the archaeological record from early colonial missionaries and officials (e.g. Father
Otto Meyer), Specht’s work at Watom and Buka in the late 60s, the Southeast Solomons
Culture History project and offshoots, Chikamori’s tours, the Lapita Homeland Project and
descendants (e.g. Kirch on Mussau and Summerhayes on Anir), Terrell’s Bougainville
research (Terrell 1976), Irwin’s Shortlands survey, David Roe’s thesis research on
Guadalcanal (Roe 1993), the Solomon Islands National Museum Site recording scheme
(e.g. Miller & Roe 1982, Reeve 1989) (and equivalents in PNG, which have not been
accessed). Of these, some surveys provide little information of relevance, and have been
omitted or only briefly discussed. Others are unpublished or difficult to access, so the data
presented below are incomplete, but are data nonetheless. Recent work on Garua is
omitted from this quantitative review, as “site” densities in that case can be expected to far
exceed those recorded elsewhere due to a combination of a well preserved buried
landscape of Lapita age and a more intensive programme of “off-site” testpitting than has
been conducted elsewhere.
What distributional information regarding site density can be extracted from these
field surveys? While none are prepared to consider the results of their surveys as unaltered
and contemporaneous settlement patterns, an approach to this problem adopted here is to
look at the range of density values for the Lapita period, assuming that many of the lower
site densities will reflect low rates of site preservation or visibility (together, these create
140

low probability of detection), or low-intensity survey coverage/bad luck, and thus low
detection rate rather than absence of sites in the past. This leaves a problem in that some
of the lower site densities may be an accurate reflection of the intensity of Lapita
settlement in the past, but discriminating between absent, never present, and absent, not
found is a problem that has to be temporarily sidestepped given the embryonic state of
existing knowledge. What the data can tell us is roughly what the maximum recorded site
densities are for coarse periods.
In addition to coastline length, a secondary measure of survey region area
employed is simply to calculate the area of a rectangle on a map roughly encompassing the
area surveyed, subsuming land, sea, coastline length, arable land, reef resources, and any
other factors likely to be significant for site location frequency within this gross
measurement. Density was calculated by dividing the number of Lapita sites found by the
number of kilometres of coastline surveyed, as far as could be determined from the reports.
The definition of Lapita sites was largely left as per the report, although in the case of
Nissan (Spriggs 1991) the Halika phase site (Lapita without pots) was not counted as the
review was limited to early ceramic sites where the characteristics of the ceramics was the
primary means of assignment to the Lapita period. Survey methods were characterized in
terms of the following criteria:
• whether survey was predominantly through the informant-prospection method,
expected to create a late-site bias,
• or whether more systematic searching was conducted, expected to yield results
with less of a time-bias;
• whether systematic terrestrial pedestrian coastal searches were made, and/or
whether intertidal survey was conducted.
• A further descriptor was added regarding the intensity of survey, high, moderate
or low, which was arrived at largely through a subjective assessment of details
supplied in the reports on the amount of time spent in the survey region.
141

1984 Lapita Homelands Project Survey:
The primary aim of the 1984 reconnaissance was to explain the aims and nature of the
proposed research and to seek reaction to it. A secondary aim was to locate regions to
form base locations for individual projects in the future: there was no aim to carry out full
and extensive surveys of such localities (Allen et al. 1984:1). Nevertheless, site surveys
were carried out in Manus, the Mussau Islands and along both coasts of New Ireland,
centred at Manggai/Lemakot, Lasigi and Hilalon on the east coast, and Limpos on the west
coast. Site surveys were carried out also in the Jaquinot Bay area of East New Britain
(Allen et al. 1984:3,6-7).
Specht and Allen spent seven days in the Mussau area, searching principally the
small islands of Eloaua and Emananus. Fourteen of sixteen new sites were recorded on
these small islands. Of the two sites on the main island, one was reported by the provincial
member, while the other was spotted during a brief visit to Lomakuauru village for
purposes of consultation. These fourteen sites on the small islands comprised Lapita
middens, other or undiagnostic pottery-bearing middens, and aceramic middens. This seven
day survey of roughly thirty square kilometres (land and sea, my calculation) must rate as
one of the more intensive in Near-Oceania. An approximate measurement of coastline
length from the published map (Kirch’s section in Allen et al. 1984:146) yielded a
measurement of thirty-three kilometres, or a recorded Lapita site density of 0.09 sites per
km.
The 1984 New Ireland survey was more in the nature of a tour, with heavy
emphasis on cave sites. The Lemau pottery-bearing mounds are assumed to have been
discovered in passing by road, rather than by pedestrian survey, attention being drawn to
these high-visibility features by the favourable location of this village on a deepwater bay
(Allen et al. 1984:17-18).
Specht, Ambrose and Allen spent five days in the Pomio-Palamal region of
Jaquinot Bay, and found the area poor in archaeological resources despite good shelter in
the bay from monsoon-season seas. No pottery sherds were seen in this time. The
142

possibility of poor preservation as a result of high local rainfall was noted (5-6000mm per
annum). Their assessment of the area as potentially having a low level of detectability
means this negative result should be disregarded for the present purposes.
The Talasea area was not re-surveyed in 1984, as Specht had surveyed previously
(Specht 1974). Specht and Kamminga had surveyed a 60km stretch of coast on the north
coast of the Huon peninsula in 1972, without finding Lapita pottery or obsidian. The
discovery in 1973 at Talasea of nine coastal ceramic sites, including three identifiable
Lapita sites was therefore notable at the time, as this seemed to Specht to add weight to
an interpretation of an Oceanic distribution of Lapita, extending close to, but not to, the
north New Guinea coast.
Specht notes a total of 15-20 sites with some dentate sherds, but only three or four
sites with large quantities of dentate sherds, in a total coastal survey region of 25km,
surveyed by a mix of intensive pedestrian cover and informant prospection (the site at
Pasiloke was found by the latter method) (Specht, pers. comm. 2001). Coastal Talasea
Lapita sites were located in some cases in the intertidal (for example the FDK obsidian
source and Lapita findspot (Specht 1981), and in others on the elevated reef shore platform
slightly above sea level, and eroding into the sea. Major disturbance by bulldozers for
roading gravel seems to have been a factor in the discovery of at least site FCS (Specht
1974). Four sites in 25km yields a Lapita distribution of 0.16 sites per km, or if the larger
total is used, the site density is 20 ÷ 25 = 0.8 sites per km.
The 1984 party revisited previously reported offshore sites at Boduna and Garala
islands, and mention water-rolled pottery above the high water mark on islands then
thought to be sinking, suggested either a complex recent sea-level history or storm
deposits relocating materials from now-submerged sites. Another explanation not
considered in the report might be an intertidal distribution of settlement in the past. Large,
volcanic-tempered sherds were present in the sea at Boduna and Garala islands, and were
notable for their strength compared to the soft Mussau sherds which in contrast were
exposed on land and calcareous tempered (Allen 1984:19-20, Ambrose & Gosden 1991).
143

Recent collection of ceramics from the reef flat on Boduna (Torrence & White 2002,
White et al. 2002), separated from Talasea by a 1km channel, heightens the likelihood of
an intertidal component of the archaeological distribution in the Talasea area, rather like
the situation at Buka, where both terrestrial and intertidal ceramic sites are found.
It has been pointed out that there seems to be ample evidence that Lapita
settlement pattern at Talasea differs from the stilt-house patterning noted by Gosden for
the Arawe Islands (Specht et al. 1991). Specht notes evidence for massive landscape
changes in the Willaumez region, and also suggests that high site density is related to
intensity of survey and monitoring over an extended period (Specht, pers. comm. 2001).
Lapita Homelands Project, 1985 Season:
While the 1985 season was reported in more final form in 1991 (Allen & Gosden 1991),
some significant additional information regarding survey is contained in the 1985 field
season report. In relation to work at Balof,
“Explorations for coastal sites were also carried out. Three were found
(ELA, ELB, ELD), each with small obsidian flakes and red, shell tempered
pottery. At ELD two crenulate rims, three sherds with applied nubbins,
some incised decoration and applied strips all link this pottery with that
previously found at Lesu. However, all sites had been very disturbed, and
no undisturbed deposits were found (White et al. 1985)”
No details were given of the extent or type of survey which located these coastal sites, and
they make no appearance in the 1991 volume, although presumably do exist in the PNG
museum records.
Kirch’s group conducted further survey on the smaller islets near Eloaua, locating
seven new sites, principally aceramic middens (Kirch 1985). Kirch and Gorecki visited
Lavongai, New Hanover, for two days and surveyed beaches, river banks and valleys in
the area (Gorecki 1985). Middens and low mounds were noted along virtually the entire
Lavongai bay, with no pottery found, although the National Museum had previously
received some sherds from there. One of the research questions for the 1988 Mussau field
season was to define the chronology and nature of the transition from Lapita to post-
144

Lapita, involving excavation of an additional aceramic midden site, but no further survey
seems to have occurred by the termination of the field phase of the project (Kirch 2001:10-
24)
There is considerably more detail on survey method in the Duke of Yorks Islands
in Lilley’s contribution to the 1985 field season report (Lilley 1985) than in the subsequent
1991 volume. Intensive investigation of the western coasts of Duke of York, Makada and
Mioko Islands located six Lapita sites. Inland survey transects fostered the opinion that this
was a zone of low archaeological potential, and the eastern coastlines were cliffed or
swampy and considered unsuitable for settlement, or possibly having poor archaeological
visibility: this point is not clear. Makada, Duke of York and Mioko Islands were surveyed,
predominantly along favourable sections of the coastlines for settlement (Lilley 1991). The
survey comprised a total land and sea area of about 75km 2 , yielding a Lapita site density
of 0.08 sites/km 2 . This included about 28km of coastline, yielding a site density of 0.214
sites per km. A subsequent 5-week intensive pedestrian survey of 60km of coastline (my
measurement) located well-preserved Lapita potttery below the water table at sites SDQ
and SDP, site SEE was found to have a high density of Lapita sherds, and SEP had some
Lapita sherds in the lowest Layers (White & Harris 1997). The more exclusive definition
of what constitutes a site, compared to the earlier survey, means that site dentity decreases
compared to Lilley’s survey, to 67 sites per 1000km, or 0.025 sites per km 2 (I measured
total area, land and sea at 160km 2 ).
In the Arawe Islands, Specht spent five nights on Pililo, and three surface pottery
sites/findspots were recorded, two with Lapita sherds (Specht 1985). On Kumbun, local
residents identified three ceramic sites, Pukwo site with Madang/Sio-Gitua pottery, Iolmo
cave with similar pottery, and Emol beach flat, with similar pottery again. Arawe Island
was visited briefly, and Lapita was immediately discovered together with Madang/Sio
Gitua, and other styles, on the Makegur spit. Specht left at this point, leaving Gosden to
continue the survey (Gosden 1985).
After an unrewarding search for in situ stratified deposits in the Arawe island site,
145

Gosden spent some days excavating in Paligmete village on Pililo island, then “surveyed
the inhabited islands”, revisiting Arawe and Kumbun, and briefly visiting Maklo and
Kauptimete, where an extensive midden with obsidian and a single Lapita sherd was
discovered. A similar site with unspecified pottery was discovered on Maklo. Gosden
describes these surveys as “by no means comprehensive”. Agussak island was visited, and
a pottery site with one possible Lapita sherd was recorded. Seventeen sites (six Lapita)
were recorded on six islands totaling 110 km 2 approximately (land and sea, my calculation
from the details given in the 1985 and 1991 reports).
Gosden finally reported a cluster of seven Lapita sites within an area of around
100km 2 (land and sea) in the Kauptimete/MakIo/Kumbun/Adwe/Pililo region of the Arawe
area, although he was not specific about the extent and method of survey. This yielded a
site density of roughly 0.07 sites per km 2 (Gosden 1991b). Gosden questioned whether
these sites were contemporaneous or not, and expected other sites would be found in the
future. In spite of this caveat, he subsequently suggests the clustered distribution
comprised a social landscape (Gosden & Webb 1994). My coastline measurement yielded
a total of 42km for the islands surveyed, or 0.17 sites per square kilometre. It seems sites
with small amounts of Lapita pottery are included in these totals.
In the Kandrian district, Specht revisited terrestrial Lapita find spots on Apugi
island and conducted test excavations, without any survey coverage details beyond the
level of village names (Specht 1985). During this fieldwork he received insistent reports
of pottery on the reef at Kreslo (Specht 1985, 1991) and spent a day visiting the site. His
initial conclusion was that this was a defluved beach deposit but on reflection and in view
of results since obtained on Mussau and in the Arawes, Specht noted the difficulty posed
to this explanation by the absence of sherds on the adjacent land, and suggested stilt houses
as a possible site-formation process (Specht 1991).
For the seven hundred square kilometres or so of the Siassi Islands, a single Lapita
site on Tuam (Lilley 1986: 109&166) gave a rough site density of 0.0014 sites/km 2 , or
using my coastline measurement of 130km (probably an underestimate due to a crenellate
146

coastline drawn at a smaller scale than the others) 0.008 sites per km. Lilley noted coastline
subsidence and emergence (Lilley 1986:105) and the effects of the 1888 Ritter tsunami and
storm action on this coastline. More intensive coastal survey was conducted on the small
islands of Tuam and Malai, with a combined coastline of around 10km, and along a 10km
stretch of the northeast coast. This yielded a much higher return of one Lapita site for 10-
20 km of coast, depending on how one views the preservation potential of the northwest
coast, giving a figure of 0.1 to 0.05 Lapita sites per km.
On Nissan, three or four Lapita sites were discovered (Spriggs 1991), depending
on whether Halika-phase sites can be regarded as “Lapita without pots”. Spriggs notes that
“Systematic surveys of former village sites were not undertaken, although
coverage was good for the south part of Nissan. Most surface sites from
Tanamalit to Nahoi were destroyed by construction of a large American
base in WWII, as were some areas of Barahun island and Iachtibol.
Nachman had collected pottery from 105 named localities in the vicinity of
Balil, Siar, Salepen and Poriwan villages and on Sirot island, and so no
resurvey of these areas for village sites was undertaken, although many
such sites were noted in passing on Sirot and Northwest Nissan.” (Spriggs
1991:226)
Spriggs’ survey had a strong emphasis on rockshelters, with 45 rockshelters and 21 open
sites recorded (Spriggs 1985). Of these, the reef site at Tarmon village was the only open
Lapita site, the other two being rockshelter deposits with small samples of sherds
recovered. The atoll covers an area of about 100 square kilometres, yielding an open-site
density of 0.01 sites per square kilometre, or for an estimated 34 km of lagoon-side
shoreline (70km total including ocean-side shoreline) 0.029 sites per km.
Southeast Solomon Islands:
Survey and excavation on Santa Ana (Davenport 1972) located pottery at Feru, leading
Davenport to hypothesize an early, Lapita-related ceramic settlement, but no information
on site distribution can be derived from the report. Survey method is not reported in any
detail, and is assumed by default to comprise a tour of villages seeking local information,
with opportunistic examination of areas en route, with prospection rather than
147

distributional information as the primary objective.
On Bellona, Poulsen reported a number of terrestrial sites sealed by earth mounds,
and demonstrated that the Sikumango mound sealed deposits that contained a sherd that
“looked like Lapita” (Poulsen & Polach 1972). It is clear from Poulsen’s account and
radiocarbon dates that site visibility for this type of deposit may bear little relation to the
distribution of earth mounds, and therefore the 33km 2 Bellona survey (land and sea) can
only tell us that there was at least one (probable) Lapita site there. Survey methods seem
to have been more intensive than the subsequent reconnaissance efforts of others elsewhere
in the Southeast Solomons, with a month spent covering the 9km 2 or so of the inland
garden zone and limited coastal area suitable for settlement. Survey seemed principally
aimed at mapping the conspicuous mound and pit features, abandoned pre-WWII
habitation sites, caves in the bush, and freshwater sources on the coast; the distribution of
which were recorded in detail. The results of Poulsen’s excavations during the following
month raise the possibility of an early Lapita phase of low archaeological visibility, that
may bear little distributional relationship beyond general environmental constraints to the
mapped distribution of mound sites, which seem to have formed partly through occupation
during the period 480-1290 AD, approximately.
Southeast Solomon Islands Culture History Project:
Green conducted a week-long terrestrial survey of the Surville peninsula on Makira (San
Cristobal) in 1970 (Green 1976c), with the aim, “...to determine the types and range in
sites in this area.” The survey encompassed an area of 250km 2 (my calculation, land and
sea) or about 60km of coastline (measured from map presented in Green 1976c), and in
spite of good surface visibility of artefacts as a result of crab burrowing, and a relatively
rich archaeological landscape of late-prehistoric coastal terrestrial sites, no early ceramics
were found. Methods are assumed to have been predominantly informant-prospection.
Green regarded the area as favourable for settlement, with good canoe access to a
relatively sheltered harbour, and the complete lack of any pottery must have been food for
148

thought, particularly in comparison to the record in the Reef/Santa-Cruz group. Green
noted the dramatic effects of 1971 cyclone waves on coastal rockshelters, a comment
which now seems relevant to the question of intertidal site preservation on Makira. While
this survey is robust evidence for the absence or rarity of terrestrial Lapita pottery sites,
it cannot tell us whether Lapita was deposited in the sea in the past from stilt villages.
A series of surveys of Ulawa (Hendren 1976, Ward 1976) covered an approximate
area (land and sea) of 209km 2 (my calculation) and revealed a coastal terrestrial focus of
late prehistoric settlement. No details of prospection/survey method are reported. Hendren
noted an almost continuous distribution of tool and midden scatter in coastal areas. Ward
notes evidence for exposure of this coast to cyclone waves, and effects on terrestrial site
preservation. The occupation sites excavated by Ward postdated about 700AD. Results
suggest terrestrial pottery sites of any period are rare or absent from the extant
archaeological record.
Green’s 1970 reconnaissance survey of Uki located sites from the last 500 years
on a series of uplifted shorelines (Green 1976b). No pottery was found. In the present
context these results raise the question whether survey methods were sufficiently intense
to locate low-visibility ceramic deposits had these been formerly located in the intertidal
areas presently forming a back-swamp adjacent to the paleoshorelines.
Inland survey of Santa Cruz (Nendo) (Yen 1976) located late prehistoric or historic
occupation sites and no pottery.
Green found six Lapita sites over the 78 km 2 (land and sea?) of the main Reef
Islands (although in remote-Oceania, included for comparative purposes), and two sites
on the roughly 660 km 2 of Santa Cruz. Green reported extensive searching and questioning
of local people, without further result (Green 1976a), from which I calculated a range of
site densities from 0.077 to 0.00303 sites/km 2 . By my measurements, the main Reef
Islands, and Santa Cruz have coastlines of 33 and120km (eastern portion as per Green
p249) respectively, giving site densities of 0.182 sites per km of coast for the Reefs, and
0.012 sites per km for Santa Cruz.
149

A subsequent brief resurvey of the western end of Santa Cruz, the Graciosa Bay
area (Mccoy & Cleghorn 1988:106) made heavy use of informant-prospection. The main
emphasis of the fieldwork was on area excavations. A two-day survey of the coastline of
To Motu Neo (about 20km from the map published by McCoy and Cleghorn) located an
additional three sites, bringing the To Motu Neo total to five. There is no mention of
Lapita at the two unexcavated rockshelters, but the Novlao Rockshelter site yielded plain
ceramics and Lapita diagnostic decoration, and C14 dates spanning the Lapita period. The
intensive 20km coastal survey thus located two Lapita open sites without structural
features, and one Lapita rockshelter/open site with structural features. Site density works
out at 0.15 sites per km of coastline for an intensive survey method.
Central Solomon Islands:
Roe’s surveys on Guadalcanal had the following aims:
• to identify the range of site types known to local informants as a key to the late
prehistoric/historic settlement patterns
• prospection, to locate archaeological sites with excavation potential
• to search specifically for rock art sites and inland agricultural sites
(Roe 1993: 31)
In forested areas with dissected limestone geology, local informants were the
dominant means of prospection for rockshelter sites. In the open grassland of the
Northwest coast regional survey, site survey was conducted without prior local
information, but no further information on method was given (Roe 1993:31-32). No
pottery was seen during these surveys, and no sites of any kind were found during the
north coast survey. Roe suggested that construction of the North Guadalcanal coastal
road, clearance for plantations, and large-scale WWII bulldozing may have been major
factors in creating this archaeological blank but didn’t consider the effects of cyclone storm
surge or tsunami on site preservation along this coast.
In a review of Solomon Islands prehistory prior to his research, which was
150

considered suggestive of a different cultural history than that recorded elsewhere, due to
the absence of pottery in surface collections, Roe included a footnote to the effect that
Spriggs’ work on Nissan revealing Lapita occupation, Wickler’s Buka work that found
Lapita on the reef flats, and Reeve’s data on the intertidal ceramics in the Western
Province were all unknown to him at the time of fieldwork (Roe 1993:6, 10). What Roe’s
work does tell us is that there is a contrast in the terrestrial record between Guadalcanal
and the north Solomon Islands in the absence of late-prehistoric pottery, but absence of
evidence regarding the intertidal/reef flat record means that this contrast cannot be safely
extended back to the Lapita period given current knowledge. It seems clear enough that
Roe did not set out to find the type of reef-flat evidence found on Buka and Nissan, and
that the absence of such evidence in his results cannot be taken as evidence for absence.
Roe’s results therefore are limited to negative locational evidence, which in itself is weak
due to recent coastal landscape alterations in the area surveyed.
Cultural Resource Management (CRM) surveys by the National Museum of Solomon
Islands:
The Solomon Islands National Sites Survey listed among its objectives the assessment of
archaeological potential of each island (Miller & Roe 1982). This is a similar objective to
many of the surveys conducted within the Lapita Homeland Project. It appears that the
larger regional surveys were all predominantly reliant on the informant method of
prospection. Of these only a survey of Simbo located pottery deposits. These surveys
included the southeast Choiseul survey, the survey of the Bughotu area on Isabel, the
Kwaio area survey on Malaita, a survey of Simbo (all prior to 1978); Paripao district,
inland Guadalcanal, Santa Catalina survey, and survey of Dorio coastal district on Malaita
(between 1978 and 1980).
Within the National Sites Survey, survey in advance of construction or logging has been
more likely to turn up pottery deposits, possibly due partly to more intense and unbiased
151

survey methods. Such surveys include four surveys on Kolombangara, all of which located
pottery, coastal plain survey in the Arosi district on Makira, which located no pottery,
survey in the Viru harbour area of New Georgia which located a range of inland sites but
no pottery (given the results of the NGAS project, its absence is anomalous). Survey in
advance of mining on Vaghena resulted in test-excavations on pottery-bearing sites. Survey
in advance of logging on north New-Georgia recorded significant pottery-bearing sites.
Survey on the north coast of Nggela in advance of road construction located no pottery
sites, but a range of inland sites. Survey on the Shortland islands in advance of logging
recorded pottery sites. Miller and Roe concluded that:
“Pottery, previously regarded as being restricted in distribution to the
south-east Solomons, northern Choiseul and the Shortland Islands, appears
to have been in use at some time throughout the greater part of the western
Solomons (Miller & Roe 1982).”
None of the sites located in these surveys can be described as Lapita, although Reeve
characterized the intertidal Roviana pottery seen later at Paniavile and at other unspecified
locations as possibly Lapita-derived.
These surveys yield no information on the distribution of Lapita sites in near-
Oceania, other than some negative evidence against a terrestrial distribution. As for Roe’s
work on Guadalcanal, in view of what seem to be many examples of intertidal Lapita
settlements across the northern part of the region at present, as for many of the researchoriented
surveys, these CRM surveys offer no strong evidence for absence of Lapita in
southern near-Oceania, but may be evidence for a relative rarity of Lapita sites, or less
visible locational patterning than in the Northern region.
A number of more recent CRM survey reports are held at the National Museum
and in provincial offices, but were not accessible at the time of writing.
Bougainville and the Shortland Islands:
A 1966 survey report (Lampert 1966) was not accessed. In the Buka area, Specht
conducted an extensive coastal survey (Specht 1969), while Wickler restricted his survey
to coastal areas suitable for settlement (Wickler 1995:62-68). Wickler’s survey yielded
three Lapita sites on the reef flats, in some cases extending (as separate sites) on to shore,
152

and at least one other non-cave site (DAI) (Wickler 1995: 714-16). The Lapita site density
for the approximately 600km 2 region works out to 0.007 sites/km 2 .
Wickler’s survey aims were to locate sites suitable for excavation rather than to
achieve extensive areal coverage (Wickler 2001:63). Survey of the east coast of Buka was
by vehicle, with likely locations on raised coral flats, beaches and reef flats examined. In
common with Specht and Terrell (for the Silao Peninsula), Wickler noted the virtually
continuous distribution of archaeological materials along the raised coral portions of this
shoreline, and an arbitrary classification into sites. It would appear from the detail given
(Wickler 2001:63-68, 215-216) that around 90km was covered in this manner. For the
Lapita period, three early-phase sites were found on reef flats, and two open terrestrial
sites. For Wickler’s more intensive survey of favourable locations these results yield a site
density of 0.056 Lapita sites per km of coastline surveyed (total km rather than coastline
length of targeted locations).
A surface collection of sherds from Teop was reported in 1964 (Shutler & Shutler
1964). No details of this were available to the author. Terrell surveyed four regions on
Bougainville, Silao Area, Teop area and Numa Numa area on the east coast in
north/central Bougainville, and Paubake area of south Bougainville around Buin. Terrell
noted the pioneering exploratory nature of his surveys, and the extensive rather than
intensive approach (Terrell 1976:218). He wanted to sample extensively to find generalities
rather than the specific sequence of a local area, and wanted to include both Austronesian
and Papuan language areas in his sample (Terrell 1976:224-227). Terrell conducted a
second survey at Buin following the initial extensive reconnaissance, undertaken in view
of the divergent results obtained for this locality on the initial visit.
Methods included informant-prospection, and can be expected to have yielded
results strongly biased toward the late-prehistoric and historic settlement pattern. Site
survey on the Silao peninsula revealed the northern coast of the peninsula to be “one long
archaeological site” for the approximately 14 km of coastline surveyed, echoing Specht’s
description. Fifty-two “sites” were abstracted from this near continuous distribution, from
which almost 100 locality collections were made. No sherds in the Buka (Lapita) style
were identified, but the low number of these in Specht’s samples was considered by
153

Terrell to make sampling error a sufficient explanation for their absence. The pottery all
fell within Specht’s Sohano and later sequence.
In the Teop area, approximately 19km of coastline was surveyed, as well as some
inland journeying. Much of the coast was either mangrove flats or swampy, and only four
prehistoric pottery sites were recorded on the upraised coral terrestrial coastline, these
being styles from Specht’s sequence, back to at least the Sohano style. With the exception
of the Nanava form, thought to have been manufactured to the south near Kieta, these
were locally produced. (Terrell 1976:240). Sherds were scarce in the interior.
In the Numa Numa area, approximately 45km of coastline were surveyed, locating
32 sites in spite of low visibility caused by dense coastal bush along much of this sparsely-
occupied coast. Mararing to recent styles were found (Terrell 1976: 252-257). Terrell
considered these surveys too preliminary and incomplete to assess temporal changes in site
density.
The Paubake survey area in the south of Bougainville seemed on prior information
to contain a ceramic sequence that differed from Specht’s Buka sequence (Terrell
1976:261-263, Figure 4.7). About 8km of coastline was included in the survey. Sixteen
sites were initially recorded, including surface pottery scatters and stone constructions,
with sherds present in smaller numbers than the northern surveys. An additional 44 sites
were visited and mapped during the second survey in this region. Only one of these sites
was located on the coast,and it is not reported whether this can be considered to represent
a distributional feature of past settlement. No Lapita pottery was seen.
Terrell’s surveys in these four areas are incomplete, in his own estimation, and
cannot be taken as strong evidence for the absence of Lapita occupation in the past. All
that can be said with confidence is that nowhere in the areas surveyed is Lapita pottery an
obvious feature of the coastal terrestrial archaeological record. We can only speculate as
to whether this is to do with past behaviour patterns that contrast with areas where Lapita
pottery is a more obvious component of the record, (this could include more subtle
behavioural variants than avoidance of the area by Lapita-using people, e.g. intertidal site
location in the past) or whether site visibility and/or preservation factors can account for
such variability in the Lapita archaeological record. Given Wickler’s survey results, I
would expect that Lapita sites were there, in the sea, 3000 years ago.
154

The coastal portions of Terrell’s surveys were all terrestrial, and total
approximately 86km of coastline on Bougainville, surveyed in a low-intensity manner,
using predominantly informant-prospection and, for much of his sample, having low
archaeological visibility due to vegetation cover. All the sites recorded belonged to the last
millennium. As noted above, such survey methods can be expected to create a strong late-
site bias.
Irwin’s survey of the Shortland Islands began by surveying
“a large area in order to get some idea of the nature and variation of the surface evidence
in addition to its distribution.” (Irwin 1972:3)
Table 5: Summary of a review of survey methods and Lapita results.
SURVEY REGION SURVEY METHOD SURVEY INTENSITY COASTLINE
SURVEYED (KM)
Eloaua/ Emananus
1984
terrestrial
?
Huon Pen. ? high but unfavourable for
settlement (specht pers comm 2001)
155
LAPITA
OPEN SITES
high 33 3 91
60 0 0
DENSITY
(SITES /
1000KM)
Talasea ‘73 terrestrial intertidal high? 25 4-20 160-800
Duke of York terrestrial
?
Moderate 28 6 214
Arawe terrestrial moderate 42 7 167
Siassi terrestrial high 10 1 100
Nissan terrestrial
intertidal?
late bias for open sites
moderate 34 1 29
Bellona terrestrial, late bias high - 0 0
Makira terrestrial ? not meas 0 0
Ulawa late bias low 0 0
Uki paleo-shoreline ? 0 0
Santa Cruz terrestrial
inf. prosp.
Reef Islands terrestrial
inf. prosp.
Guadalcanal terrestrial
varies
low to mod. 120 2 12
high 33 6 182
modified coastal landscape ? 0 0
Buka (Wickler) terrestrial intertidal moderate 90 5 56
Bougainville terrestrial
inf. prosp.
Shortland terrestrial
inf. prosp.
Low 86 0 0
targeted 70 0 0
Duke of Yorks Terrestrial intensive 60 4 67
To Motu Neo Terrestrial intensive 20 3 150

Survey was concentrated on the east and south coasts of Alu Island, and appears to have
comprised a mix of informant-prospection and highly targeted terrestrial coverage of the
about two-thirds of the lengthy Alu coastline, within an overall survey block of about
500km 2 . A rough estimate of the length of coastline covered is 70km, of a similar
magnitude to the total coastal portion of Terrell’s Bougainville survey. His approach was
highly targeted towards reef-passage locations (Irwin 1972:19), and made use of
informant-prospection method. Twenty-two sites were recorded, most of which yielded
surface pottery collections, which were fitted into a hypothetical post-500AD sequence
comprising two widely-spaced early-period sites, nine closely spaced middle-period sites,
and five closely spaced late-period sites (Irwin 1972:193-196). No Lapita sites were found.
Irwin’s survey seems to provide reasonable evidence that terrestrial Lapita sites were either
rare or absent at the time of the survey, and likely to have been so in the past. No strong
evidence for the absence of either terrestrial or intertidal Lapita can be claimed though. The
metric results of this analysis of survey method and results in Near Oceania and the
Southeast Solomons are summarized in table Table 5.
Discussion of Review Results:
Many have commented on the gross patterning evident from these survey results, which
is the absence of Lapita pottery in near-Oceania south of Buka, and on the Continental
New Guinea coast (Huon Peninsula). These broad-scale discussions (see for example
Anderson 2002) have taken place within a data framework with minimal information on
or discussion of site preservation and visibility, despite comments within the individual
156

survey reports on such factors, for example on differential exposure to cyclone and
tsunami waves. With an average return for these 17 surveys of 2-3 Lapita sites each,
absence of sites through sampling error alone is a distinct possibility, even without
consideration of survey coverage, site preservation and site visibility. None of the surveys
in this review which returned a negative result for Lapita can be viewed as evidence for
absence, when survey methods and preservation/visibility are taken into account, if one
allows that settlement in the past may have been over water in the Lapita period for the
survey regions which returned negative results. It must also be remembered that there is
very little information about relative sea-level changes in most of the surveys.
For Near-Oceania, in regions where Lapita occupation has been demonstrated,
Lapita site density ranged from 29 to 800 sites per 1000km of shoreline, with an average
density of 112-183 sites per 1000km, depending on the definition of a Lapita site. The
sample of surveys includes two remote-Oceanic surveys, and seven near-Oceanic surveys.
Much of the information on which these densities are based is uncertain, the distance
surveyed in the Duke of Yorks, for example, is something of a semi-educated guess. The
definition of what is a Lapita site varies between researchers too, as does survey intensity.
There is full recognition that these figures have a high potential error and bias. That does
not mean they are meaningless though, as they result from a review of the best information
we have on the subject, the accumulated work of professional archaeologists.
There is possibly a hint of geographic decay in the data, as the lowest useful
densities were for Nissan and Buka, both to the south of the other near-Oceanic data, but
this may simply be a sample-size effect given the low number of observations, or may
157

Figure 6: Roviana and Kaliquongu surveys as samples of the New Georgia lagoon and
barrier island system.
relate to survey methodology and/or the effects of preservation and visibility on probability
of detection.
Sampling Theory and the Roviana/Kaliquongu Survey Regions:
Orton notes that an archaeological sampling region may be of any size, but should
have some geographic or cultural coherence (Orton 2000: Chapter 4). The Roviana and
Kaliquongu surveys reported below can be seen as two single building block sample
populations in two de facto regional sampling strategies, but what is the target population
(Orton 2000:18) being sampled? They could be viewed as samples of the Lapita/Post
Lapita culture region, which stretched (according to present understandings of Lapita)
from the Bismarck Archipelago to Tonga-Samoa. Roviana can also be considered as a
sample of the near-Oceanic biogeographic region, in extent from the Bismark Archipelago
158

to Makira/Ulawa. Perhaps the Roviana survey is best considered as a sample of the
geographically, geologically and tectonically coherent lagoon circuit of New Georgia
(Figure 6). Ultimately the significance of the Roviana results need to be considered at
more than one of these scales in comparison with other “building block” surveys.
This raises questions of comparability, and highlights the importance of the details
of survey methodology and assessments of archaeological visibility and archaeological
preservation for comparative data. In this respect, the Roviana survey is more comparable
to most other regional surveys than the Kaliquongu survey at the broader scales of
comparison mentioned above. The Kaliquongu survey is in some respects comparable to
the recent Buka survey (Wickler 1995) in that Wickler states that examination of reef flats
was purposive, but information in that case on site visibility and quantification of survey
coverage was not given, limiting comparative utility.
Preservation and Visibility:
In contrast to the Bismark archipelago, the geographic and geological setting of the
Roviana/Kaliquongu surveys is an area of low seismicity, and despite the presence of
active undersea volcanism off southern New Georgia, is not blanketed with any airfall
tephras. Recent work explicating sea-level history (Mann et al. 1998) suggests sherd
scatters in the intertidal zone are currently in the process of emerging from the sea into the
swash zone, and being destroyed in the process. The relatively high visibility of
archaeological sites in the intertidal zone of coralline gravel shorelines and the ease with
which sherd samples can be obtained enable a regional spatial sampling approach along
such shorelines with the aim of obtaining intersite distributional information. There are
other soft-sediment shorelines bordering these lagoons for which archaeological visibility
is much reduced. One might expect preserved ceramic/lithic scatters to be obscured to
159

some degree along such soft-sediment shorelines.
Even along the sheltered shorelines of the lagoon, archaeological preservation in
zones of high intertidal site visibility is less than ideal, as evidenced by estimates of the
parent vessel populations from which sherd samples derive (see Chapter 5), and in general
the recovered samples seem to only be a tiny fraction of the absolute quantities of material
in the minimum breakage population. There is thus great potential for site density in the
zone of high visibility to under-represent total site density, although some lithic lag-deposit
scatter should outlast the ceramics (but these are not chronologically diagnostic to the
same extent as the ceramics).
Both the Roviana and Kaliquongu surveys comprise sample regions that are
geographically most directly representative of the lagoon systems of the New Georgia
Group. While a feature of the geology of the New Georgia group is very substantial and
rapid uplift of the forearc region (Ranongga/Southern Rendova/Tetepare), the bulk of the
group, and particularly the main island of New Georgia, is more stable, surrounded by an
intricate system of elevated Pleistocene-age coral reefs and lagoons (Dunkley 1986, Mann
et al. 1998, Stoddart 1969a, b). As discussed in Chapter 1, this system of lagoons,
mainland shorelines and barrier islands forms a near-continuous circuit of New Georgia,
and is relatively unvarying in terms of the diversity of preservational and visibility
zonation over much of its length, although uplift is less at Marovo Lagoon. The area
covered by these lagoons and elevated barrier islands and the crenulate drowned mainland
shoreline total over 900km 2 (Stoddart 1969a: 387) (15,000km 2 including New Georgia
mainland and the surrounding waters as measured in the review). From the presently
published geological/geomorphological literature it would seem that the 200km 2 Roviana
lagoon is roughly comparable as a settlement/preservation/visibility context to the 700km 2
of barrier lagoons along the whole north-east coast of New Georgia, comprising Tongavae
160

Lagoon, Ngerrasi Lagoon, Marovo Lagoon, Kolo Lagoon and Kalikolo Lagoon, and also
the Panga Bay area south of Vangunu, and can be reasonably regarded as a 2/9 sample of
this area. The Kaliquongu intertidal archaeological survey is best seen as most directly
representative of intertidal patterning within this 900km 2 area, and results should only be
extrapolated to other environmental contexts with caution.
In characterizing the archaeological distribution in a sample region, the aim is to
obtain estimates of the values of variables that can be measured in principle on our target
region (Orton 2000:18). For the Roviana and Kaliquongu surveys aims were to obtain
estimates of the ratio variable site density (number of sites per km of coastline and number
of sites per square kilometre) for two coarsely defined periods, Lapita and Post-Lapita (as
measured by ceramic styles).
In summary, features of this environment pertinent to the approach taken are:
• reduced wave-exposure through the protection of the raised coral barrier
islands, leading to enhanced preservation of intertidal sites,
• a low degree of sedimentation on the barrier island shorelines, resulting in
high probability of site detection in some locations given appropriate
survey methods and reasonable preservation. Additionally, the use of a
single sample region for each of the two survey methods means the results
do not comprise estimates of New Georgia site density in a statistical sense.
One further set of geological features worthy of particular note is the number and location
of deep-water reef passages between the barrier islands. There are around 40 passages in
total but many are shallow, and the deeper of these may be favourable settlement
localities, due to navigational advantages and marine biodiversity. Reef passages give
161

access to pelagic resources in the channels surrounding New Georgia, as the outer coasts
of barrier islands are (in general) unsuited to settlement (there are modern exceptions, for
example Himo bay in the Saikile district) These geofacts and biofacts are relevant to any
archaeological sampling of the lagoon system, as their influence on late-Holocene site
location is likely to be profound.
Data Quality:
The aim of acquiring estimates of site density depends not only on good archaeological
data from Roviana Lagoon, but also on a repeatable measure of survey area and distance.
Shoreline length is a more objective measure than gross area, as there are a number of
problems in defining the areal limits of survey in an islandic environment, and the land-to-
sea ratio and area-to-coastline ratio will vary from place to place in such environments.
Shoreline length is regarded as a useful predictive unit for the New Georgia lagoon
intertidal target region, with reef-passage count less clear (not all reef passages were
created equal, despite use of this measure elsewhere (Felgate 2002).
Probability of Detection is a major archaeological sampling issue and source of non-
sampling errors (Orton 2000:26). Sites on surveyed reef flats or gravelly sediments were
considered to have an equal probability of detection, although variation in the degree of
preservation could complicate chronological assignment. Sites on soft-sediment shores
were considered to have a lower and more variable probability of detection. The absolute
values of these probabilities can only be guessed at, a problem by no means unique to the
Roviana survey (Orton 2000:39), so that no standard errors can be calculated. For hard-
sediment shorelines the probability of detection of extant deposits is probably close to
100%, although if taphonomic processes have removed ceramic stylistic information the
162

probability of detection by period is less than 100%, while for soft sediments it is probably
near to zero, unless as findspots.
If Lapita does favour reef passage locations, which tend to have hard-sediment
shorelines and high probability of detection (assuming sufficient preservation), this might
lead to a higher probability of detection for this site type in comparison to more widely
distributed site types.
Survey Methods:
Design of the Roviana and Kaliquongu surveys is considered in relation to the 12 stages
of sample survey recommended by Orton (Orton 2000:27, 76-100). These are:
1. assimilation of existing knowledge
2. objectives of the survey
3. population to be sampled
4. data to be collected
5. degree of precision required
6. method of measurement
7. the frame
8. selection of the sample
9. the pre-test
10. organization of the fieldwork
11. summary and analysis of the data
12. information gained for future surveys
163

Methods of the Roviana Survey:
Assimilation of existing knowledge:
What do existing surveys tell us about how much work needed to get the information
desired? It has been suggested throughout this chapter that pioneering informant-
prospection surveys of the type conducted on Bougainville by Terrell and in Roviana
Lagoon by the NGAS have a late-site bias. This suggests that a substantial amount of work
would need to be done to fortuitously discover a proportion of early sites. Terrell’s 86km
of Bougainville coastline surveyed in this manner failed to locate early (Lapita-era) pottery
sites (either because there never were any there or because they were not
preserved/detected by the scale and method of survey), as did Irwins’ Shortland survey,
although in the Shortlands case it must be noted that Irwin targeted terrestrial reef passage
locations, which may have improved the probability of early site detection. Current
evidence from Buka and Roviana suggests that a) this method will have a low return as a
prospection technique for Lapita sites and b) even large surveys such as Bougainville may
yield no return for the early ceramic period.
The Roviana Lagoon has an inner coastline length of 180km (estimated by walking
a 1km divider along the mapped coastline at 1:50 000 scale) and therefore is double the
amount of coastal survey conducted by Terrell by similar methods on Bougainville.
Additionally, this lagoon shoreline is not exposed to tsunami or cyclone storm swells
(discussed in detail in Chapter 6), although storm surge associated with cyclones passing
to the south through the Bellona area does affect water-levels in the lagoon, flooding the
lower shore platforms on occasion (personal observation, 1998).
164

The Roviana survey therefore represents a significant addition to the knowledge-set gained
through informant-prospection and fortuitous observation coastal survey methods.
Objectives of the Roviana survey:
The stated objectives of the Roviana Archaeological survey were to conduct a preliminary
archaeological survey of the region to determine the location and type of sites present in
this region (Sheppard 1996:4). Clearly these are not density estimation objectives (Orton
2000:79). For the purposes of the present thesis these objectives are recast as :
to test the hypothesis of Lapita settlement of New Georgia in the past, and to gain
if possible an estimate of Lapita site density. Further, to estimate the density of other early
post-Lapita sites known to be located across the target region.
Population to be sampled by the Roviana archaeological survey:
The hypothetical Lapita sites of the coastlines of the New Georgia lagoon system was the
target population to be sampled. This region comprises an area of 900km 2 approximately
(Stoddart 1969a), and is geologically and geographically homogenous and bounded
through relative uniqueness in these respects in the broader region of near-Oceania. This
was a surface survey, and for the present purposes, a survey of early ceramic sites (the
majority of archaeological evidence resulting from the 1996 field seasons was of other
types of site, but these do not directly relate to the present objectives). This region was
considered as a series of environmental zones between which site obtrusiveness,
preservation and visibility could be expected to vary. These were:
165

• the mainland shoreline, consisting for the most part recent alluvium
• the mainland intertidal, consisting for the most part soft sediments with high
preservation potential, low archaeological visibility for the sought-after site types,
and low site obtrusiveness;
• the intertidal zone of barrier-island shorelines, predominantly although with major
exceptions sandy to coralline gravel/ coralline rocky shorelines, with low site
obtrusiveness, good archaeological visibility and moderate preservation; the
terrestrial coastlines of the barrier islands, the recent emerged coral/reefal detritus
shore platform with for the most part fairly dense vegetation, low site visibility and
moderate preservation potential due to modern and possibly prehistoric gardening
activities.
Data collected:
The primary unit of measurement was the site, with offsite data recorded as findspots. For
site I use Schiffer’s definition: “a high density area of artefacts (Schiffer et al. 1978:2,14).
By findspot I mean the location of a group of three or less sherds in an area of otherwise
low ceramic density. Aceramic lithic scatters are similarly regarded as sites, while aceramic
findspots were not recorded. Late-Lapita sites were initially defined as those which yielded
ceramics of multi-part slab vessel construction and Lapita design structure (defined in
detail in Chapters 4, 8,9 and 12) (see Figure 7, Figure 8, Figure 9).
166

Figure 7: Slab-built carinated vessels from Honiavasa initially assigned to the Lapita
period (this assignment is examined in more detail in Chapters 8 and 9): the solid
vertical line in HV.2.464 represents an estimated location of the central vertical avis
(CVA) of the pot.
167

Figure 8: Carinated vessels from Honiavasa, showing double-line markes on some
design zones, bands of nubbins at the neck in two cases, and a band of fingernail
impression in one case (top).
168

Figure 9: HV.4.202 has an incised motif laid out in double lines; HV.1.314 is dentatestamped,
with similar design structure and a double carination; HV.2.341 is carinated,
with the design laid out in applied fillets, bounded horizontally by decorated lap joins
between slabs; HV.2.297 and HV.4.379 have bands of single fingernail impressions and
nubbins at the neck respectively.
169

Sites with mainly simple one-piece globular forms, together with Miho-style or
Gharanga/Kopo-style decoration and rim forms (decorative styles named for sites in which
they were first collected in quantity, locations shown below) were classified for the coarse
periodization of the current chapter as post-Lapita. This classification is examined in detail
in other chapters. These styles are defined below as follows:
• Miho style includes a diversity of vessel forms, but sharply carinated forms are
virtually absent, and simple one-piece pots generally have a thick neck decorated
most commonly with a band of opposed-pinch fingernail impression. Decorative
techniques include also incision, perforation and applique, and incised motifs are
usually made up of triangular secondary design zonation within a primary design
zone band, constructed with single lines rather than double. There is no incised
horizontal upper or lower boundary to the primary design zone band. On the rim
the triangles are usually filled by parallel lines, but are unfilled on the shoulder.
Examples are illustrated in Figure 10, Figure 11, Figure 12, Figure 13 and
Figure 14.
• Gharanga/Kopo style can be defined for the present purposes by three diagnostic
attributes, a band of punctation in the neck region and/or multiple bands of
opposed-pinch fingernail impression on the shoulder, and/or a short, heavily
everted rim. Examples are illustrated in Figure 15, Figure 16, Figure 17, Figure
18, Figure 19, Figure 20, Figure 21, Figure 22, Figure 23, Figure 24, Figure
25, and Figure 26.
170

Figure 10: Incised rims with opposed-pinch fingernail impressed band at the neck,
diagnostic of the Miho subgroup of Post-Lapita styles.
171

Figure 11: Miho-style post-Lapita sherds.
Figure 12: Miho-style post-Lapita sherds with incised shoulders.
172

Figure 13: Miho-style post-Lapita sherds with CVA measurements shown (MH290
was measured at to locations on the profile; at the interior of the neck orifice and the
exterior shoulder).
173

Figure 14: Miho-style post-Lapita sherds; dashed CVA lines are measurements based
on non-circular (uneven) curvature at the profile locations indicated by arrows.
174

Figure 15: Gharanga-style post-Lapita sherds. (Gharanga is a short-rim subgroup
of Gharanga-Kopo which may have multiple bands of opposed-pinch fingernail
impression on the shoulder.
Figure 16: Gharanga-style post-Lapita sherds.
175

Figure 17: Gharanga/Kopo style post-Lapita sherds (top) and a large Gharanga-style
sherd (bottom) (four measurement points used as arrowed to estimate CVA.
176

Figure 18: Intermediate between Gharanga and Kopo styles: all post-Lapita.
Figure 19: Kopo-style post-Lapita rim sherds.
177

Figure 20: Another large Gharanga-style post-Lapita sherd from an unrestricted vessel
form.
Figure 21: Gharanga-style small-orifice post-Lapita sherd showing rolled rim and thin
wall common to this style.
Figure 22: Kopo-Style post-Lapita sherd (taller, less everted rim than Gharanga style)
from a large-orifice vessel, with bands of impressions along both inner and outer edges
of the lip.
178

Figure 23: Less decorated variant of Gharanga-Kopo post-Lapita style, without a
strong corner point in vertical section at the neck.
Figure 24: Large-orifice Gharanga/Kopo-style sherd with deformation of the lip into a
wave pattern.
Figure 25: Large Gharanga-style post-Lapita sherd with typical decoration, including a
band of impressions along the inner edge of the lip.
179

Figure 26: Gharanga/Kopo-style post-Lapita vessels: most are weakly restricted at the
neck, with short, heavily everted rims. Punctate band at the neck is the most common
decoration in this group, while multiple bands of fingernail pinch are common on the
short-rim examples. One sherd (MH.33) had exotic quartz-calcite hybrid temper (see
Chapter 4).
180

Most collection sites yielded a mixture of these two periods to a slight extent (this will be
discussed in detail in Chapter 12), but showed predominance of one period in the styles
represented.
Degree of precision required:
Precision could not be calculated due to use of a single sample region.
Method of measurement:
Analysis of ceramic variability of surface collections was used to assign sites to period. Site
locations were recorded on 1:50000 topographic maps. Site visibility for the intertidal zone
was classified as high or low using 1:25000 air photographs (gravelly reef flats and coral
sand flats show pale or white while soft sediments show dark) combined with ground
truthing. Site density was calculated for both the total 180km of shoreline and for the
70km of high-visibility shoreline
The frame:
The sampling frame for the Roviana survey was as shown in Figure 6.
Selection of the sample:
Survey was conducted on land and in the sea as directed by informants.
The pre-test:
No formal pre-test was conducted, other than the informant-prospecting survey reported
by Reeve (Reeve 1989)
Organization of the fieldwork:
Sheppard, Felgate, Keopo and Roga spent five weeks in the Kaliquongu area of the
Lagoon in January-February of 1996, using the informant-prospection method to locate
traditional sites, rockshelters and early ceramic sites and findspots. A further season of
similar fieldwork was spent in the Saikile area, including participation by Sheppard,
Walter, Felgate, Roga, Jones, Nagaoka and informants, most notably Mr.Sae Oka of
181

Patmos, Nusahope area. An observer-bias in the results is likely in that Mr.Sae Oka had
developed a particular interest in intertidal pottery scatters and actively located a number
of these on the Ndora Island coastline, while in the Kaliquongu area the informants initially
consulted reported only one such locality, at the mouth of a mainland stream, Gharanga,
where a landowner (Mr Phillip Lanni) had developed a personal interest in the pottery and
artefacts noticed during gardening/copra activities at that specific locality (the stream is
also used today to obtain water in the dry season).
Summary and analysis of the data:
Overall survey results supported the pattern noted previously (Reeve 1989) of early
ceramic evidence located exclusively in the intertidal zone. Thin plain ceramics were found
on land, associated with late-prehistoric ritual stone structures. Similar sherds were found
in an inland midden deposit on the mainland dated to 1403-1490AD at one sigma (Site 25
in Sheppard 1996). Only sites with sufficient ceramic evidence to allow coarse temporal
characterization were counted in the analysis. Findspots comprised sites with three or less
sherds in the general locality. No aceramic lithic scatters were noticed during the survey.
Lapita sites were not initially located, although in a subsequent field season the Nusa
Roviana site was located, which had one sherd with dentate-stamping. Results were
initially interpreted (prior to the Kaliquongu survey and the Nusa Roviana dentate find) as
suggesting Lapita was absent in the New Georgia group, and possibly elsewhere in the
near-Oceanic Solomon Islands (Sheppard et al. 1999). A number of post-lapita sites were
located, including Hoghoi, Paniavile (previously recorded by Reeve), Ririgomana, Punala,
Humbi Quongu, Rangu, Gharanga, and Kopo. Intertidal ceramic findspots were noted at
Mare point, Kazu, Pikoro, Mbaraulu, and Sasavele (Figure 27).
Information gained for future surveys:
Early ceramics seemed exclusively intertidal, while late ceramics were principally found
associated with terrestrial ritual structures although a broader scatter of tiny fragments of
this thin pottery was noted in garden areas.
182

Munda
Malangari
Island
183
Saikile
Figure 27: Locations mentioned in the text in relation to Roviana and
Kaliquongu surveys.
The Kaliquongu Survey:
Objectives of the survey:
A primary reason for surveying in 1997-8 in the Kaliquongu region of Roviana Lagoon
was to test the hypothesis that Lapita pottery sites were absent. Additional objectives were
to describe the distributions of archaeological materials by stylistic period. An important
secondary objective was methodological, and in this regard the Roviana survey is regarded
as an experiment in intertidal-zone/shallow-water survey and sampling. What would the
results of such a survey be? What implications would these have for archaeology of the
Lapita era in Near-Oceania?

Population to be sampled:
As for the Roviana survey, the Kaliquongu survey region was regarded as geographically
representative of the New Georgia lagoonal system. The populations to be sampled were
the intertidal/shallow water Lapita sites of New Georgia, and the intertidal post-Lapita
early pottery sites.
Data to be collected:
The primary unit of measurement was the site, with offsite data recorded as findspots. Sites
were defined as high density scatters of artefacts, sufficiently numerous to allow a coarse
ceramic periodization. This decision was taken with recording technology in mind: the
relative remoteness of the location and lack of availability of electronic recording aids or
power supplies dictated a reduction in the spatial resolution of data recording below what
is fast becoming the norm in distributional or off-site archaeology, i.e. point-provenancing
of artefacts. Artefacts were collected rather than documented in situ, both to prevent rare
categories of finds being missed due to marine encrustation, and to enable development of
classificatory systematics in the laboratory. Some sites were collected as single samples,
some as samples from unequal sub-areas, one as a linear transect of 5x12m “squares”, and
one as a series of 1m transects at 5m intervals, each comprising a linear arrangement of
5x1m “squares”.
Degree of precision required:
In terms of the hypothesis that Lapita is/is not present in the Western Solomon Islands, the
184

prime difficulty in specifying the degree of precision required arises from unknown
probability of detection. The existing knowledge used to formulate the expected Lapita site
density was not sufficiently detailed regarding factors contributing to site detectability to
allow any precise estimate of the size of a survey region required to establish the absence
of Lapita. In this regard bigger is better theoretically, but in practice resources were
limited, as was access, imposing limits on the size of the sample region that could be
accessed and surveyed intensively in a manner that allowed biases to be assessed. The
review of 17 coastal surveys given above suggests for high-intensity survey Lapita site
density could be around 91 sites per 1000km of coastline or upwards, or on average at
least one site per 11km of coastline. This suggests that the survey region would need to
be greater than 11km of high-detectability coastline favourable to Lapita settlement to
have a reasonable chance of locating a Lapita site at the minimum recorded site density for
intensive survey.
Method of Measurement:
Site locations were recorded on air photographs or 1:50 000 topographic maps, augmented
on featureless coastlines with compass bearings to prominent landmarks, usually to nearby
islets in the lagoon. Artefacts/sherds and manuports were bagged by intrasite transect or
in the absence of transect collection, by site. Transects were set out using fibreglass
measuring tapes and marked out with nylon monofilament on stakes. Field desalination and
laboratory cleaning were important aspects of measurement, particularly of sherd
decoration and temper characterization. Site density was calculated in relation to the
185

distance of high-visibility shoreline surveyed, and as number of sites/findspots/lithic scatters
per square kilometre.
The Frame:
The frame was as shown in Figure 6, which comprised the intertidal zone of the
Kaliquongu region of the lagoon, chosen for reasons of access and because this appeared
to be a favourable environment for settlement due to presence of a deepwater barrier-island
passage and harbour, many small islets, reef shallows and seagrass flats, and a variety of
barrier-island and mainland resources within easy paddling distance of each other (Figure
28). This area encompassed 70km 2 , approximately one third of Roviana Lagoon and
approximately 8% of the total 900km 2 of the New Georgia lagoonal system.
Selection of the sample:
The sampling transect comprised the firm-sediment walkable sections of a natural
landscape unit, the intertidal zone, as shown on Figure 28. The sampling fraction was
intended to be 100% of the intertidal zone within the survey limits, but in practice some
areas were impassable, and others were not covered in the time available. About 15km of
high-visibility intertidal was surveyed in this manner, and a further 10km or so of
mangroves and soft sediments was surveyed where passable.
The Pre-test:
The 1996 field seasons had effectively provided a pre-test of methods. Site location
186

technique was initially predominantly through reports by interested local residents or
through fortuitous discovery during the process of traveling to villages by canoe to consult
with communities regarding the archaeological survey. This was regarded as
unsatisfactory, and the determination to take a more systematic sampling approach to the
landscape developed out of it. The January/August surveys of 1996 involved a substantial
terrestrial component, as did later work, which had shown a dominant terrestrial pattern
of a thin, largely plain pottery style associated with late-prehistoric archaeological evidence
(Sheppard et al. 1999:319).
Organization of the Fieldwork:
Survey was conducted during the season of low tides at the low-water spring tides of each
month. This had a major beneficial effect on site detectability. Sites were surface-collected
in most cases for laboratory classification of site type. Fieldwork assistants and guides
were recruited from Sasavele village for the Kaliquongu survey and surface collections.
Summary and Analysis of the Data:
This was as for the Roviana survey, except site visibility was noted qualitatively in the field
during survey rather than assessed from air photographs.
Information Gained for Future Surveys:
Time and cost information, intertidal site detectability under different conditions, intrasite
archaeological potential and the effects on this of grab sampling, and the solutions to
187

technical and logistical problems encountered during the survey, should all be of interest
to archaeologists interested in the Lapita ceramic series in near-Oceania. This information
will be of use to future researchers and is regarded as of immediate importance to cultural
resource managers in the region.
Results:
Results of the Kaliquongu survey are shown in Figure 28. In tabulating these results
(Table 6), two site types were lumped, aceramic lithic manuport scatters, and lithic
scatters with rare ceramic finds. Two of the aceramic scatters recorded during the
Kaliquongu survey may be of fluvial rather than anthropogenic origin, so the site density
of these site types may be too high.
Table 6: Site densities by period for Roviana and Kaliquongu surveys.
Site Type Roviana
High vis.
(70km)
Roviana
low vis.
(110km)
Roviana
Total
(180km)
188
Kaliqu.
High vis.
(15km)
Lapita 0 0 0 67/1000km
n=1
Post-
Lapita
129/1000km
n=9
9/1000km
n=1
56/1000km
n=10
333/1000km
n=5
aceramic 0/1000km 0 0 267-
400/1000km
n=4-6
findspots 43/1000km
n=3
0 17/1000km
n=3
Kaliqu.
Low vis.
(10km)
0 200/1000km
n=2
Kaliqu
Total
(25km)
0/1000km 40/1000km
n=1
0/1000km 200/1000km
n=5
0 240/1000km
n=6
80/1000km
n=2

Figure 28: Kaliquongu survey transects: the unfilled symbols represent sites discovered
by informant-prospection during the Roviana survey.
189

Discussion and Conclusions: the Two Surveys.
Disregarding the emphasis on intertidal ceramics, and considering the Lapita site density
figures obtained, the results of the Roviana survey do not seem significantly different to
those of Terrell’s Bougainville survey. Where Terrell surveyed 86 km of coastline by the
informant prospection method and found no Lapita, the Roviana Survey covered 180km
of Roviana lagoon coastline by similar methods and found a site with a single dentate sherd
(Nusa-Roviana-Zoraka). Neither do the results seem all that significantly different to
results from Bellona, Guadalcanal or Santa Cruz. It must be borne in mind that the results
of extensive, low-intensity survey using informant prospection will be highly variable
depending on informant selection, as evidenced by clustered site distribution in the Roviana
survey results (Sae Oka’s site discoveries on Ndora Island for example). Also, factors
such as coastal site preservation and particularly intertidal site preservation, have not
entered in any systematic way into discussions to date. Locating a single Lapita site in a
survey is awkward in that site density calculation clearly suffers from inadequate site
sample size. The distances surveyed in the Kaliquongu survey are low, resulting in this
potential source of sampling error.
Taking these potential sources of error in the Roviana and Kaliquongu survey data
into account, and bearing in mind also the imprecise abstraction of the comparative data
from the survey results of others, it seems impossible to conclude there is evidence for any
significant difference in terms of site density between the Kaliquongu survey and the
general Lapita record elsewhere in Near-Oceania, particularly if one allows for an
190

unknown level of site preservation. If, for example, the aceramic lithic scatters at Mbolave
and Elelo were at one time Lapita sites, Kaliquongu Lapita site density would jump to 200
sites per 1000km We would have to have much more extensive data from the New
Georgia Lagoon system encompassing a number of sites identifiably within the Lapita
series and much more intensively and systematically surveyed data from other locations in
near-Oceania to demonstrate significant differences in Lapita site density.
The implication for the question of the presence of other Lapita-era sites in the
New Georgia group is clear. A larger survey frame, with careful attention to models of site
preservation and visibility, i.e. probability of site detection, is likely to locate additional
sites from this general period. What should the sampling fraction be to obtain a sample
suitable for seriation analysis? To answer this we must first ask how many Lapita sites
might be expected from the New Georgia lagoon circuit in total (sampling fraction 100%)?
The Kaliquongu survey of 70km 2 located one Lapita site, so a best guess for the 900km 2
of coastlines and lagoons of the New Georgia lagoon circuit on current data would be
around 13 sites using this measure. Similarly, roughly 65 post-Lapita sites might be
expected, and about 78 aceramic sites. A corpus of ceramic samples of this magnitude
would allow a much better seriation analysis, which together with testing by direct dating
of styles could potentially enable a fine-grained chronology of the entire period of intertidal
site formation. A chronology of this sort would allow abandonment of the sort of coarse
Lapita/Post-Lapita chronological construct currently in use.
It is hoped that this consideration of a sampling approach to regional
archaeological questions in the near-Oceanic setting highlights the a-regional and overly
191

culture-historical status quo in Lapita studies (in the published literature at least) which
threatens to stifle the archaeological potential of the region by largely dismissing the bulk
of the Lapita record, surface or turbated deposits, as “disturbed” and therefore of low
archaeological value. It is hoped that this analysis of the Roviana and Kaliquongu surveys
will promote re-evaluation of the New Georgia early-ceramic surface record as the most
complete witness of the history of early ceramic variability. Such a re-evaluation of survey
objectives beyond a search for “good” (stratified) sites demands as a prerequisite
acceptance of the tenet that survey design matters, a simple point, but one that has largely
failed to penetrate Lapita studies.
Uncontrolled direct reading of site densities in a comparative manner (as done by
Anderson 2002) is counterproductive when the fundamental inferential issues of site
preservation and visibility, or site detectability in relation to survey methods, are simply
ignored in the rush to calculate higher-level behavioural variables like migration rates. The
poor type and extent of our sampling becomes evident when a sample-surveying
perspective is engaged. If we wish to talk about migration rates, or demographic history,
or colonization rates, or the distribution of “Lapita”, or when “Lapita” ends, research
design which obtains data matched to the research question is needed, which for any of
these behavioural questions requires that we concentrate on the difficult formational
questions of evidence, such as, “what is the probability of a pottery sample being
preserved in a particular location and observed by the survey method?’, rather than easy
questions, like “did we find anything?
192

Introduction:
CHAPTER 4:
CERAMICS: UNITS OF DESCRIPTION
The units of description employed for ceramic form and decoration were designed to
enable a detailed description of ceramic variation at the level of the sherd. The problem of
assessing sample formation processes required careful attention to the description of sherds
as sedimentary particles. Thus sherd area, mass, thickness and composition were recorded
for all sherds. Description of decoration was detailed to the level of the layout, size and
spacing of decorative elements, in an attempt to capture variation to the level of individual
potters if necessary, in the process of deciding on units of classification (classification is
covered in Chapters 8 and 9).
There is a descriptive and classificatory divide between Lapita and post-Lapita
pottery (discussed in Chapter 1), in that the Frost-Irwin attribute-combination grouping
approach has generally been applied to materials post-dating Lapita, while the Mead
system of classification, or its non-classificatory splitting variant, the Anson system of
description, have dominated Lapita studies. Studying the transition between Lapita and
post-Lapita requires a descriptive system that is capable of dealing both with complex
motifs and with simple geometric pattern such as bands of repeated elements.
The difference between Lapita and post-Lapita is already well-understood: Lapita
has complex decoration including dentate-stamping in a range of designs and a diversity
of (often slab constructed) vessel forms, some of which are characteristic. Post-Lapita
lacks these distinguishing criteria (as might pre-Lapita). If at the end of this thesis I
conclude there was a transition from dentate to incised and simplification in vessel form
I would be merely restating the definition of Lapita rather than attempting to understand
the details of ceramic change. Description in a context of reduced decorative complexity
still needs to be sensitive to the temporal dimension of variability. For Roviana pottery
this required that careful attention be paid to decorative location, as the sensitivity of
193

simple decorative techniques and simple patterns in themselves to change was not expected
to be sufficient to capture fine temporal patterning. The structure of decoration across the
vessel and in relation to vessel form was therefore of interest.
Understanding why change, as expressed in the historical events of pot-making,
happened, requires a materialist approach to the analysis of variability (Dunnell 1986:153).
An inventory of designs as in the motifs of the Anson system is not sufficient when the goal
is to understand design change. The units of description must reconstruct design into
component attributes that are salient to design evolution. The process of how one design
may have changed into another is of interest. Also, we have no reason to expect all
attributes of designs to change at the same rate. Some aspect of vessel design and
decoration might be highly constrained by functional adaptation or may be changing rapidly
as a result of some selective pressure. Also some form or decorative attributes may be free
to change stochastically, i.e. drift, in the sense of genetic drift (Dunnell 1978, 2001).
Sufficient sherds in the overall sample were large enough to permit such a
structural approach to the patterned layout of elements and the structure of this patterning
across the vessel, and some designs were compact and simple enough to allow an
element/motif design analysis. This is not to imply that elements and motifs are always
hierarchically structured: some patterns are constructed as a series of repeats of an
element, while some designs are more complex, and are just inventoried, being classified
in the next chapter according to criteria such as zone markers (Mead 1975:24). Vessel
morphology is an aspect of design that, like decoration, may be constrained or directed in
some ways but free to shift or drift in others. Analysis of vessel form, using mainly interval
measures like curve template fitting, was designed to capture slight variation, and enable
analysis within the theoretical framework of form variability in Chapter 1.
Quantitative analysis of vessel design from sherds of broken pottery requires
specification of a relationship between the pieces and inferred wholes. Some of the units
of description pertain to this requirement, and are used extensively in Chapter 5 to
ascertain the most appropriate method of quantification. In addition to behavioural-
historical and behavioural-explanatory objectives, units of description were designed with
194

the objective of understanding other non-cultural aspects of site/assemblage formation
processes. Accordingly, several measures of sherd size and thickness are recorded, in
more detail than usual, for all sherds, allowing detailed taphonomic study in Chapters 5,
6, 7 and 11.
Database Structure:
Figure 29: Data structure for each sherd record; each sherd can have many
records in the detail tables pertaining to the various parts of the vessel
represented.
The core ceramic relational database used for description of sherd attributes consisted of
a linked arrangement of tables, with the master record linked in a one-to-many relationship
to thematic tables about the sherd (Figure 29).
A master table contained data pertaining to the entire sherd. Fields of data in the
master record for each sherd were about sherd identity and provenance, area, weight, date
recorded, time recorded, general purpose notes, fabric description code, and Estimated
Vessel Equivalent (EVE) (after Orton 1993), (this is the same measurement as Egloff’s
“Percentage Factor”) (Egloff 1973) (Table 7).
195

Table 7: Structure of master record for each sherd.
FIELD PURPOSE ATTRIBUTE /
UNIT
sherd ID identify site, unit
and sherd
196
7-character
alphanumeric
METHOD OF
MEASUREMENT
re-join any postrecovery
sherd
breaks
sherd area measure sherd size cm 2 grid transparency
sherd weight measure sherd size g balance
date recorded quality control date automatic input
time recorded quality control time automatic input
notes general purpose text
fabric description preliminary
variability
special code iterative qualitative
EVE measure sherd size percent rim percent chart
Vessel Family link to other sherds
potentially from
the same vessel
vessel family
identity number
iterative
comparison
(methods detailed
in Chapter 5)
Data pertaining to the various vessel parts represented on the sherd was distributed
across a number of tables. A table of sherd thickness for each of the vessel parts
represented on the sherd was related to the master table via the sherd identity code.
Similarly, a range of data about vessel form for each part of the vessel represented on the
sherd was related to the master table via the sherd identity code. A fourth table comprised
information about decoration on each part of the vessel, again related to the master record
for the sherd via the sherd identity code. These tables are appended on CD (Master.db,
Thickness.db, Form.db and Decoration.db).
Master Record Table: Explanation of Data Codes for Each Field:
The sherd ID field was a 7-character assignment with the following site prefixes used:
R37 Paniavile (2 nd collection) (R37 had two redundant unit characters of theform1a, 2a,

P Paniavile (1 st collection)
C Paniavile (villager collection stored in cave)
A Zangana (1 st collection)
Z Zangana (2 nd Collection)
B Zangana (3 rd collection)
GH Gharanga (1 st collection)
GE Gharanga (2 nd collection- east of stream)
GW Gharanga (2 nd collection- west of stream)
MH Miho collection
HV Honiavasa collection
HG99 Hoghoi (1 st collection)
HG Hoghoi (2 nd collection)
K Kopo collection
NR Nusa Roviana collection (Zoraka)
etc (so that R371a*** is from the same collection event and
spatial unit as R372a***, R373a*** etc.)
Sherds prefixed P, C, A, B, GH, GE, GW, MH, HG99, K, and NR were from single-unit
collections. For these sherds the remaining digits indicate sherd identity within the unit.
Sherds prefixed Z, HV and HG have a collection unit number following the prefix, as
follows.
Zangana: Zxxxyyy where xxx is a three digit unit identifier and yyy is the sherd
number.
Honiavasa: HV0xyyy where 0x is unit 01-05 and yyy is sherd identity
Hoghoi :HGxxsyy or HGxxayy where ‘xx’ is the unit, ‘yy’ or ‘yyy’ is the sherd
number, ‘s’ means subsurface sample and ‘a’ means additional collection
(a control for additional time spent in collection, to prevent intensity of
collection bias in the sherd density pattern across the site). The usual form
is HGxxyyy. Where multiple fabric readings were taken as a quality control
measure there are additional records in the descriptive database with ‘b’,
‘c’ or ‘a’ added into the unit number or sherd number in place of a zero.
197

Sherd Size:
Sherd size was recorded as sherd area, sherd weight and in the case of lip and carination
sherds, estimated vessel equivalent (EVE) (Egloff 1973, Orton 1982, 1993, Orton & Tyers
1991). Sherd area was counted using a gridded PVC transparency, which could partially
conform to sherd external curvature. Measurement was made on the exterior of sherds, as
a count of square centimetres, with an accuracy estimated at ± 10%. Sherd weight was
recorded to the nearest gram using Swedish rounding in general, although initially decimal
places were recorded also. Date and time were input automatically by the database and
were used to correct some temporal inconsistencies in data recording. An additional
measure in this regard was the re-analysis of the Paniavile assemblage which had been the
first assemblage analyzed. A “notes” field was used primarily in initial database
development, to record additional information usually coded subsequently in added fields.
Fabric/Temper:
Fabric descriptions were initially made using a point-counting system, which proved too
time-consuming; and also cluster analysis on the results was inconsistent with sherd
groupings established by visual comparisons. This initial approach was discarded. Fabric
grouping of sherds was subsequently done using iterative manual sorting and visual
comparison. A corner of each sherd was snapped, and most sherds had previously been
acid-etched in the process of cleaning marine encrustation. Examination of etched sherd
surfaces and the fresh un-etched break allowed reliable discrimination between calcite
fragments, quartz/feldspar fragments, volcanic rock fragments, and Ferromagnesian
mineral fragments, when examined under 10x binocular magnification.
Iterative comparisons were used to develop a fabric type-series, but this also
proved overly time-consuming. An ordinal notational system describing type and size of
inclusions in declining order of abundance was finally used. The approach is convergent
with that of Wickler (Wickler 2001: 97-99). Major classes of grain identifiable under
binocular microscope are listed in Table 8.
198

Table 8: Classes of mineral identified at 10x magnification in reflected light.
v ferromagnesian minerals ( opaque ferromagnesian, pyroxene, olivene,
hornblende) black or green
c calcite reef detritus, white/grey, soft, often partially sintered and easily
acid-etched
q hard translucent mineral grains: quartz or feldspar
t polycrystalline lithic fragments, usually volcanic, black-grey and become
light grey-silver when etched
w white volcanic glass fragments
r red volcanic glass fragments
b black volcanic glass fragments
A microscope eyepiece was fitted with a 10x10 grid of 0.5mm squares to aid grain size
estimation. Size descriptor suffixes for the mineral groups were as follows: x (poorly
sorted sand incorporating a range of sand-sized grains from 0.1mm to 2mm), l (large), m
(medium), s (small) and f (very fine). These size classes corresponded to >0.5mm, 0.49-
0.25mm, 0.24-0.1mm and

Table 9: Examples of descriptive syntax for tempers.
VXCS Volcanic-Calcite hybrid temper comprising dominant Volcanic
ferromagnesian minerals of miXed size range with subordinate Calcite
(Small) reefal detritus grains
CSQM Quartz-Calcite hybrid temper comprising Calcite reef detritus (Small)
dominant, with subordinate Quartz/feldspar grains (Medium-sized)
VFSMCF Volcanic particles (very Fine, probably opaques) dominant with a few
Small sized and occasional Medium-sized volcanic particles,
subordinate Calcite grains (very Fine)
established for sites other than Miho and Honiavasa, and for the “P***” sample from
Paniavile, which was etched. These two Groups (1 & 3) while distinct in the sample sent
to Dickinson, form something of a continuum, where some sherds are intermediate
between the types, suggesting a more moderate degree of placer effect than seen in the
placer volcanic samples. Groups 1 and 3 are therefore regarded as subgroups of a single
class.
Group 2, un-placered hornblendic-feldspathic volcanic sand tempers, include sherds
of similar unplacered feldspathic composition, but the relative abundance of hornblende is
thought to be highly variable within this group. The sherds examined by Dickinson and
assigned to this group were all relatively rich in hornblende, and were interpreted as most
likely from Vella Lavella. Many sherds in this group have pyroxene rather than hornblende
as the predominant ferromagnesian mineral, as far as could be established from
examination of sherd surfaces under 10x binocular magnification, so this group should be
regarded as comprising a particularly low level of placering, consistent with stream sands,
rather than as being entirely exotic. Some sherds within this grouping might be exotic to
the New Georgia Island Group, but many could turn out to be local were further
petrographic analysis carried out.
200

Low-power reflected-light discrimination between Groups 4, 5, and 6 was problematic in
Table 10: Temper groupings after Dickinson 2000.
Group 1 Ferromagnesian Placer Volcanic sand tempers
Group 2 Unplacered Hornblendic-Feldspathic Volcanic Sand Tempers
Group 3 Pyroxenic-Lithic Volcanic Sand Tempers
Group 4 Quartz-Bearing Volcanic Sand Tempers
Group 5 Quartz-Calcite Hybrid Sand Tempers
Group 6 Calcite Tempers (some with possible subordinate fine ferromagnesian
component)
Group 7
(Group 1?)
Placered volcanic with plagioclase-rich microlitic rock fragments
Group 8 Temper not recorded
many cases. While there were many examples of a distinctively quartz-calcite hybrid
temper (Group 5), there were other sherds with a greater proportion of volcanic fragments,
which tended towards the volcanic-quartz grouping (Group 4). Also, there were quartz-
calcite sherds with very low amounts of quartz, difficult to detect in sherd surfaces under
binocular magnification, but which, when analyzed petrographically, turned out to have a
terrigenous fraction of similar composition to the quartz-calcite sherds (see Appendix 1:
Table 196-5) It seems clear that varying proportions of terrigenous component to calcite
reef detritus component can account for such a variety of visual temper characteristics yet
allow derivation of these materials from a single coastline (Felgate & Dickinson 2001).
Equally however, it is possible that some of the calcite-tempered sherds for which a
quartzo-feldspathic fraction cannot be discerned in the sherd surface could be from a non-
quartzose source area. Further petrographic quality-control is necessary to assess the
situation in this regard between Groups 5 and 6.
Group 7 in the database refers to a fabric in which placered volcanic temper is
dominant, but in which subordinate dark grey microlitic rock fragments occur. Due to
201

constraints on sample size, none of these sherds were included in the sample analyzed by
Dickinson. These are provisionally considered to be a less placered variant of the
ferromagnesian placered volcanic group (Group 1) with affinities also to pyroxenic-lithic
volcanic sand tempers (Group 3) and thus intermediate between Groups 1 and 3, although
calcite grains were rare or absent in Group 7, suggesting a separate origin.
Group 8 comprises those sherds for which fabric was not recorded, usually due to
difficulties discerning temper type under binocular magnification, most commonly as a
result of a dark reduction core which hid darker-coloured mineral grains in reflected light.
Vessel Parts:
Definition of vessel part terminology is fundamental to the ceramic data structure (Figure
29, Figure 30). In some instances these vessel part definitions seem more specific than
those of potters in the past, as will be argued in Chapter 9. Identification of vessel part for
smaller sherds was not always possible, and sometimes a default assignment was used
when a sherd met some criterion. For example all sherds of hyperboloid form (Rice
1987:219) other than those from Honiavasa conical-shouldered carinated jars were
classified as rims.
Thickness, Form and Decoration of Various Vessel Parts Represented on Sherds:
Three tables containing information on thickness, form and decoration were held in a one-
to-many relationship with the master table via the sherd ID field. This allowed information
pertaining to multiple specific parts of the vessel to be recorded separately, yet as a group
for that sherd. By way of example, for a sherd where the lip, rim and neck were
represented, the thickness table could contain measurements taken according to a set of
conventions at each of these points on the vessel, allowing thickness trends to be
calculated for that sherd, and also allowing comparison of assemblages in terms of
location-controlled thickness measurements. Similarly, for vessel form, characteristics of
the lip, rim, neck, shoulder, carination and body could be recorded where present for each
sherd, allowing virtual reconstruction of sherd form from the database, and also allowing
202

Figure 30: Major vessel form variants showing part
terminology; L=lip, R=rim, N=neck, S=shoulder,
C=carination, B=body; inverted or unrestricted vessels have
no neck, while for restricted vessels with everted rims (the
first seven) the only distinction in neck types is between the
double neck (top centre) and the single neck (including all
unlabelled).
detailed comparison of assemblage variability controlling for position on the vessel.
Decoration was recorded in the same way, i.e. the decorative characteristics of each vessel
part represented on the sherd were recorded in some detail. There could be more than one
decoration record for one part of a sherd, provided that the decorative technique differed,
as the key for the decoration table was a composite of more than one column.
A similar effect could have been achieved using a single flat table of data with a
203

large number of attributes or variables, but this would have been unwieldly for data entry,
requiring a multi page electronic data entry form to achieve the same level of detail as
entered in the relational system.
Thickness: For the table of sherd thicknesses by vessel part, the “ID” field linked
to the master table record for that sherd (Table 11). “Part” could be either the lip (l),
rim(r), neck(n), shoulder(s), carination(c), body(b), base(a) or unknown(u). “Part”
comprised a secondary index, to enable only one measurement for each part, and to give
each part measurement a unique identity. Thickness was measured according to the rule
for that part, using digital calipers and measured in mm, using Swedish rounding in most
cases, although some initial measurements recorded thickness to one decimal place.
Table 11: Data structure of table of thicknesses for
each part of the sherd.
Field Name Type Size Key Required Value
Id A 7 Yes
Part A 1 Yes Yes
Thickness N Yes
Thickness at the lip was measured perpendicular to the tangent of the interior surface of
the rim where it meets the lip. Thickness of the rim was measured equidistant between the
lip and the point of maximum vertical curvature of the neck (minimum vertical radius). In
the case of broken rims, where the full depth was not present, thickness was measured at
the lowest point. Thickness measurements were an average of the reading from both ends
of the sherd (in the horizontal dimension) where the sherd could be oriented. For sherds
of unknown part, which could not be oriented, thickness was recorded as the average of
several caliper readings across the sherd.
204

On the data entry form for the sherd, on which the thickness table resided, once a part had
been entered, a thickness for that part was required by the database, and use of the
secondary index meant that only one record per part could be entered.
Vessel Form:
Vessel forms were recorded using interval variables as far as possible (the set of curve-
fitting templates used provided the intervals). Secondary classification was undertaken
when these observation suggested essential classes in the data as opposed to continuous
variation. While some vessel form descriptive schemes measure the orientation of the rim
and the angle between neck and shoulder in some manner (see Irwin 1972:60: inclination
angle), the effects of both assemblage brokenness and rim/neck/shoulder curvature on the
measured angle are seldom discussed. Also, necks tend to be treated as though they are
the apex of two straight lines, these lines being the vertical sections of the rim and
shoulder. In reality, these sections are made up usually of a series of complex curves rather
than straight lines and corners.
The approach taken in this study was to regard these complex curves as reduceable
not to a series of straight lines of measured length and angles, but to a series of simple arcs
of circles of measurable length, oriented by measurable angles in relation to each other.
In my opinion this approach provides a better approximation of the actual shape of
ceramic vessels in vertical profile, and therefore does a better job of describing variation
in vessel form. This approach is less sophisticated than the two-curvature method of
Hagstrum and Hildebrand, where the exterior vertical section can be represented as a
continuously varying curve (Hagstrum & Hildebrand 1990), but at the level of sherd
assemblage description there is substantial convergence between my approach and theirs.
Curvature measurements were taken at various parts of the vessel as represented by
sherds, and covariation between interval attributes was assessed by means of bivariate
plots.
205

Field
Name
Type Size Key _Require
d Value
Id A 7 Yes Yes
Part A 1 Yes Yes
Type A 1
Angle N
Vcurve N
Hcurve N
Sherdfor
m notes
Curvatures in all cases were measured using brass radius templates, manufactured
by the Auckland University Engineering Workshop for the Anthropology Department.
Internal radii of the templates, used to measure the horizontal curvature of the exterior of
the vessel (including lip Hcurve), ranged from 50 to 300 mm in increments of 10mm (5mm
increments were manufactured but provided spurious accuracy given the hand-formed,
irregular nature of the pottery). The brass templates had a width of 7mm, so external radii
measurable by the set of (concave) curves ranged from 57 to 307 mm, also in 10mm
increments as used.
Data structure for vessel form (Table 12) used “Id” and “part” information as for
the thickness table. How the remaining fields were filled out depended on both the part and
the part form type (the first column is the list of data fields which form the columns in the
data table, the other columns detail properties of each data field).
Variations in the way different vessel part forms were recorded are listed below
Table 12: Data structure for the table of records of sherd form attributes by vessel part.
A 240
206
_Min
Value
_Max
Value
Rimdepth N 0.00 200.00

under sub-headings:
Lip Form:
For lips, nine lip “type” options were used (Figure 31), with remaining fields filled out the
same regardless of type. Lip angle was measured to the nearest ten degrees using a
goniometer, as the angle between the uppermost tangent to the interior surface of the rim,
and the innermost facet of the lip, except in the case of rounded lips, for which lip angle
was not measurable. Measuring this angle, and another which gave the degree of rim
eversion, allowed a reduction in the number of lip types recorded below the norm for
Pacific potsherd studies, without loss of information on variability.
The “Vcurve” field (Vertical curvature) was not used for describing lip form. The
“Hcurve” field (Horizontal curvature) was measured by curve fitting for 45 lips, these
tending to be the larger sherds where the measurement is less likely to be spurious. Of the
Figure 31: Lip form variants showing
database codes
207

42 sherds in this category for which EVE was recorded, EVE ranged from 3% to 17%,
with an average value of 7.2%. A larger number of lip sherds were measured in this way
during preparation of sherd drawings, where multiple measurement points were used to
establish the best location of the central vertical axis (CVA) of vessels, but these
measurements, while shown on some sherd illustration figures, were not entered into the
database. In the drawings, irregular Hcurves which did not accurately conform to any of
the curve templates were shown as a dashed CVA estimate, while solid lines indicated a
good circular fit.
“Sherdform notes” was a general purpose field used particularly in the development
stages of database design to note any data that required systematization. “Rim Depth” was
a general purpose field applicable to rims, some types of neck, and shoulders, but was not
used for lip data.
Rim Form:
Sherd ID and “part” were used as for lips. Eight rim types were distinguished (Figure 32),
and measurement rules for rims varied slightly with rim type in some cases. Generally
speaking, the rim descriptive terminology of Snow was used (Snow & Shutler 1985:20)
which is similar to that used by Irwin (Irwin 1972:55, 1985). Rims which diverge with
height from the CVA of the vessel are referred to as everted. Rims which approach the
CVA with increasing height are inverted, while those which do neither are vertical. Snow
and Shutler provide a finer descriptive terminology that subdivides everted rims into
straight everted, excurvate everted, or flared everted, excurvate having a concave exterior
208

vertical contour (hyperboloid form) and flared being convex in that plane. The “type” field
was sufficient to encompass all these variants, in addition to some sub-types of everted
rims. Eight rim types were defined (Figure 32). Rim profile (Snow & Shutler 1985:22)
was not recorded, as thickness measurements for lip, rim and neck captured this variation.
For excurvate rims (x) or straight rims (s), where the lip was present, rim angle
was recorded as the angle between the CVA and the interior tangent (in vertical contour)
at the lip. Where local profile concavity was present through thickening of the lip, the
goniometer would sit stable on two points, while on convex-profile interiors, the
Figure 32: Rim types (database codes shown and
measurement method for rim angles).
209

goniometer would be positioned so that it contacted as a tangent below the lip, and was
then rocked outwards until the point of contact reached the top of the rim (see “rim slope”
in Smith 1985:263). For flared everted rims (f), the goniometer could be rested stable
across two contact points, one at the neck and one at the rim.
This process was not particularly accurate, but had the advantage of being
measurable independent of the neck (except in the case of flared rims), and required only
that there was enough of the lip present to allow the sherd to be oriented, itself a
procedure fraught with error, particularly with uneven hand-formed pottery. This meant
a larger sample of measurements could be obtained, as sherds having both the lip and the
neck present were rare. This measurement is similar to the more commonly measured
“orientation angle” (see for example Irwin 1972:60) but not the same, except in the case
of flared or straight rims. Where the rim is excurvate this rim angle measurement will be
substantially wider than Irwin’s measure of orientation angle, dependent on the amount of
excurvature as expressed by the Vcurve measurement (see below) and the depth or height
of the rim.
For excurvate everted rims (x), brass curvature templates were used to determine
the best expression of concavity in the vertical plane (Rim Vcurve, Figure 33), recorded
as radii in millimetres, in increments of 10mm. The smallest curve template was 47mm,
curves of smaller radius were estimated using a ruler or gridded transparency. For Flared
rims (f) the curve measurement indicates the radius of concavity in the vertical plane. If
both lip and neck were present, horizontal curvature (Hcurve) was sometimes recorded
210

midway between lip and neck, on the exterior surface. No analytical use was made of this
measurement in the present study. Rim depth (Figure 33) was measured in millimetres
with calipers between the outer edge of the lip and the topmost point of maximum
curvature (minimum radius) of the neck, where the Vcurve template would be tangential
to the Vcurve template of the neck (see below). Where either the neck or the lip was not
present rim depth was noted as greater than the measured distance to the broken vertical
boundary of the rim, and recorded as a “Sherdform note” rather than in the “Rimdepth”
field.
Taken together with the lip and rim thickness measurement, these approximately
reconstruct the form of rims, and therefore allowed a more realistic record of rim form
variability, more likely to capture subtle variation than previous schemes, while boosting
sample size. As discussed in the preamble to the ceramic analysis, this level of detail
seemed desirable in attempting to characterised stylistically depauperate assemblages, and
to analyse variability in detail. For more complex rim forms, hard flare (h), compound(c),
hard compound (k) or modeled (m), measurements were made in the same way, but on the
upper portion of the rim only, so that measurements could be obtained for more broken
examples where the neck was not present. In such cases rim depth too was measured from
the outer edge of the lip to the first corner point rather than to the neck.
In practice, the advantage of being able to measure rim angle independent of the
neck was mitigated slightly by the difficulty of orienting more broken sherds which did not
include the neck. For some of the slab-built Honiavasa pottery though, which tended
211

to have broken at the neck, this system allowed more measurements than it would have
if Irwin’s or Snow’s system had been used.
Neck Form:
The vessel neck in this study was defined as the orifice restriction between body and rim
(Figure 30). Unrestricted vessels of incurvate or direct form (Snow & Shutler 1985:20)
therefore do not have a neck. The double neck vessel form (d) (Figure 30) is the only neck
type with depth, and for which neck depth must be recorded to record the shape of the
vessel. In double necks an upper and lower tight radius are separated by a distance, over
which the vertical external section is roughly straight. Other neck types are restricted (r)
or unrestricted (u), meaning there is no neck or a slight inflection (indirectness) in the
external profile. These are described only in terms of neck type, angle (Figure 33), Vcurve
(Figure 33) and Hcurve. The neck seems to have been a rugged portion of the
Figure 33: Measurement of rim depth, rim Vcurve, neck
angle and neck Vcurve at different levels of brokenness.
212

vessel, well represented in the sherd assemblages in most of the site samples, and
conveying considerable information on vessel form when thus described.
Neck Hcurve was measured on the external surface of the vessel. Where sherds
were smaller than 10cm 2 , or Hcurve was too irregular to fit any of the templates well,
measurement was either not taken or disregarded by use of a sherd size filter during
analysis. Orifice radii can be obtained by subtracting neck thickness from neck Hcurve.
Neck “Vcurve” was estimated using templates for curves of radius $47mm, and
by estimation for smaller curves. The neck Vcurve was the largest circle that could fit
vertically and snugly against the neck in the middle of the sherd (Figure 33). Minor
irregularities might cause light to show in places between the neck and the template, but
usually this measurement was a straightforward and unambiguous procedure.
The neck Vcurve measurement forms the basis for the neck angle measurement
Figure 34: Shoulder form measurement.
213

shown in Figure 33. The advantage of this approach is that the neck angle is measurable
at the neck, rather than requiring the lip to be present, as is more common (e.g. Irwin
1972:60). This means that for a relatively broken assemblage where few lip-rim-neck
sherds are present, neck angle measurements can still be obtained in quantity, yielding a
valuable increase in the sample size for this type of information on vessel form for the
assemblage. Also, this method does not treat the neck as though it is an “inflection point
(emphasis added)” (Wickler 2001:93), but rather as a curve, which is a less arbitrary
determination of where the measurement should be taken, particularly where the neck
Vcurve is gentle.
Shoulder Form:
Four shoulder types were defined (Figure 34). For soft shoulders (s), the most numerous
category, shoulder, Vcurve and Hcurve radii were defined as for necks, with one important
difference: the templates used to measure curves were too long to be of much use
measuring tight convex curves on shoulders, and Vcurve and shoulder angle were thus not
measured as accurately as in the case of necks. Soft shoulder angle measurements on small
sherds (

characteristics. Shoulder angle for hard shoulders was effectively the angle between the
tangents to the upper and lower halves of the shoulder at the apex of the halves. Notes
referring to bevel distance on these sherds are the depth measurement, recorded also in the
Rimdepth field.
For slab-built vessels where a lip-rim-neck slab is joined directly to a shallow body
(Figure 32:“k”), there was no convex (in external profile) formed shoulder and these were
therefore classified as shoulderless (l) type, where vcurve measures the concavity of the
external profile from the lower boundary of the neck to the carination/modeling junction
edge. Hcurve was either not measured or disregarded in subsequent analysis.
Where the shoulder was straight in external profile or gently convex, and typically
extended, the shoulder type was designated conical ( c ), and was expected to have, or did,
join to the maximum width of the vessel. In these cases Vcurve measured the degree of
convexity of this flattish shoulder, and the distance from the lower boundary to the neck
to the beginning of the carination was recorded in the “Rimdepth” field.
Carination Form:
Carinations could be either hard (h) (small Vcurve) or soft (s) (larger Vcurve) or double
(d) (there were two examples: see Figure 9 for one of these). Carination angle was
measured by the same rules as neck angle or shoulder angle. Carination Hcurve was
measured on the exterior plan of the vessel (horizon).
Body Form:
Body sherds were all convex in form by definition, and were classified as either of conical
215

type (c), where orientation could be established and where the vertical curvature was slight
in comparison to the horizontal curvature, or of globular (spheroidal) type (g) where both
dimensions were similar (Figure 35). Latitudinal rows of anvil marks, and horizontal
wiping marks were useful in inferring body sherd orientation in many cases. In cases where
the sherd could not be oriented, maximum and minimum curvature were entered in the
notes field of the form table, rather than in Hcurve and Vcurve fields, unless maxima and
minima were the same (spherical form). Where body sherds were smaller than 10cm 2 ,
Vcurve and Hcurve data were regarded as deriving from an unknown vessel part,
irrespective of coded vessel part, as these could equally be portions of rounded shoulders
rather than body sherds.
For those sherds that could not be oriented, but could be characterized as conical
(C) or globular/spheroidal (G), a number of body forms are possible, and some confusion
with conical shoulder forms is possible also. Relative abundance of various body sherd
Figure 35: Derivation of conical/cannister ( C ) and spheroidal (G) sherds from
various body forms.
216

forms in assemblages must be interpreted with these various possibilities in mind, and may
serve to provide a basis for establishing difference between assemblages, but not similarity.
Interior Form:
It was noticed during analysis that some interiors were smooth, with even body wall
thickness, while others had anvil marks arranged often in latitudinal circles around the pot.
This was thought to relate to whether or not a final or secondary smoothing had been
performed, using a larger anvil than that used to rough out the body shape. It seemed also
that there was some correlation between this secondary smoothing and a more distinct, or
sharper edge to the interior of the neck (orifice). To test whether this
correlation was real, the larger sherds were re-analyzed to record whether the body
interior was even (e) (Figure 36) or uneven, the latter involving the following options:
finger impressions (f), anvil impressions (a) or other (o) specified in the notes field. Also
recorded was whether the interior of the neck was sharp or rounded. These data were
subsequently filtered to exclude small sherds from analysis, to preclude the possibility of
Figure 36: Form codes for vessel interiors.
217

misinterpretation based on too small a sample of the vessel.
Unknown Part Form:
These were probably mostly soft shoulders or body sherds, most being nondescript
fragments of more-or-less globular form. Where there was some basis for orienting the
sherd, Vcurve and Hcurve were recorded. Descriptive statistics of these measurements,
filtered to sherds over 10cm 2 in area were subsequently used in analysis. Some sherds
recorded as rim sherds on the basis of their hyperboloid form (concave in one surface axis
and convex in the other) could be parts of shoulderless vessels below the neck, but this
seems unlikely, as all decorative evidence points to these being rims. Shoulders of that
form were all decorated with bounded incision, except for one vessel from the Paniavile
site which had idiosyncratic decoration, and one plain carination.
Base Form:
Very few base sherds were identified, and those tentatively. No evidence was found for flat
bases, and only tenuous suggestion of some more conical forms, and this is regarded
Figure 37: Possible conical base sherds from robust vessels.
218

as indirect evidence for predominantly rounded bases in the entire assemblage. Base sherds
were identified through having a parabolic rather than circular curvature in any dimension
around a point (see the base profile of the bi-conical form in Figure 35). One or two
sherds have slight flattening and extreme thickness also (see Figure 37 for illustrated
examples).
Decoration:
Objectives of the decorative study shifted significantly through the course of data recording
and analysis, from an initial poorly-defined aim of a detailed social/behavioural analysis,
to a position where understanding production variability was seen as the key to sample
evaluation (allowing the construction of vessel families) and to constructing a seriation
chronology. This shift was a consequence of the realization that the sherd sample displayed
a low degree of vessel completeness for all collection sites, and must have been heavily
sampled from a breakage population by post-discard processes (Felgate 2002). Nowhere
in the analysis is this shift more apparent than in the analytical scheme used to describe
surface decoration of the pottery. Initially, I was interested in identifying the work of
individual potters through examination of variation in technical execution, but it became
apparent, as will be discussed, that there was no particular reason to expect a good sample
of the work of a single potter to have been preserved in the assemblages, and the primary
focus of research shifted from the fine details of decoration to a slightly broader analysis
of production variability. There is substantial data-redundancy in the decorative
information recorded as a result of this shift, although some use was made of the mark
variability data in testing units of decorative classification (particularly for
219

impressed or notched lips).
Data structure for the decoration table are shown in Table 13. Use of the fields
“Id”, “Part” and “Type” were as for the form table of description, with multiple records
for sherds linked to the master table of sherd properties via the “Id” field. These fields, and
Table 13: Data structure for decoration records.
Field Name Type Size Key Required
Value
Id A 7 * *
Part A 1 * *
Type A 1 *
Dec_tech A 3 * *
Pattern A 3
Pattern_rep_dist N
Notes_1 A 240
Element A 3
ERD N
Ms_MarkSection A 1
Ml_MarkLength N
Mw_MarkWidth N
Md_Markdepth N
Hand A 5
Notes A 240
LinDecClass S
LipDecCode S
NeckDecCode S
a fourth, decorative technique (“Dec_tech”) were hierarchically indexed in the database
to allow each vessel part to have more than one decorative technique recorded. In practice
it was found more convenient to have multiple decorative techniques on the same part
recorded as coded combinations rather as separate lines in the table, in which
220

Table 14: Decorative techniques.
Code Description
? Decorative technique uncertain
a Applique (Figure 38, Figure 39, Figure 40)
b Burnishing or polishing (not illustrated: not used in analysis)
d Lip deformation (Figure 41, Figure 42, Figure 43)
e Excision (Figure 44, Figure 45)
f Fingernail impression (subdivided by the “element” field as
detailed below)
g Incision
h Brushing (rare, identified on the presence of multiple parallel
bristle marks sharing a common termination)
i Impression not identified as fingernail impression, usually on
lip or carination.
l Linear stamping (rare, but see Figure 12: Z.10.573, which
has linear stamping on the shoulder, in a pattern of opposed
lines forming triangles)
m Spatulate tool impression (during vessel forming)?(see Figure
46)
o Other unclassified techniques (see “Notes_1")
p Punctation (Figure 15 )
r Excision by punctation/rotation (see Figure 47)
s Surface smoothing without striation (self slip?)
t Perforation (see Figure 48)
v Anvil impression, exterior (unimportant- one unconfirmed
example)
w Wiping - identified by presence of multiple shallow rounded
parallel grooves.
y Wavy stamping or dentate stamping as specified in element
field (Figure 45 and Figure 49)
cases the predominant technique was described in detail in subsequent fields, and sub-
techniques for that vessel part were described in the notes fields of the decoration table.
The fields of decorative information were hierarchically structured from the most general
to the most detailed. Decorative techniques were coded as in Table 14.
221

“Pattern” (Table 13) is roughly equivalent to Anson’s concept of motif for more
complex patterns, (Anson 1983:57) or Mead’s concept of “alloform” (Mead 1975:23),
being defined as the structure or arrangement of decorative elements, and is defined
independent of decorative technique. Mead did not separate technique of execution and
the pattern in which any technique was applied, thus generating unnecessary proliferation
of some units of decoration, such as the various “Zone Markers” (see for example Mead
1975:Figure 2.11 page 26, GZ2-GZ4). In this study such a distinction has been maintained
as far as possible, with the objective of reducing the number of decorative classes.
Pattern entries in the decoration table fell into two classes, formulaic patterns, from
which the pattern could be generated by a formulaic repetition of elements, and more
complex patterns described using sherd illustrations. The formulaic patterns are mostly
simple latitudinal bands of decoration (pattern prefix “b” in the field “patt” in the
decoration data table) running around the vessel parallel to the vessel horizon (with the
central axis of symmetry oriented vertically). Pattern definitions are given in Table 15. The
complex patterns were later classified into two classes based primarily on zone markers,
with a third dustbin class, mostly for pattern fragments.
“Patt_rep_dist” (Pattern repeat distance - Table 13) was a measure (in mm) for simple
formulaic patterns of the vertical distance between element repeats (from vertical centre
to vertical centre). Thus in a pattern comprising multiple latitudinal bands of opposed pinch
fingernail impression, the pattern repeat distance was a component in specifying the overall
expression of the pattern on the sherd.
222

Figure 38: Examples of applied decoration.
Figure 39: Examples of applied decoration.
223

Figure 40: Applied decoration on compound rims.
Figure 41: Examples of deformation of the lip into a discontinuous band.
224

Figure 42: Horizontal deformation of the lip into a continuous band in a wave pattern.
Figure 43: Examples of discontinuous deformation.
225

Figure 44: Excision of outer lip to form a band of notches.
Figure 45: Compound rims from Nusa Roviana. NR.34 has excised lines forming the
triangle pattern on the upper rim.
226

Figure 46: Impressions on the top of the lip; and spatula impressions on the neck,
thought to indicate forming of the neck using a tool.
Figure 47: Excision by rotation: one end of a small twig or rod has been poked into the
clay and the other end moved in a circle, leaving a conical hole.
Figure 48: Perforation (upper hole).
227

Figure 49: Examples of wavy stamping and an example of dentate-stamping.
Figure 50: Applied decoration (in combination with incised decoration) with
detachment scars indicating a v-pattern, and also possible attached disc.
228

Table 15: Decoration pattern definitions
“Patt” Description Anson Motif # Count
? incomplete, insufficient to characterize 284
?pi (continuous?) band of parallel elements on inner edge of part 1
?po (continuous?) band of parallel elements on outer edge of part 1
?pt (continuous?) band of parallel elements on top face of part 1
acc accidental marks? 1
am1 (complex applied pattern) see R37.11A.1 in Figure 40 2
as1 (complex applied fillets) see R376a14 in Figure 38 1
as3 see HV.1.3 in Figure 49 3
as? (applied fillet ?) Not illustrated 1
avc (applied v-shaped fillet) see Figure 50 1
b continuous latitudinal band of an element (used for circular elements such as punctations or nubbins) 117
bar incomplete, twinned vertical bars made up of fingernail impressions 1
bd continuous latitudinal band of elements oriented diagonally to horizon (see Figure 51) 16
bdi band, elements diagonally oriented, inner edge of lip (not illustrated) 4
bdo band, elements diagonally oriented, outer edge of lip (equivalent to “bpi” below) 8
bdt band, elements diagonally oriented, top face of lip (equivalent to “bpi” below, see Figure 57) 3
bnd fragment of complex pattern? Repeats of bent lines, see MH266 (Figure 52) 272?, 273?, 498 sans zone markers? 3
bo band, outer edge of lip (as for “bpi” below, but for circular elements) 6
bot band of parallel elements in diagonally, symmetrically opposed sets, top face of lip (Figure 57) 1
bp band of parallel elements, most commonly fingernail pinching but sometimes applied fillets (Figure 7) 124
bpb bands, parallel elements, outer and inner edges of lip (Figure 19: Z.89.549, Figure 22, Figure 26: GE266) 16
bpi band of parallel elements, inner edge of lip (Figure 54) 39
bpo band of parallel elements, outer edge of lip (Figure 55, Figure 56) 69
bpt band, parallel elements, top face of lip (Figure 57) 38
bt band of elements on the top face of the part (applies to circular elements such as punctation) 4
c10 complex linear pattern: see HV.4.164 (Figure 8) similar 154, 155 4
c11 complex linear pattern: see HV.4.202, HV.2.302, HV.2.297 (Figure 8, Figure 9) 155 3
c12 complex linear pattern: see HV.2.227 (Figure 49) 1
c13 complex linear pattern: see HV.1.314 (Figure 9) 2
c14 complex linear pattern: see HV.02.419 (Figure 58) 155 in fingernail impression? 4
c17 complex linear pattern: see MH 169 (Figure 13: MH169) 4
c18 basic linear pattern: see Z.61.654 (Figure 12) 162 sans zone marker 16
c20 complex linear pattern: see sherd MH360 (Figure 59) similar 163 1
c21 complex linear pattern: see GE 155 (Figure 60) 1
c23 complex linear pattern: see Z.49.751 (Figure 61) 1
c24 complex linear pattern: A11 (Figure 59) 1
c25 complex linear pattern: similar to ch5, see A.2 (Figure 62) 1
cf (crow’s foot) see P205 (Figure 53) compare 197-223 1
ch3 complex linear pattern: see R37.11A.16 (Figure 62) 3
ch5 complex linear pattern: see R37.2A.14 (Figure 64) 157-159 sans zone marker, 307 4
ch6 complex linear pattern: GW.258 (Figure 59) 2
ch7 complex linear pattern: incomplete, see HG.17.8 (Figure 63) 2
ch9 complex linear pattern: see HV.2.124 (?), HV.2.464 (Figure 7), HV.4.175 (Figure 49) 169161 3
ch? complex linear pattern: in triangular pattern category but too small to determine 32
chd complex linear pattern: see MH.185 (Figure 61) 1
chv complex linear pattern: see Z.60.326 (Figure 65) and MH.14 (Figure 66) 187, 188, 190 sans zone markers 45
chz complex linear pattern: see HG.99.061 (Figure 59) 1
cl1 complex curvilinear pattern: R37.7a.21 (Figure 67) 1
cl3 complex curvilinear pattern: see 3711a22 (Figure 67) compare 259 1
cl4 complex curvilinear pattern: see HV.5.214 (Figure 7) 259 with zone marker added? 1
cl5 complex curvilinear pattern: see HV.04.379 (Figure 9) compare 133, 141, 494, 496 1
cl6 complex curvilinear pattern: see GW.159 (Figure 67) 1
cl7 complex curvilinear pattern: see sherd C.14 (Figure 67) 1
cl? curvilinear pattern fragments, not illustrated 3
crw complex linear pattern: see C.101 (Figure 68), post-depositional damage? 1
ct3 (cross-hatch 3) see MH.290 (?), R37.9A.14 (Figure 62), MH.69 (Figure 69) see 230, 237, 7
ctw (cross-hatch/wavy) see MH.290 (Figure 13) 1
cvh (chevron/hatch) see MH.273 (Figure 70) 2
d pattern unknown, element oriented with long axis diagonally oriented to CVA 4
gm3 complex linear geometric pattern: see HV.2.313 (Figure 8) compare 278, 279, 280, 283, 325, 441-448 1
gm4 complex linear geometric pattern: see HV.1.369 (Figure 7) 1
gm5 complex linear geometric pattern: see HG.11.72 (Figure 71) 4
gm6 complex linear pattern: see A.23 (Figure 60) 1
gm7 complex linear pattern: ref R37.7a..13 in data files, faint thin lines, not illustrated 2
gm8 complex linear pattern: See NR.34 (Figure 45) 1
l lateral element, discontinuous, e.g. striation or brush mark 118
lbi transitional between bpt and bpi (one record only) 1
mb multiple stacked continuous latitudinal bands (e.g. Figure 26: HG.21.2) 29
oci discontinuous (widely spaced?) element placed inner edge of part 41
oco discontinuous (widely spaced?) element placed outer edge of part 2
olb vertical trails of paired single impressions (1 example) 1
pc1 complex linear pattern:(?pictorial incised ) see MH44 (Figure 61) 2
pc2 complex linear pattern: see A.24, (Figure 61)
pwc complex linear pattern: see GW055 (Figure 39) 2
rbr complex linear pattern: see R37.11A.2 (Figure 62) 2
rl1 see HV.5.210 (Figure 7) 1
sfl complex linear pattern: see HV.04.185 (Figure 72) 1
wav horizontal wave (created by deformation or impression) (see Figure 42) 96

Figure 51: A band of fingernail impressions (opposed pinch), each of which is oriented
diagonal to the CVA rather than vertically, or parallel to the CVA.
Figure 52: Deformation, perforation, and the “bnd” incomplete pattern example.
Figure 53: “cf” (Crow’s foot) pattern on the vessel rim above a band of pinching.
230

Figure 54: Examples of lip impression in pattern “bpi” (band parallel inner-edge of lip).
Figure 55: Examples of lip impressions laid out in pattern “bpo” (band parallel outeredge).
231

Figure 56: Fragmented examples with lip impressions assigned to “band parallel outer”
pattern.
Figure 57: Examples of lip impressions/incision laid out in patterns “bot” (band
opposing top), “bdt” (band diagonal top) and “bpt” (band parallel top), which were
regarded as equivalent in analysis due to non-exclusive nature of these descriptions.
232

Figure 58: Pattern expressed using linear arrangements of fingernail pinching.
Figure 59: Miscellaneous incised patterns: MH360 is middle row, left-hand column.
GW258 is bottom-right.
233

Figure 60: Examples of applied nubbins and some bounded incised patterns (MH259 is
an unbounded pattern).
Figure 61: Decorated sherds with quartz-calcite hybrid granitic temper.
234

Figure 62: Unbounded incised patterns.
Figure 63: Thin incised rims from Hoghoi.
Figure 64: Unbounded incised pattern on the shoulder from Paniavile.
235

Figure 65: An example of “chv” pattern, a band of unbounded linear triangles filled by
alternating fields of parallel lines.
Figure 66: Example of “chv” pattern on tall rim (pinching at neck).
236

Figure 67 Curvilinear incised patterns.
Figure 68: Crow’s foot mark, probably post-deposition scratching.
Figure 69: Example of cross-hatch pattern “ct3”.
237

Figure 70: Cross hatch pattern “cvh”.
Figure 71: Example of pattern “gm5".
Figure 72: Sole example of pattern “rl1", a double-line arrangement of single repeated
fingernail impressions.
238

For complex patterns expressed graphically, pattern repeat distance was the distance
latitudinally between adjacent homologous points, where the sherd was sufficiently large
to measure this. Pattern repeat distance was not used analytically and is included in case
somebody is interested in this property of the assemblages in the future. The number of
vertical repeats of patterns comprising stacked multiple bands was recorded along with
miscellaneous other information in the “Notes_1" field of “Decoration.db”, and could
easily be systematized into a separate field should anyone so wish in future.
Decorative Elements:
The “Element” field specifies that which is repeated in the formulaic patterns: for example,
in the pattern “bp” (lateral band parallel) the pattern consists of repeated elements oriented
parallel to each other and perpendicular to the horizon of the band, and might be an
opposed-pinch fingernail impression (“opc”), or a deformation of the lip into a horizontal
wave (“wav”), or a single impression using a tool or fingernail (“i”). The element
represents a single manual decorative operation on the part of the potter, each pinch of the
fingernails, each impression of the tool, each deformation of the lip, repeated according to
the pattern. The element repeat distance (ERD) specifies the average horizontal spacing,
in mm, between element centres as calculated from two or three measurements, using
calipers, on the sherd. These data were not used in the current analysis, other than for lip
impressions but provide a level of descriptive detail that might be of use to future analyses,
so were included. Element codes and their meanings are given in Figure 72.
239

Table 16: Decorative elements.
Eleme
nt code
Description count in table
“Decoration”
No element identified (eg surface treatment records) 218
as applied "sausages" or elongate fillets of clay 26
ds dentate stamp 7
fi finger/fingertip impression 44
hol hole 82
i linear edge impression not obviously fingernail 141
l linear element (e.g. incision) 239
nub circular nubbin 19
opc opposed pinch fingernail impression 179
pi paired linear impressions 1
psm prismatic or triangular section applied fillet 3
sin single crescentic fingernail impression (not pinched) 27
t applied triangle 1
tfm latitudinal tool forming mark (spatula?) 94
tw? Tool forming marks or striations from wiping? 13
wav wave form, repeated shear-deformation of the lip 84
wip wiping or brushing marks 11
ws wavy stamp, probably a shell-edge impression 3
wv3 a closely spaced paired repeat of “fi” deformation of the lip 8
The fields “Mark Length”, “Mark Width” and “Mark Depth” were used to record
dimensions of the decorative elements, with maximum depth recorded perpendicular to the
surface of the sherd, using a sharpened depth-gauge, in mm, as a positive value regardless
of whether decorative element was raised relief, as in the case of applied decoration, or
sunken, as in the case of incising or punctation.
Measurement conventions for these variables for each element are as given below:
240

Applied fillets (when complete) were measured across their longest and shortest
dimensions for length and width, with depth measured perpendicular to the sherd exterior
surface at a point of average depth. dentate-stamp overall length, average width and
average depth were recorded where measurable. Tooth sizes were not recorded but scale
drawings of all dentate-stamping are provided (the seven records for dentate-stamping in
the decoration table arise from four sherds with confirmed dentate-stamping and two
sherds with possible dentate-stamping: one sherd yielded two zones of dentate-making a
total of seven). Fingertip/finger impressions were measured when present as lip
deformations, with length measured around the lip, width measured from the top of the lip
to the lowest point of the impression. Depth was measured parallel to the top face of the
lip, with the base of the depth-gauge resting across the length of the impression, and the
gauge located at the deepest point. Length of holes was measured in their longest
dimension, and width was measured at right angles to the length measurement; depth was
gauged at the deepest point; several readings were averaged where there was variation.
Edge impressions (proportionately deep v-shaped impressions) were measured as
for applied fillets, except for depth, which was measured as for holes. Linear elements were
measured where length was standard for the pattern (e.g. for pattern c18). Nubbins were
measured as for holes, except that depth was measured as for applied fillets, but at
maximum depth. Opposed pinch fingernail impression was measured for length across from
nail-top to nail-top impression at their widest separation, for width at right angles to the
length measurement, and for depth at the deepest point recorded by the depth gauge,
averaging the readings of multiple elements where possible. Paired linear impressions
were measured as for linear impressions. Prism applied fillets were measured as for
applied fillets. Rotated tool excision was measured as for finger impression. Single
fingernail impressions were measured as for linear impressions. The single example of a
pattern made up of repeated applied triangles was measured for length across the widest
241

horizontal dimension, and for width in the vertical dimension, and for depth as for other
applique. Wave elements were measured as for finger impressions, and wavelength was
entered as ERD.
Handedness was recorded in many instances but data will not be presented, other
than in the appended database files on CDROM. These data were interesting though as
these provided many clues to vessel orientation during the application of decoration
(assuming the potter to be upright) during the process of decoration, and also revealed
some variety within the element category of opposed pinch fingernail impression.
Additionally, many of the deeper fingernail marks were of such small size (width) as to
suggest the decorators were women. There is potential for use of variation in the shape of
fingernail impressions as a proxy for genetic variability of the potters.
Mark cross-sections across the width of the mark (length on the case of OPC
fingernail impression) through the depth measurement point were recorded in the “Mark
Section” field as follows: v-section (v), square-section (s), oblique v-section (o), u-shaped
(u), parallel-sided pierced (p), pierced using a conical tool (x), double-v section(w) (as
though mark was incised with a fingernail and fingertip, or with a frayed tool), with “t”
denoting other unclassified cross-sections (specified in “Notes_2").
Transforming Relational Data into a Flat Table:
A flat table with one record (row) per sherd was created from the relational database using
a subset of variables pertaining to sherd size, fabric, decoration and form, to provide
summary information on the structure of vessel form and decoration as represented by the
sample of sherds. (Table 17 shows the data structure and raw data is given in table Flat.db
appended on CD).
Descriptive data for decoration was summarized as a short list of decorative
242

attributes, entered into four location fields (decoration-lip = “Part 1", decoration-rim =
“Part 2", decoration-neck = “Part3", decoration-shoulder = “Part 4"). Carination and body
decoration were so highly correlated with upper vessel decoration that these parts were
omitted from the table to prevent multiplicity of fields. Attribute value codes used in all
four location-specific decoration fields were as follows :
1. Complex linear motif Class 1 with linear design zone horizontal markers, typically
double-line zone markers
2. Complex linear motif Class2 without linear design zone horizontal markers.
3. Multiple bands of opposed pinch fingernail impression (fni)
4. Lateral band of punctation
5. Other unclassified decoration not including wiping, brushing or tool-forming marks
6. Plain
(this is a filter attribute)
7. Horizontal finger deformation/impression, discontinuous
8. Wiping, brushing or tool-forming marks (this is a filter attribute as these records
do not constitute decoration in the sense of deliberate patterned execution)
9. Staggered opposed inner and outer bands of lip impressions forming a wave
pattern without deformation
10. Horizontal finger deformation of the lip into a continuous wave
11. Fingernail impression, single band of opposed pinching
12. Band of impressions along the top of the lip
13. Band of impressions along the inner edge of the lip
14. Opposing bands of impressions along both edges of the lip, not staggered to form
a wave pattern
15. Band of impressions along the outer edge of the lip
16. Lateral band of applied nubbins
243

Multiple decorative attributes on single vessel part were discarded from the attribute data
for the flat table, but this only affected a small number of unusual cases, and does not affect
overall results. Subsequent analyses made use of both this flat data structure and the more
comprehensive attributes/variables in the decoration and form tables, through the use of
multi-table one-to-one and one-to-many database queries.
Parts of the vessel not represented by the sherd were identified in “Flat.db” by a
decorative value of 99, thus a lip sherd would be coded with a value of 99 for each of the
missing rim, neck, and shoulder parts. The remaining data fields are a transposition of basic
variables from the form and thickness tables, separated by vessel part. This draws
information about co-variation by part together into a single table of sherd properties for
analysis of the structure of decoration and form across the vessel.
Chapter Summary and Conclusions:
The preceding sections detail the overall data model of the ceramic database, and the
contents of each of the four tables that make up the primary database. A table of the vessel
family membership of lip and carination sherds, used to estimate a parent breakage
population, is detailed in the next chapter. Attributes values were listed and explained in
a series of tables and paragraphs pertaining to each attribute, and methods for obtaining
values for variables requiring measured dimensions or counts of repeats were explained.
The overall objective in designing this database was to try to capture sufficient descriptive
information to be able to draw a picture of the vessel as represented by the sherd from the
information recorded, using these rules. Fine structured decorative detail was recorded
formulaically alongside larger, more complex or incomplete patterns that required
illustration.
244

Table 17: Data structure for flat table of summary data of sherd properties.
Field Name Type Size Key
Id A (alphanumeric) 7 *
Area N (number)
Weight N
Site S (short integer)
Unit (spatial collection S
Code (fabric A 6
Fabric (temper class) S
Form A 2
Sector (EVE N
TYPE S
Decoration Part1 (Lip) S
Decoration Part2 S
Decoration Part3 S
Decoration Part4 S
Lip Type A 1
Lip Angle N
Rim Type A 1
Rim Angle N
Rim Vertical Curve N
Rim depth N
Neck Type A 1
Neck Angle N
NeckVcurve N
Neck Hcurve N
Shoulder Type A 1
Shoulder Angle N
Shoulder Vcurve N
Shoulder Depth N
Carination Type A 1
Carination Angle N
Carination Vcurve N
CariniationHcurve N
Carination notes A 240
Body Type A 1
Body Vcurve N
Body Hcurve N
Lip Thickness N
Rim thickness N
Neck thickness N
Shoulder thickness N
Body thickness N
245

I would like to think that this approach addressed the problem discussed in previous
chapters of separate analytical/descriptive schemes for Lapita and for post-Lapita
decorative design, which create an artificial disjuncture in many regional series. This
scheme has been an attempt to bridge that disjuncture, by having a hierarchy of descriptive
detail that can capture the broader complexity of larger, more complex (and usually
fragmented) designs as well as simpler formulaic repetition of basic elements. The finer
details of technique can be addressed also, down to the physical characteristics of potter’s
fingernails, and microscopic differences in technique, measurable by gauge rather than by
eye. This holistic approach to ceramic variation allows the analysis of Oceanic pottery to
a level of fine detail where variability in potting behaviour from the past can be read
accurately. The key, or a major aid at least, to recording at this level of detail is a relational
database, where many data themes can be constructed about each part of the sherd, while
maintaining the overall relationship of these parts to each other and to the whole, without
requiring enormous forms, spreadsheets or tables with hundreds of fields of information,
most of them with missing values due to vessel fragmentation into sherds.
246

CHAPTER 5:
SAMPLE EVALUATION AND QUANTIFYING
Introduction:
SAMPLE SIZE.
As outlined in Chapter 1 the development of sherds-as-vessels approaches to ceramic
analysis is recognition of the difficulty of quantifying sherd samples for comparison unless
they have the same levels of brokenness and completeness. Estimation of a breakage
population provides a solution to this problem (Felgate & Bickler n.d). Evaluation of
ceramic sample size and sample representativeness in general continues to be problematic
in Pacific archaeology, where the measured or quantified properties of recovered samples
are often regarded as representative in an unspecified way of the properties of people’s
lives in the past. In this chapter and in subsequent taphonomic and wave-exposure chapters
I try to develop ways to be more specific about the nature of that representation, allowing
a more robust behavioral inference from the recovered sherd samples. Objectives in
estimating a breakage population for the Roviana pottery sample are as follows:
• evaluate sample size and significance
• accumulations research: investigate implications for the intensity/duration of
occupation using the size of the estimated breakage population rather than the size
of the recovered sample
• estimate the extent of post-depositional sherd/pottery attrition between the
breakage population and the extant deposit, as an indication of the type/extent of
sherd-taphonomic process affecting the sites from which the samples were
recovered
247

• characterize the state of preservation of the deposits (the conditions of wave
exposure which are thought to be a key taphonomic factor are explored in Chapter
6). If most of the breakage population has vanished even in the sheltered waters
and relative stability of sea levels of Roviana Lagoon, what are the implications for
the archaeology of Lapita in Near Oceania?
• choose a unit of quantification: what does the state of vessel
brokenness/completeness tell us about how ceramic attributes of these sherd
samples should be quantified? If completeness can be shown to be extremely low,
then sherds are independent observations, and sherd counts are likely to be biased
only by differential survival of vessel types/parts of types, rather than by differences
in brokenness or bias arising from MNI-based counts.
Methods for Establishing Vessel Brokenness and Completeness:
Two characteristics of the archaeological sample form the basis of an estimate of the
breakage population: mean sherd size (measured in EVEs as explained in Chapter 4) as a
measure of brokenness (fragmentation); and vessel completeness (Orton et al. 1993:167-
168 and 178). The key methodological problem for archaeologists, regardless of which
estimation statistic is adopted, is that of identifying vessel classes/species/sherd families
in the potsherd sample. Measurement of these properties of sherd samples has generated
a substantial methodological literature (see for examples Orton 1993, Orton et al.
1993:172, Schiffer 1995c:183-186, Smith 1983:77-79). The task of identifying sherd
families can be complex, and the prospects of a reliable solution being obtained to this
problem vary from sample to sample. This is because pottery is found in an infinite variety
of states of brokenness and is manufactured with varying degree of standardization.
The following assumptions were made about the sample:
• The sample assemblage is a random sample of vessels drawn from a breakage
248

population (this is unlikely to be true, but a good idea of the departure from
randomness is obtained in subsequent chapters, and the consequences for this
analysis are discussed)
• Vessel brokenness and vessel completeness can be accurately measured for the
sample.
The vessel-randomness of a sample and the validity of ones measures of vessel brokenness
and completeness are central to the level of confidence with which the breakage population
estimate can be accepted. It is not a random sample of sherd size that is needed.
Taphonomic processes and collection intensity are likely to have structured sherd size, and
as previously discussed, we do not need to concern ourselves with what size that sherds
we do not have were or are. Rather, we must either assume or have reason to believe that
the sherds in the sample assemblage are a random sample of vessel membership of the
breakage population we wish to estimate.
Collection of sherds from broad areas rather than from testpits suggests that the
measured completeness of sherd families is unlikely to have been biased by sampling a non-
representative locality in the sites. While test excavation indicated that some pottery
remained buried in some areas of some sites, there was no suggestion of a buried well-
preserved deposit in any of these (Reeve 1989 reports the Paniavile site as such a deposit
but this was not the case by 1996). Turbation by waves and bio turbation are thought to
have created a complex history of vertical migration of sherds within sediments and some
horizontal movement across the sites, although the lack of rolling damage on larger
recovered sherds from the deeper margins of sites suggests such horizontal movement was
infrequent.
Lip/rim-based sherd families mostly comprised singleton lip/rim sherds, so spatial
dispersion data could not be obtained from the distribution of sherd families.
249

No significant departures from latitudinal circularity were observed among vessel
rims, beyond those attributed to inaccuracy of manufacture. Brokenness was measured by
estimating the percentage of the circular vessel lip/rim represented by sherds. Lip/rim EVE
was estimated for sherds by curve fitting, using a rim diameter percentage chart marked
in 5% sector intervals, with concentric circles ranging from 20mm radius to 300mm radius,
at intervals of 20mm. Where the sherd was too small to estimate lip horizontal radius with
confidence, EVE was estimated using a default radius of 140mm, approximating the
average horizontal lip radius recorded for the larger sherds, about which there was little
variation. Sherd sizes ranged up to a maximum lip/rim EVE of 25% of the vessel
circumference, a recorded minimum of 1%, and a recorded average lip and rim EVE for
all sites of 5.78% (n=471). Mean EVE by site is shown in Table 18.
Grouping lip/rim sherds into sherd families was the most subjective step in
constructing an estimation of the number of vessels in a breakage population. Sherds
Table 18: Variation in lip brokenness between sites.
Site Lip sherd count Average EVE (%)
Paniavile 76 4.56
Hoghoi 82 5.16
Miho 60 4.41
Honiavasa 104 7.14
Gharanga 38 5.78
Nusa Roviana 11 7.09
Kopo 3 8.66
Zangana 76 5.93
sorted by site, then sorted further by fabric and by form variation, formed the basis of the
primary vessel groupings, which were further subdivided by decoration information. This
required some idea of the extent to which fabric and form might vary within a single
250

vessel, and the best information pertaining to these problems was obtained from larger
sherds. Fabric and form attributes were examined at multiple points on such sherds to
develop an assessment of the amount of within-vessel variation.
No instances were found in which temper of different petrographic classes (see
Chapter 4) were incorporated into a single vessel. Grid counts of temper density and
mineralogy, under low-powered reflected-light magnification, at multiple points on single
sherds, indicated that there was only slight variation in temper mineralogy (in the relative
proportions of mineral grains for example) within vessels, but that temper density could
vary more dramatically within vessels. In sorting lip sherds by fabric, coarse variations in
the relative proportions of minerals were used. Colour variation was disregarded, as
surface colour appeared to have been affected by the aerobic regime of specific
depositional context, and colour zonation in the break could be expected to vary within a
single vessel as a result of firing variability.
Form variation of the lip and rim within vessels was examined qualitatively by
looking at larger sherds where a greater proportion of the vessel was represented (up to
about 30%). Slight variation in form around the lip/rim circumference were noted, and
small variations of this order of magnitude were ignored when sorting fabric groups into
vessel groups based on form.
Sherds for which the edges of the lip were weathered, making decorative analysis
impossible, were generally small and water-rolled, and had been collected along the
strandline. These were omitted from the analytical sub-sample used to estimate
completeness.
It was noted that few if any of the more decorated lip/rim sherds could be matched
to any other, suggesting either that decoration changed around the circumference of the
vessel lip/rim, or that vessel completeness for all site assemblages was very low. Some
sherds showed latitudinal decorative discontinuities of this type, but these discontinuities
were restricted to a small range of incised motifs, omitted from subsequent analyses.
251

This relationship between the amount of decorative information on lip and rim
sherds pertaining to vessel membership, and the sherd count for vessel groups, was
systematized by counting the number of latitudinal bands of repeated decorative elements
that made up the total decoration of the lip and rim. These were typically bands of repeated
parallel edge-impression along the edges/top of the lip, or bands of punctation or incised
decoration around the rim. The properties of the tool used to make the decorative marks
and the average spacing and depth of marks were also used to differentiate between vessels
in some cases. Here again, this depended on assessing through examination of larger
sherds, the amount of variation one might expect in this regard within vessels. For all sites,
this system yielded the data in Table 19 (vessel membership data is appended on CD in
“Vessels.db”).
Table 19: Selection of sample for estimating vessel completeness.
Sherd type/
# bands
decoration
Sherd
count
Vessel count Min. # per
vessel
252
Max. # per
vessel
Mean # pieces
per vessel
(completeness
lip/rim (0 bands) 91 53 1 14 1.72
lip/rim (1 band) 264 166 1 15 1.58
lip/rim (2 bands) 68 65 1 2 1.05
lip/rim (3 or more
bands)
lip/rim (2 or more
bands)
28 25 1 2 1.12
96 89-90 1 2 1.07-1.08
Undecorated lips or lips with a single band of decoration (usually parallel edge
impressions) yielded up to 15 pieces per vessel. The sample of more decorated sherds by
contrast yielded lower vessel completeness, with a maximum of two lip/rim sherds per
vessel, and an average of 1.07-1.08 sherds per vessel, despite similar brokenness. This
lower level of completeness is considered to be a more accurate reflection of the state of
the assemblages than the figures for the plain or slightly decorated lips. The large vessel
groups for the plain or slightly decorated lips are inferred to be faulty attributions of

sherds to vessel groups as a result of inadequate information with which to construct
groups. This inference depends on the assumption that stylistic variation between sherds
results from idiosyncratic recombination of the potters' repertoire of continuous horizontal
bands of decoration for each pot constructed, and also on the related assumption that
differences between sherds do not result from changes in decoration around the
circumference of the vessel.
Simulation Approach:
The number of pieces into which a population of vessels breaks is variable. The
probabilities of individual vessels being represented in an archaeological sample are thus
unequal. For pottery, the “death” of a pot (breakage) generally implies at least two pieces,
although there are exceptions. Moreover, breakage is not necessarily a single event. As
Orton states,
“It is obvious that sherds can either stay the same size or become smaller
(through breakage) over time (Orton et al. 1993:176).”
In the latter case, these would not only become more numerous (increasing probability of
survival/capture in the archaeologist’s sample), but once below the detection size
threshold, would become less likely to be found (decreasing probability of capture in the
archaeologist’s sample). This last factor constrains the minimum size of sherds commonly
recovered by archaeologists. An additional size-related factor in survival/detection is that
smaller fragments are potentially more mobile, either within sediments as a result of
turbation or on the sediment surface due to fluvial or aeolian transport.
While it is possible to envisage either progressive breakage (see Orton 1982 for
a simulation approach to progressive breakage) or a state of breakage stasis, it is not
possible to know from the sample assemblage which of these situations applied to a
pottery breakage population; these are equifinal in that either progressive breakage or
253

static breakage could produce any found broken assemblage. The state of brokenness in
which a pottery breakage population existed is usually unknowable from the material
record because the creation of a pottery breakage population is usually a diachronic
process, and archaeologists only have access to the deposited/fossil assemblage through
their sample assemblage as a synchronic (or nearly so) observation. A vast array of small
breakage and re-breakage events may have occurred (or not) over the period prior to
sample recovery, except in exceptional circumstances, such as where people or natural
formation processes have curated or preserved pots in their death-state upon breakage,
either acting continuously in this way over time, or where by some instantaneous process
a whole assemblage of vessels is broken at once and lain undisturbed since (as at Pompeii).
In such circumstances, preservation would be so good that there would be no need for
estimation approaches.
It is the property of potsherd samples that whole objects are broken into a number
of fragments that allows us to establish vessel representation fraction (the fraction of the
breakage population represented in the sample). If an assemblage of unbroken pots were
recovered we would have no information on whether others existed at that location in the
past and were carried off in the intervening years. The more broken an assemblage, the
more the survival chances of each vessel as a fragment improve (to a degree) as do the
chances that some fragment of any individual vessel will make it into the archaeologist's
sample.
To reiterate, basic assumptions are:
1: The sample assemblage is a random sample of vessel membership of the
breakage population,
2: vessel brokenness and completeness can be accurately measured from the sherd
sample.
Sample brokenness was defined as the mean number of rim sherds per vessel, and was
254

calculated by dividing 100% by the mean EVE of sherds in the sample assemblage. As
EVE is an estimate, assemblage brokenness is also an estimate, although no attempt is
made here to calculate precision of that estimate. Vessel completeness was calculated as
the estimated mean number of pieces per vessel in the archaeological sample, or the
calculated mean number of pieces per vessel in any simulated pottery sample
The simulation sherd sampling fraction was defined as the number of sherds one
would have to remove from a simulated breakage population of complete vessels
fragmented as observed in the sample assemblage, to achieve the level of completeness
observed. This definition implies that brokenness of the assemblage was static in the past
rather than progressive, and means that rather than asking the question 'what fraction of
an actual population of sherds near the time of breakage of the pottery is present in the
sample?', the question posed in our approach is rather 'If the pottery in this deposit had
existed since initial breakage in the state of brokenness observed in the sample, what
fraction of the initial sherd population would the sample represent?'
The simulation vessel representation fraction is the fraction of the simulated
breakage vessel population represented at any given simulated sherd sampling fraction.
While sherd sampling fraction is a stipulation device permitting the calculation of vessel
representation in samples, based on the expected number of similar-sized pieces in the
whole breakage population, vessel representation fraction seems to be a valid measure of
the relationship between sample and parent population, and informs as to the fraction of
the breakage population represented by an archaeological vessel/sherd sample. The sherd
sampling fraction range is the range of outcomes in the simulation for which vessel
completeness is consistent with the archaeological sample. The vessel representation
fraction range is the range of simulation vessel representation fractions that are consistent
with the vessel completeness of the archaeological sample.
In the simulation, brokenness is treated as static, and is determined by specifying
255

the number of vessels and the number of sherds in a simulated breakage population. Vessel
completeness in these simulated breakage populations will of course always be 100%.
Brokenness is made consistent with that observed in the archaeological sample. The goal
of the simulation is to establish, for Roviana sample assemblage brokenness, the shapes of
the curves and the range of likely values that relate:
a) simulation sherd sampling fraction to vessel representation fraction, and
b) simulation vessel completeness to simulation sherd sampling fraction.
This is achieved by simulating a large number of sherd sample assemblages with a level of
brokenness consistent with the archaeological sample. Vessel membership of sherds in all
samples is assigned randomly, but overall, is consistent with the expected number of pieces
per vessel at that sampling fraction. Completeness (mean number of sherds per vessel) and
vessel representation fraction are calculated for each sample and are displayed in a plot to
show the relationship between vessel completeness and vessel representation fraction. By
this means a good estimate can be gained of the sherd and vessel sampling fractions in the
simulation consistent with the vessel completeness and brokenness estimated for the
archaeological sample.
To reiterate the statement made earlier, this is not a suggestion that the breakage
population had the particular static level of brokenness of the archaeological sample, and
of all the simulated samples. The question of how many bits someone's pottery broke into
when they dropped it is probably of little historical or archaeological interest, and is
usually unknowable. Stipulating a level of brokenness consistent with the archaeological
sample assemblages is rather a way of calculating, which allows the question 'how much
needs to be taken away from the whole to give it the characteristics of the sample? We
also do not really have to be concerned with what size any extant missing sherds are or
were in the total discard assemblage, as ultimately it is vessel representation fraction we
are after as an entry to quantification of the breakage population, rather than sherd
256

sampling fraction itself.
Estimating the number of vessels in a breakage population requires approaches that
allow a variable number of members in classes (for example Chao 1984, Chao & Lee
1992). Accordingly, the simulated samples are created as datasets containing a variable
number of sherds per vessel. In the simulated breakage population, sherds are randomly
assigned to vessel membership based on a probability derived from brokenness as measured
by the archaeological sample mean EVE.
Statistical Approaches:
While the simulation approach is useful for elucidating the relationships between
brokenness, completeness, and vessel representation in the sample, a number of statistical
algorithms apply to the problem of estimating a breakage population from a sherd sample
(estimating the number of classes/species in a population). These have the advantage of
producing formal confidence limits. Of particular relevance to the problem are the capture-
recapture statistics used in ecological research where a sample of animals are captured,
tagged and then released back into the population. Once they have redistributed across the
landscape, animals are captured and then released again in a series of trials. The frequency
of capture of particular tagged animals is used to generate an estimate of the total
population. While there is no attempt here to summarise the vast literature on this topic
and the complexities involved, the following points are particularly relevant to the
archaeological problem.
The capture-recapture process works because the probability of capturing a tagged
animal from a population tells us something about the population size. Getting one tagged
animal from a single sample does not really tell you anything about an unknown
population size. However, the more tagged animals you get and the more times the same
257

tagged animal is recaptured in successive trials, the more certain you can be about the size
of population that it probably derives from. These statistics have been reworked to
estimate the number of species in a population, in which form these are directly applicable
to the problem of estimating the number of pottery vessels represented by a sherd sample.
Here the number of ecological capture trials is analogous to the maximum number of
pieces from a vessel in the archaeologist's sample, with incidence of singleton sherds
comprising the sample of animal captured in the first trial, the incidence of doubletons
(pots represented by two sherds in the recovered sample) being the second capture sample,
the incidence of tripletons (pots represented by three sherds) the third, and so on.
This approach does not make explicit use of the sherd-size information inherent in
pottery samples, which informs on the probability of capture (brokenness), which seems
wasteful of information. Such information on the probability of capture could potentially
be used to narrow the confidence limits obtainable for a given sample size, or to reduce the
sample size required to obtain meaningful confidence limits.
The Chao (1984) estimate was tested experimentally (Felgate & Bickler n.d) using
data simulated from hypothetical breakage populations. Capture-recapture data calculated
from these simulated samples was used to estimate (with confidence limits) the known
breakage population, using the program “EstimateS”(Colwell 1997). The estimates from
the Chao 1984 statistic were compared with the known breakage population.
The key findings were that:
• Sherd sample fractions of 5% or more allow a close estimate of the breakage
population (this figure is a product of the size of the breakage population used in
the trial, and could be improved using a larger population)
• Even at 2% sampling fraction, the result is useful if the breakage population is
large
• Higher mean EVE (less broken assemblages) make the statistic less accurate,
258

equiring a larger sampling fraction (but at the opposite extreme, in very broken
assemblages, although the statistic works well, the data will be impossible to
acquire; if the pottery is too broken the measurement of EVE itself is
compromised).
In contrast to the simulation method, the Chao statistic does not require discrete
information on assemblage brokenness. The structure of the capture-recapture data
provides this information. While the Chao statistic appeared to give good results only if
the sherd sampling fraction is greater than 5% of the breakage population, this was a
sample size effect. If the simulated breakage populations used by Felgate and Bickler had
been larger than 10,000 sherds, confidence limits ought to have narrowed at lower
sampling fractions.
Simulation Results:
Using the two-or-more-bands-of-decoration approach outlined above, completeness was
estimated to be between 1.07 and 1.08 mean sherds per vessel. Brokenness was calculated
by taking the average EVE of all lip sherds in the site regardless of the amount of
decoration present (5.78%). Site samples were combined to boost sample size. The
question being asked of the following simulation is therefore “What is the total breakage
population for these six sites?”
An initial simulated breakage assemblage of 10 000 sherds forming 578 vessels was
stipulated, consistent with overall sample Brokenness (5.78% EVE). Initial simulation with
500 sampling fractions from 0% to 100% indicated that vessel completeness of 1.08 sherds
per vessel corresponds with simulation sherd sampling fraction of somewhere less than 3%
(Figure 73).
259

A second, detailed simulation was run using an enlarged breakage population of
Figure 73 Roviana vessel completeness against sampling fraction, sampling
fraction against vessel representation fraction; random assignment of shreds to
vessels; mean EVE of 5.78%.
100,000 sherds (5780 vessels) at 11 sampling fractions between 0% and 2%, iterated 100
times, plotting 1100 experiments in total. Vessel completeness of 1.07-1.08 sherds per
vessel correspond on the simulation plot with vessel representation fraction of between
8.9% and 17.5% (Figure 74).
260

The total archaeological rim/lip sample was 471 sherds. Based on vessel completeness
estimated to be from 1.07 to 1.08 sherds per vessel, this sample was inferred to physically
represent between 436 and 440 vessels. This suggests limits for a total breakage population
for the analyzed sites of 3259-4944 at 1.07 completeness, or at 1.08 completeness, 2491-
4152 vessels. The combined approximate limits of the breakage population for this run of
the simulation program would be 2491-4944 vessels.
Repeating the simulation two more times yielded vessel representation fractions
(using completeness of 1.07-1.08) of 9.6%-19%, and 8.8%-18.9%. Combining the three
ranges obtained yields vessel representation fraction range of 8.8%-19%, suggesting a
more conservative approximation of breakage population confidence limits at 2sd would
be 2295-5000 vessels.
For comparison with the Chao calculation given below, a single completeness
Figure 74: Roviana vessel completeness against vessel representation fraction;
random assignment of sherds to vessels; mean EVE of 5.78%.
261

value of 1.08 was used, yielding a total vessel count of 436 in the sample of 471 rim
sherds. Vessel representation fraction averaged limits were 10.5%-17.5%, 11.2%-19%,
and 10.6%-18.9%. Combined vessel representation fraction limits at this level of
completeness were 10.5% to 19%. The total sample assemblage of 436 vessels (at 1.08
completeness) represents a breakage population of 2295-4152 vessels.
Statistical Results:
The results of analysis of the same Roviana data using the Chao (1984) statistic (Table 20)
are similar. The results do not coincide, but the ranges overlap, which is encouraging,
especially as the simulation results indicate that the Chao statistic is operating at its limits
of usefulness at such small sampling fractions. The 89 vessels in the sample yielded parent
population limits at 1s.d. of 293.38-716.16 vessels, or vessel representation fraction limits
of 0.3034-0.1243. (At 2s.d., i.e. ±422, the breakage population represented by the 89
vessels in the sample would be 83-927, illustrating how at these very small sampling
fractions and limited sample sizes the Chao statistic has insufficient information on the
probability of “capture” in the sample to give meaningful confidence limits.) The 436
vessels in the total rim/lip sherd sample assemblage, at 1s.d., represent a breakage
population of 1437-3508 vessels (compare this to a breakage population of 2295-4152
vessels using the same data in the simulation).
Table 20: Breakage population estimate for the combined Roviana highly decorated lip
sample using the statistic of Chao 1984.
Vessel count Sherd count Singletons Doubletons Breakage population 1s.d.
89 96 82 7 504.77 211.39
262

Discussion:
In cases where the Chao statistic yields estimates with reasonably tight confidence limits,
it is clearly a more systematic solution to the estimation problem than our simulation
approach as it permits an estimate with confidence limits without the archaeologists
specifically inputting brokenness information, which may itself be subject to error. In the
case of the Roviana data the sample size is too small for the structure of the multiple
capture data to inform as to the probability of capture, leading to an overly large standard
deviation (at two standard deviations the 89 vessels in the sample would represent between
83 and 927 vessels, which is not a very convincing result). The Chao estimate does rely on
the structure of the capture-recapture data to establish the brokenness of the sample. This
structure breaks down at small sampling fractions and small sample sizes. In these
circumstances the simulation approach provides a better estimate than the Chao statistic,
as brokenness is independently established from the EVE characteristics of the sample of
lip/rim sherds.
Summary and Conclusions:
Two methods of estimating a breakage population were applied, a simulation approach,
and a statistical approach using the algorithm developed by Chao (1984). Both require that
sherd families be identified in the sample. The simulation approach requires a measure of
completeness (in mean number of pieces per vessel) and brokenness (expected number
of pieces per vessel) from the archaeological sample. The simulation approach proceeds
by simulating many samples of a breakage population with a degree of brokenness
consistent with the archaeological sample, to develop plots of the relationship between
vessel completeness and vessel representation fraction. The measured vessel completeness
263

from the archaeological sample is then compared with these plots to discover the
corresponding vessel representation fraction indicated by the simulation. Dividing the
number of vessels represented in the archaeological sample by this vessel representation
fraction range yields the likely limits of vessels in the breakage population.
The statistical approach uses a version of the capture-recapture approach for
estimation of animal populations, adapted to the problem of estimating the number of
species in a population. In this application to archaeological pottery, the number of vessel
sherd families in a breakage population is equivalent to the number of species in a
population, and the presence of multiple sherds from the same vessel in the archaeological
sample is equivalent to multiple capture events. This approach was tested on a range of
simulated breakage populations and simulated archaeological samples, and for the sample
sizes tested, found to be accurate for sherd sampling fractions of 5% or more. Moreover,
the statistic was more accurate when the vessel assemblage was more broken because of
the enlarged sherd sample for each vessel, which increases the probability of multiple
“captures” at any given sherd sampling fraction.
Drawing a theoretical distinction between a breakage population and a discard
assemblage means that there is no required assumption that all of the sherds of pottery
were discarded at the site. Sherds discarded or destroyed off site are simply sherds that did
not make it into the sample. Only when seeking a history of the pottery outside the sample
assemblage does such transport or destruction become significant.
Both estimation approaches require close attention to the nature of the total
sampling processes. A vessel-random selection of sherds from the breakage population is
the ideal. In archaeological terms this is most likely the case when sherds from the same
vessel show a high degree of dispersion across the site. This suggests the approach is most
suited to analysis of surface collections, or collections from large areal excavations, where
dispersion information is attainable. Alternatively, other sampling approaches to
264

archaeological sites such as multiple test excavations across a site might yield adequate
information on the dispersion of sherd families.
For both approaches, the accuracy of estimates of breakage population obtained
depends on the accuracy of the estimate of vessel completeness. For the simulation
approach an estimate of sample brokenness is also required. The Roviana data illustrate
some of the difficulties in attaining reliable data of this sort. This requirement places a
heavy burden on the analyst to develop reliable ways of sorting a sherd assemblage into
vessel lots. This is not a simple analytical task and there is vast scope for methodological
development in this area.
For the overall Roviana sample, where vessel completeness is uniformly low, we
can infer from the simulation that around 99% of the sherd material of the breakage
population has not made it into the sample, and secondly, can appreciate that all is not lost,
and that means exist for estimating the size of the total breakage population represented
by the sample collected.
Sample sizes were too small to estimate the composition of the breakage
population for individual sites, and an argument is made elsewhere that some production
styles could be better estimated from other vessel parts such as necks or carinations, due
to fragile rim form (elaborated in Chapters 8 and 9).
The extent of sherd attrition shows that assemblage composition is potentially
biased and is most likely comparable only with other assemblages which have been subject
to a similar taphonomic/sampling regime. It is clear that the vast bulk of sherdage deriving
from the breakage population has not made it into the archaeologists sample. The count
(assuming breakage stasis) or weight of sherds recovered would underestimate the
intensity/duration of occupation by a factor of 100. The estimated number of vessels
physically represented in the archaeological sample under-represents the size of the
breakage population by a factor of between five and twelve.
265

While some pottery was found buried in sediments in some sites, through test
excavation, this could account for only a tiny fraction of the missing pottery, and as
sample collection of the reef-flat surface scatter was intensive, it is inferred that the bulk
of the breakage assemblage pottery had been removed by taphonomic processes.
Observation of exotic pyroxene minerals in the local calcite sand of one site (Hoghoi)
suggests much of the breakage population has dis-aggregated entirely into constituent
mineral grains.
The potential for bias in the composition of recovered sample assemblages is
therefore inferred to be high for the sites as a group. The conclusion for any analyses that
use relative abundance of styles or attributes (frequency seriation for example) must be
that external comparisons should be restricted to sites that have undergone similar
taphonomic processes. Direct comparisons should be avoided even with sites that seem
to have similar cultural formation processes, but different taphonomic regimes (for
example the Mussau sites).
For accumulations research, the estimate will under represent some fragile lip
forms, and also cannot represent types or sites that have not made it into the sample at all.
It is entirely possible, indeed likely, that many more than 4000 vessels were broken around
the intertidal zone of Roviana lagoon in the past, but the result obtained is sufficient to
confirm the intuitive understanding that most of the sherdage must have gone, in order to
produce the observed pattern, and that at least five times as many vessels as the number
observed in the sample were broken here.
How fragile is this site type? The degree of wave exposure for sites is compared
in Chapter 6 but all sites are broadly similar due to a broadly comparable situation within
the Lagoon. It is clear from the simulation that around 99% of the physical quantity of
ceramics has not been captured in the sample, and as sampling was quite intensive, aiming
at total collection of lip sherds, it is clear that site preservation even in the sheltered waters
266

of the Lagoon is marginal, and that this type of site is almost entirely removed by
taphonomic processes.
This result suggests that aceramic lithic scatters in similar locations may have had
a ceramic component in the past, removed by differential weathering of ceramics and
lithics. Had similar sites been located in more wave-exposed settings elsewhere in Near
Oceania, we can be certain that few of these would have survived to be detectable as
ceramic scatters by an archaeologist today. The poor state of preservation of the Roviana
intertidal-zone sites suggest settlement patterns, site density etc for this site type cannot
be read directly from survey results, and that preservation and visibility factors ( probability
of detection) must be carefully considered to read survey results.
Finally, the low state of completeness of the Roviana vessels indicates that sherd
counts of decorative attributes are likely to be independent in the vast majority of cases,
and that sherd count is the best unit for counting decorative attributes (quantifying sample
size) for seriation analysis.
267

268

Introduction:
CHAPTER 6:
WAVE EXPOSURE
Systematic consideration of the effects of wave exposure on archaeological deposits are
rare or superficial in recent Oceanic archaeology, despite the situation of the subject in the
largest wave basin on the planet. In the 1920s the prevailing paradigm in Oceanian
archaeology was that there was no Oceanic subsurface archaeology due to the erosional
effects of cyclones, floods, high temperatures and damp conditions (an example of this
paradigm can be found in Linton’s introduction to his “Archaeology of the Marquesas”
published in 1925 (Linton 1925:3-4), but from the 1930s through to the present the
pendulum of scientific opinion has swung to the opposite extreme, with the archaeological
record read relatively directly, and minimal attention has been paid to formation processes
in general and the degree to which wave exposure structures Oceanian archaeological
distributions. A sample surveying approach by contrast demands attention to probability
of site detection, including preservation and visibility, as lenses through which the
archaeological record is read more accurately.
In Chapters 7 and 11 the ceramic scatters are viewed as lag deposits, evidenced by
a high degree of size sorting. The value in constructing this model of wave exposure in
the current chapter lies not so much in being able to test it against sherd size data, but in
having some means other than the sizes of recovered sherds to assess relative exposure.
Sherd size is subject to a collection intensity effect, making highly size-sorted
assemblages difficult to distinguish from poorly sorted assemblages unless scrupulous
control for collection intensity can be maintained (difficult unless one knows the entire
269

history of collection of the site). Modelling wave-exposure allows a cross-check on size
data, and promotes the view that factors additional to natural transforms are structuring
the taphonomic data in assemblages (in the next chapter some conflicts between expected
size distribution from wave exposure and actual size distribution of the archaeological
sample will be highlighted which illustrate this).
Review:
Some studies in which wave exposure receives particular mention are reviewed below
(passing reference is made in many studies to wave-laid sediments in excavations, but these
are reviewed elsewhere (Tarlton 1996).
Green’s survey of the Surville peninsula on Makira (San Cristobal) in 1970 noted
the dramatic effects of 1971 cyclone waves on coastal rockshelters. For Ulawa in the
southeast Solomon Islands (Hendren 1976, Ward 1976) Ward notes evidence for exposure
of this coast to cyclone waves, and effects on terrestrial site preservation. The occupation
sites excavated by Ward postdated about 700AD. Results suggest terrestrial pottery sites
of any period are rare or absent from the extant archaeological record.
Lilley in a study of islands in the Vitiaz Strait noted the likely impact of the 1888
Ritter tsunami on the north coast of Umboi, parts of coastal Sakar, western New Britain
and the low-lying Siassi Islands, and coasts of Malai and Tuam islands (Lilley 1986:20),
and suggested that low levels of site visibility and survival could be expected as a result of
similar events and also as a result of storm action (Lilley 1986:106). Lilley did not make
any sampling-based inferences regarding archaeological distribution in the past, or use this
type of information in inferring site formation processes, despite his Lapita-phase evidence
(pottery from the KLK site on Tuam) being,
“...associated with a very restricted quantity and range of other cultural
material in a clean beach sand matrix which is devoid of structural features
270

such as hearths and postholes (Lilley 1986:455).”
He inferred, like Wickler for Buka, that the Lapita-phase pottery represented, “...probably
only intermittent activity in Siassi generally and on Tuam in particular and second, that
people did not actually live on the KLK site”, but did not regard the lack of occupation
features as related to natural formation processes.
The effect of storm surges and cyclone waves has been studied in some detail for
Tongatapu archaeological sites (Spennemann 1987). Spennemann concluded that waves,
including storm waves, could play an important part in site formation, site visibility, and
site destruction, but gave examples also of beach scatters that were the transformed (lag
deposit) remains of terrestrial archaeological sites. In his Tongan case study major
erosional effects were limited to exposed sand cays lying of the northeast coast of
Tongatapu. He did not seek to develop any predictive use of the information, arguing
instead for continual monitoring and rescue excavation of sites subject to coastal erosion.
A general study on the subject of tropical cyclones and their effect on Pacific
archaeological sites (Tarlton 1996) considered cyclone frequency for various island groups
in the southwest Pacific and their impact on various island types. The study focused
exclusively on terrestrial archaeological sites, while in the Roviana case the evidence
suggests maritime site location in the past. Thus her discussion of increases in water level
as having largely negative impacts on site preservation is largely inapplicable to stilt
occupation over water in sheltered locations, where increases in water level during storms
should theoretically protect archaeological deposits from swash processes.
Tarlton also concluded that high islands with indented coastlines were likely to
have sheltered areas relatively unaffected by cyclone waves (Tarlton 1996:108, 114), and
that landforms resulting from cyclone activity or tsunami were readily identifiable in many
instances (Tarlton 1996:118), both conclusions pertinent to discussion of Roviana
intertidal formation processes. Tarlton considered that awareness of cyclone or other
271

wave-generated landforms would enable an archaeologist to choose locations which were
relatively undisturbed where stratified archaeological deposits might be discovered (Tarlton
1996:123).
Tarlton reviewed archaeological reports of cyclone or other wave damage to sites
or landscapes. Among these the most sophisticated distributional inferences seem to be
from Australia, for the Queensland coast (Rowland 1989), and the Torres Strait islands,
where whole coastal landscapes have a lack of archaeological evidence thought to relate
to wave damage of sites or wave erosion of coastlines (Vanderwal 1973:187). A similar
explanation for the distribution of Lapita and other early sites in New Caledonia was cited
as a personal communication from Jean-Christophe Galipaud in 1995 (Tarlton 1996:136-
138).
Tarlton provided an extensive section on Kirk’s numerical modeling of cyclone
wave exposure for the Rarotongan coastline and suggested that cyclone wave impact on
archaeological sites could be modeled in a similar way using GIS, perhaps in combination
with remote-sensing identification of cyclone wave-generated landforms (Tarlton
1996:189-196), but did not extend her discussion to the interpretation of regional
archaeological distributions, seeing this approach as principally as an aid to site
prospecting, for locating undisturbed archaeological sites. It is in this section though, that
discussion of a predictive modeling approach for a specific coastline is closest to the type
of analysis conducted in the present chapter. Her emphasis on storm surge models and
wind models was appropriate for exposed coastlines and/or terrestrial sites, but these are
inapplicable in the Roviana case, where collection sites comprise reworked intertidal and
sub-tidal evidence rather than in-situ terrestrial sites.
Roviana Intertidal Ceramics and Waves:
272

Intertidal sites in the Roviana study were all thought to have had broadly comparable post-
deposition natural formation processes, due to location in an almost landlocked lagoon
setting bordered by either mainland high-island shorelines or upraised Plio-Pleistocene
reefs. Discrimination between some of the models of formation processes constructed in
Chapter 7 require some independent assessment of wave exposure variation between sites.
A model of wave exposure is presented in this chapter which deals with seasonal weather
patterns which can be expected to have prevailed in varying degrees in the past, and a
model of wave formation based primarily on the variable “fetch”, which is the distance in
a windward direction between a shore and the nearest land.
Waves coming onto a beach increase in height and steepness until they break. As
waves move into shallow water, circular orbits of the water particles become flattened, and
some wave energy will be used in moving sediments to and fro on the sea bed. This is
particularly so for shallow-sloped sea beds, such as the reef flats on which some of the
Roviana ceramic sites are located. As a wave breaks, most of the energy is dissipated in
the final mixing of water, sand and shingle (Open University Oceanography Team 1989:27-
29). It is these two “swash zone” effects on sediments which are of interest to the shallow-
water/ intertidal archaeologist. The generation of waves by seismic processes and by wind
will be considered separately below.
Tsunami:
Exposure of shorelines within Roviana Lagoon to large waves generated by seismic events
is comparatively low. The Blanche Channel, to which the Roviana reef passages open, is
sheltered by Rendova/Tetepare islands, from tsunami originating in the Solomon Sea
(Figure 6). Earthquake epicentres recorded between 1962 and 1967 show only a single
shallow epicentre within the Blanche Channel, the vast majority of seismic events having
their epicentres to the north, in the Bougainville area, or well south in the Southeast
273

Solomons (Dunkley 1986:9, Mann et al. 1998). In the event that seismic activity did create
waves in the Blanch channel, the effect within the Lagoon would be negligible due to the
narrowness of the reef passages through the uplifted coral barrier islands. It is possible that
refracted waves generated within the Blanche channel by seismic activity could have some
impact on sites like Honiavasa and Miho, located at the inner end of a reef passage,
adjacent to deeper water, but only a few small events of this type are known of , in contrast
to their common occurrence on the Solomon Sea coasts of Rendova and elsewhere in the
New Georgia group.
The lagoon basin is interrupted with many linear and patch-like shoals, thought to
indicate a submerged karst-like landscape (Stoddart 1969a:394), and generation of
substantial waves within the lagoon by seismic activity seems unlikely in view of the
baffling effect of this dissected bathymetry.
Wind-generated Waves:
Waves in this context are defined as the process by which wind kinetic energy is
transferred to and transported across water, as a progressive wave. For an idealised wave,
wave height refers to the vertical distance between trough and peak, which is twice the
wave amplitude, while wavelength is the distance between two successive waves.
Steepness is wave height divided by wave length. Period is the time interval between two
successive analogous points on the wave pattern passing a fixed location. The number of
peaks or troughs passing a fixed point per second is the frequency. Wave speed is related
to wavelength (Open University Oceanography Team 1989:7, 10).
On the open ocean, after a period of calm weather, when the wind starts to blow,
small, steep waves form as the wind increases, continuing to grow after the wind has
reached maximum speed until they reach a wavelength equal to one third of windspeed.
Fetch is the unobstructed distance of sea in which the wave can build. Wave size depends
274

on the fetch (Open University Oceanography Team 1989:11-12). Fetch imposes the prime
limits on wave size, speed and energy in sheltered waters like Roviana lagoon and the
Blanche Channel, and provides an easily measured proxy for wave exposure when
combined with wind direction.
“The factors that are important in increasing the amount of energy that
waves obtain are (1) windspeed, (2), the time during which the wind blows
in one direction, and (3) the fetch.....When both maximum fetch and
duration are reached for a given wind velocity, the sea is said to be fully
developed (Thurman 1981:219).”
This is why windspeed and duration are less significant than fetch for the Roviana intertidal
sites, as the limits imposed by short fetch are quickly reached, beyond which waves do not
build. The following three anecdotes illustrate the primacy of fetch in determining wave
exposure.
During the 1998 cyclone (which tracked through the Rennell/Bellona area)
northerly winds were estimated by the author to have exceeded fifty knots. In these wind
conditions, children were sailing their canoes at respectable speed in the bay near the Miho
site, powered by coconut fronds, across water barely more than rippled, due to the limits
on wave energy imposed by short fetch across that bay. For a sea to become fully
developed in similar windspeed to this would require a fetch of about 2500km and a
duration of 65 hours (Thurman 1981:219).
By way of contrast, moderate tradewind conditions in the Blanche Channel of
around 25 knots windspeed, of extended duration, where the fetch to the southeast is
about 65km, allow a sea to become partially developed, with short and relatively steep
waves hazardous to small craft a fetch of about 180km is needed for a sea to become fully
developed for this windspeed).
Wave refraction is well understood in oceanography, allowing oceanographers to
determine regions that are likely to experience high waves due to this effect. Refraction
275

can change wave height and direction, as can shallowing water (Open University
Oceanography Team 1989:25). Locations such as the inner margins of reef passages may
be exposed to refraction waves even though the measured direct fetch is low.
Seagrass, Mangroves and Reefs:
Seagrasses, mangrove forests and reef shallows all serve to mitigate wave exposure, by
effectively reducing fetch at low tide, acting as an energy absorbing barrier to wave
formation and transmission.
Roviana Lagoon nowadays has extensive shallow mixed seagrass meadows,
characterized most visibly by the long strap-like Enhalus acoroides, in which Dugong
dugon and Chelonia mydas graze, and which shelter a profusion of fish, mollusc and
crustacean species. Besides being attractive locations for finding a meal (easily traversed
by paddle canoe, and less easily if reliant on a rotating propellor), these meadows also have
a sheltering effect on marine archaeological deposits, trapping organic and reefal detritus,
and acting as sediment stabilizers (Stewart 1999:581). Seagrasses have the potential to
create stratigraphy by sealing archeological deposits, and also are potentially agents of
bioturbation through the growth of root systems, as are many of the creatures that live
within the shelter of seagrasses, some of which are burrowers and some of which pluck
grasses from sediments when grazing.
Seagrass shallows are found principally in the eastern two-thirds of the Lagoon,
east of Honiavasa, where the lattice of shallow reefs nears the surface of the water, as a
result of a slight tectonic tilting (Mann et al. 1998). West of Miho site, the coral heads tend
to be deeper below the surface, as far west as Nusa Roviana, although there is a lattice of
shallow reefs along the northern (mainland) coastline of the lagoon in this region.
Seagrasses are not common west of Miho, other than sparse occurrences along the shallow
mainland coast and among the forest of islets between Miho and Zangana.
276

Forty-three percent of the worlds mangrove species are represented in Solomon
Islands, with 26 species in 13 families (Oreihaka 1997). Mangroves protect many of the
Lagoon’s shorelines from wind and waves, having a similar stabilizing role to seagrasses,
and harbouring crustaceans, molluscs and many different fish species, some of which are
agents of bioturbation. Mangroves may create sedimentary strata post-deposition. At the
Hoghoi site, for example, testpits revealed an orange-brown organic layer beneath the top
10cm of sediment, composed almost entirely of living (mangrove?) rootlets, with sherds
found both above and below this layer. This could end up looking like bedding planes in
a deposit of pottery fragments after the death of the mangrove system, which would leave
a thin dark organic band in an excavation section profile. Mangroves are found principally
along the mainland coast of the lagoon, where soft sediments predominate, and are present
in smaller areas along parts of the raised-coral barrier system, where they are found
associated with softer sediments.
Solomon Islands Tides:
The reason tides are significant is that land and sea breezes follow a diurnal cycle, as do
tides in the region for much of the year, thus creating a situation in which wind directional
changes or changes in wind strength are in phase with tidal changes for extended periods,
as will be detailed below, leading to differential wave exposure of coastlines in some
seasons. Numerous shallow reefs within the lagoon shelter many of the shorelines at low
tide (when intertidal scatters are most at risk of wave damage) but for those shorelines not
sheltered by reefs within the Lagoon, sea-breeze wave exposure at low tide is thus likely
to have had a more severe effect on the pottery scatters, which are more exposed to
swash-zone energy at low tide.
Lunar-induced tides have a fundamental periodicity of 12 hours and 25 minutes,
277

and operate over a cycle of 27.3 days governed by the rotation of the moon-earth system.
Usually, in most parts of the world, for this reason, the time of the high tide shifts slightly
each day (the diurnal calendar being based on a 24 hour solar periodicity). Lunar tidal
forces are also influenced by the moon’s declination either side of the equatorial plane,
which varies over a cycle of 27.2 days, and controls diurnal variation. The Moon’s
elliptical orbit also causes variation in tide-producing forces over a 27.5 day cycle
(Macmillan 1966:48-49). Solar tidal forces are of smaller magnitude than lunar tidal forces,
with a semi-diurnal period of twelve hours.
The declination of the sun varies over a yearly cycle, and the elliptical orbit of the
earth around the sun has a slight effect on a yearly cycle also. Diurnal inequalities (the
tendency for the two tides of a 25 hour period to be of unequal height) are affected by
declination of the sun and moon, and will be greatest at times of maximum declination, and
least at times of minimal declination (Macmillan 1966:44). This diurnal inequality is
significant for the Roviana case, where for much of the year a solar tidal cycle
predominates, as will be explained below. When solar and lunar effects are in phase, either
in conjunction or opposition, large tidal ranges occur, and when they are out of phase the
range is small. These are referred to as spring tides and neap tides respectively, changes
which follow a 29.5 day cycle (Macmillan 1966:50).
The Pacific ocean appears to favour the occurrence of small-range diurnal tides,
and furthermore, the solar contribution sometimes exceeds that of the lunar in the Pacific
area,
“...probably due to the resonance of the solar disturbing period with the
natural period of the body of water affected. In consequence, under these
circumstances the diurnal tide due to solar influence dominates that due to
the moon (Macmillan 1966:60).”
Solomon Islands tides seem to be of the mixed type, varying seasonally between diurnal
and semi-diurnal with the changing influence of solar forces. Tide predictions for Honiara
in year 2000 and 2001 (Sea levels for Honiara were obtained from the National Tidal
278

Facility, The Flinders University of South Australia) show a mixed tide regime where a
diurnal tidal cycle (1 high tide every 24 hours) is dominant for most of the year, with high
water occurring near noon in January, February and early march, tending to have a
morning high water in March, and shifting to a semi-diurnal regime in April (two tides in
24 hours), with only small tidal range for this cycle. May sees a return to a diurnal cycle,
but with low tide during the day and high tide in the early hours of the morning. June is
dominated by a diurnal cycle with low tide at mid-day and high water at night, and this
pattern continues through July and August, being modified towards a semi-diurnal cycle
in September and October. November sees a return to a stronger diurnal cycle, but with
high water during the day, with the same pattern in December. Although tidal records for
the New Georgia group were not accessed, this pattern is consistent with the author’s own
observation and information from local informants during 1996-1998 at Roviana Lagoon.
This annual pattern shows a strong influence of solar declination, dividing the year
into four tidal seasons, with the sun creating a semi-diurnal effect at the equinoxes and a
diurnal regime at the solstices. Within this annual pattern is a monthly cycle created largely
by the changing declination of the moon, and the rotation of the moon-earth system. There
is a daytime high tide from November to March (October to January according to Aswani
1997:235), and a night time high tide from May through September. The timing of daily
high-water is following a solar cycle rather than a lunar one (as outlined by Macmillan
1966:60).
The daytime solar tide is superior around the summer solstice, and the night-time
tide becomes superior around the time of winter solstice. Why this should be in the daytime
through the months of northern solar declination, and at night through the months of
southern declination is explained by a the mechanism of diurnal inequality (Macmillan
1966:42), whereby the declination of the relevant tide-producing body (in this case the sun)
produces, instead of the theoretical equilibrium tide, superior and inferior semi-diurnal
tides (in the Roviana case the inferior tide is completely suppressed when the declination
of the moon is low, but is present as a slight tidal oscillation when declination of the moon
is high, on a monthly cycle).
The broad characteristics of this tidal system as explicated above seem likely to be
a pattern of antiquity as well as the present, as there seems to be few reasons for this
to have varied over the timescale of the past few thousand years. Only at the geological
timescale, where the Pacific Ocean basin undergoes major changes in plate tectonics, can
this resonance with solar tidal forces be expected to change, as a result of the changing
279

size and depth of the basin, within the current (and longstanding) harmonic theory of tides.
Winds:
The nearest meteorological station to the Roviana Lagoon is situated in Munda, away from
the coast, surrounded by trees and buildings, and while this location is no doubt convenient
for staff who take the readings, recorded windspeeds and directions cannot be expected
to bear much relationship to winds affecting the Lagoon. This section is therefore written
using a mix of data pertaining to the general region acquired from the Pacific Pilot
(Hydrographic-Office 1971: diagrams 8-11), data from personal observation, and data
from Aswani’s thesis (Aswani 1997:238-239). (Author’s note: the airport building was
shifted circa 2001, allowing slightly more exposed location for weather equipment.)
The southeast trades (Gevasa) prevail from April/May/June through to about
October, and their strength and duration varies from year to year. Direction is recorded as
blowing from 125 degrees true (Hydrographic-Office 1971:diagram 10). Personal
observation over the course of three years suggests the local direction of this wind is more
like 135 degrees, as at the Honiavasa reef passage it blows consistently directly from
Tombatuni Island in the Blanche channel. This wind tends to strengthen during the day,
probably due to the local effect of the sea breeze, at a time of year when tides are low
during the day, so that any pottery scatter at such a depth as to be exposed to the
Southeast, or to refraction waves from the Blanche channel curving around the inner points
of the reef passages, can be expected to have undergone a relatively harsh taphonomic
process. Low tides during the day during the season of the SE trade mean that any
intertidal pottery sites with a large fetch to the SE (e.g. Gharanga, or the Mbolave lithic
scatter/ceramic findspot) can be expected to undergo a harsh swash-zone taphonomic
regime, unless sheltered by reefs or soft sediments and vegetation.
The Northeast trade blows strongly for a period of a month or so in late summer,
but locally from the Northwest, and is known by local expatriates as the Northwest
Monsoon (this may be the “Peza” referred to by Aswani, but the period and direction
differ from Aswani’s suggestion that this wind is a westerly. The Roviana dictionary lists
Peza as the northwest monsoon (Waterhouse 1926:91). The recorded direction of origin
of the NE trade is about 70 degrees true (Hydrographic-Office 1971: diagram 9). Personal
observation suggests that direction is locally influenced by the New Georgia landmass
280

with its volcanic peaks, tending to blow more from the north to NNW, and tending to
strengthen in the evening and at night with the addition of a slight land breeze. Tides at this
time of year are predominantly semi-diurnal (two highs and two lows every 24 hours), so
are not in phase with this strengthening of the wind in the evening. Sites exposed to
significant Northwesterly fetch in this direction at low tide can be expected to show some
effect of this harsher taphonomic regime.
Transitional periods between these seasonal prevailing winds are dominated by a
diurnal cycle of light land (night) and sea (day) breezes (Sea breezes blow in from the
Blanche Channel toward the centre of New Georgia during the day, and land breezes blow
off the coast towards the Blanche Channel at night). These breezes are most noticeable at
midsummer, in the absence of other winds, but the forces generating these winds, the
differential heating and cooling of land and sea, act in combination with other winds in the
cooler months also. When tides are high during the day, sites with seaward fetch are
unlikely to be affected as the pottery is well submerged, but sites with landward fetch at
low tide at night (Hoghoi for example) might be expected to be exposed to a build-up of
small waves through the course of the night.
Tropical depressions, tropical storms, and cyclones tend to originate between
Latitude 10 degrees south and Latitude 15 degrees south (Hydrographic-Office 1971:36
and diagram 12). There are an average of about two cyclones per year in the region, and
these generally pass to the south of the Solomons, and can be expected in most instances
to create strong winds veering (progressing around the compass clockwise) through
northerly as a result although other directions are not ruled out, as some rare cyclones
originate in Solomon Islands. Holocene climate changes may have affected cyclone tracks
and other climatic zonation in the past (Tarlton 1996:18-28), but no information is
available at present on the nature of such changes in the period since deposition of the
pottery. It seems likely that the presence of the high landmasses of Choiseul and Ysabel
to the east are moderating influences restricting the formation of cyclones in the region,
and that this effect is of great geological age.
The season for tropical storms is December through April, with most occurring in
January, February and March. As their average track speed is about eight knots, they can
affect an area for many days and nights (Hydrographic-Office 1971:36), so tidal periodicity
is irrelevant to wave exposure during tropical cyclones.
Cyclone storm surge may (but does not necessarily) raise sea levels temporarily,
reducing the effect of waves raised during a cyclone on the intertidal pottery scatters,
281

elative to effects of lesser winds prevailing at times of low tide. Positive tidal surges of up
to 4m occur during tropical cyclones (Open University Oceanography Team 1989:61), and
in some cases this property of tropical storms and cyclones would be expected to reduce
the effect of storm generated waves in the lagoon on submerged pottery scatters. Storm
surge is caused primarily by winds driving water ashore and this will depend at a larger
scale on whether wind is onshore or offshore, but within the “radius of maximum wind”,
several kilometres from the storm centre, surge response to barometric pressure anomaly
is universal for cyclones (Anthes 1982:5-7, 161), although this effect is slight. Where
cyclones pass to the south of the Solomon Islands, as is usually the case, and winds can be
expected to veer through northerly, storm surge would be expected to be slight or absent,
and was observed to be slight during such an event in 1998, but if a cyclone were to take
a more northerly route, storm surge from Easterlies, Southerlies or Westerlies could be
expected in the embayment comprising the Blanche Channel. In terms of cyclone events,
the circumstance most likely to affect intertidal pottery scatters would be the more
common northerly-tending cyclone winds with attendant storm surge only slight or absent,
as pottery might be exposed to wave action in the lagoon at low tide in these
circumstances. Sites with significant fetch in a northerly direction (NE through to NW)
could therefore be expected to be more severely impacted by cyclones.
Measuring Fetch:
Fetch for cyclones was measured as the maximum at low tide in a northerly arc between
NW and NE. As the direction of the NE trade is uncertain on present information, fetch
for this wind is measured as a maximum between 320 and 350 degrees (consistent with
personal observation) at low tide; and also at 70 degrees at low tide, consistent with the
direction of the NE trade shown in the Pacific Pilot. Fetch for the SE trade is measured at
low tide bearing 140 degrees, consistent with personal observation. Fetch was not
measured for the summer sea breeze because it coincides with the diurnal summer-solstice
daytime solar tide, and is unlikely to affect the site areas studied. Fetch in the direction of
the land breeze was measured as the maximum at low tide 25 degrees either side of North.
Fetch was measured using air photographs at a scale of 1:33 000.
282

Results:
Fetch under various combinations of wind and tide are given in Table 21.
Table 21: Fetch measurements for collection sites as an indicator of wave exposure.
Site Cyclone
(NW-NE)
Paniavile 2km
(moderated
by seagrass
and soft
sediments)
Hoghoi 6km
(seagrass)
“NE” trade
(Peza?)
283
SE trade
(Gevasa)
land breeze
(low night)
Maximum
fetch (max
exposure)
0 0

pottery discard is less likely to have survived.
Hoghoi had a wave exposure measurement of up to six km to the NNE, with wave
effects moderated due to protection at low tide by an extensive shallow seagrass flat with
numerous small coral heads nearing the surface. If sea levels were higher in the past this
protection would have been absent or reduced, but the ceramics would have been in deeper
water, mitigating swash effects to some extent.
Gharanga could have up to six km exposure in SE tradewind conditions allowing
swash to affect the site at intermediate stages of the tide (this effect may have been
extended in the past if sea levels were slightly higher, with the materials in shallow water
rather than exposed at low tide). Gharanga and Kopo are different to the other sites in that
they are situated at the mouth of mainland streams. Sherds did not have the cushioning
protection of marine growth to the same extent due to reduced growth of sponges in the
low-salinity location, and could also be subject to taphonomic effects arising from stream
flooding.
Miho and Honiavasa both yielded maximal wave exposure of 6km but Honiavasa
was more exposed to refraction waves in SE tradewind conditions (Miho is better
protected from such waves than Honiavasa due to more extensive shoals adjacent to the
passage and situation in a small embayment, while Honiavasa, by contrast, is located quite
close to the deeper water of the passage, and the author noted refracted waves while
collecting there on a rising tide in tradewind conditions. While the seaward portions of the
site were in deep enough water to be protected in these conditions, shoaling towards the
centre of the site left the latter area exposed to swash, while the eastern end of the site was
protected by the central shoal. Sediments across the site support the idea of a central area
affected by swash from refraction waves, where ceramics are exposed as large fragments,
and a more sheltered easterly zone where ceramic visibility drops away as sandier
sediments build up. The picture could look quite different after a strong northerly though.
284

Nusa Roviana has up to 3km of fetch in a NNE direction, and can be expected to
show some signs of damage or size sorting in view of this relatively exposed location (in
modern times a wharf/breakwater has been constructed here to shelter the landing in
northerly conditions). Sherd surfaces were generally quite damaged in this site with temper
grains slightly pedestalled in many cases (the single dentate sherd was an exception,
retaining a smooth finish).
Unanalysed assemblages from Ririgomana, Punala, Humbi Kongu and Rango were
all in poor condition with sherds showing signs of water rolling in many cases. More
recently, large sherds in good condition have been reported from the deeper margins of
Rango (Rigeo, Pers. comm., 2002). Unlike Paniavile, Ririgomana is free of seagrass, and
although the water is shallow generally in the area between Ririgomana and the mainland,
I can envision a chop of small waves building up along this shore in a northerly. Punala and
Humbi Kongu were similar in this respect.
A number of findspots and lithic scatters on the rocky shore of Honiavasa Island
to the east of Hoghoi had a lesser degree of seagrass shelter than Hoghoi, as attested also
by rockier coral shorelines. The low-density scatter of sherds, especially lithic manuports,
along this shoreline does not preclude relatively intensive stilt-house settlement in the past,
with most ceramics removed by swash-zone rolling post-deposition. Similarly, the ceramic
findspot at Kazu was on a stony shoreline, and may be a chance survival of significant SE
trade exposure. Lithic manuport scatter at Mbolave, where a single sherd was found, could
also be a case of a ceramic/lithic scatter transforming into a lithic scatter through time.
The fetch data predicts that Honiavasa, Miho, Gharanga and Hoghoi should show
the greatest level of sherd damage and size-sorting, closely followed by Nusa Roviana.
When the sheltering effect of seagrass shallows is taken into account at Hoghoi, this
285

seems to partly match the size-data from sites (Table 22) in that Honiavasa and Miho had
larger average sherd sizes than other sites (the 25 sherds from Kopo were largest but this
may be a sample size and/or collection intensity effect).
Table 22: Total sherd count and average sherd area by collection site,
resulting from combined effects of collection intensity and wave exposure.
Site Sherd Count Average Sherd
Area(cm 2 )
Paniavile 644 14.65
Hoghoi 861 10.94
Miho 382 22.08
Honiavasa 442 28.25
Gharanga 277 17.96
Nusa Roviana 115 15.83
Kopo 25 33.24
Zangana 860 13.59
Chapter Summary and Conclusions:
It was difficult to construct an unequivocal model of wave exposure given conflicting
information about the direction of prevailing winds, and uncertainty as to the relative
importance of the various winds. Personal observation and local anecdotal information
conflict with the information in the 1971 Pacific Pilot, and the latter is probably only
intended as a very general guide for navigational use. Aswani’s recording of Peza as a
westerly may be a local effect due to topography at Baraulu Village (Canaan), where he
was mostly based, and a northwesterly direction for this strong wind is preferred here.
Wave exposure for some sites (e.g. Gharanga and Hoghoi) is much higher using
286

Figure 75: Location of Honiavasa and Miho collection sites
at the inner ends of a long passage through uplifted barrier
reef. Miho is more sheltered from refraction waves than
Honiavasa as it is situated in a small bay oriented to the
NW, while the western edge of Honiavasa drops away into
Honiavasa passage.
personally observed wind directions, and any future modeling of this sort would benefit
from more formal recording of wave exposure at various locations in the various prevailing
conditions.
This analysis provides a caution against a direct reading of aceramic scatters as
having a chronological or behavioural significance, suggesting that some of these at least
may be the result of a harsh taphonomic regime, which has biased the observed assemblage
composition towards relatively robust coarse-grained lithic manuports.
Kopo and Paniavile seemed to be the more sheltered locations, and this seems
consistent with the information initially recorded regarding Paniavile, which was a rich site
prior to a series of collecting and fossicking events (Reeve 1989). The sample of sherds
from Kopo is too small to yield useful taphonomic analysis results, but collection intensity
was fairly high (Sheppard, pers. comm. 2002) suggesting either a harsher taphonomic
regime than predicted by the wave exposure model or a low intensity-duration of
287

occupation in the past. It seems likely that a number of such sites may be well preserved
along the mainland shoreline under soft sediments, in locations sheltered from all the major
wind/fetch combinations (the sort of locations which can be expected to be characterized
by a buildup of soft sediments). The reported ceramic find-spot at Pikoro may be one such
location, although limited test-pitting to the water table revealed a sparse distribution of
lithic manuports and some small sherd fragments from the adjacent stream banks.
None of the ceramic sites had more than 6km maximum fetch, suggesting this may
be an upper limit for site preservation in the water depths in which the sites were found
(intertidal-subtidal). This analysis thus provides a starting point for designing a model of
site preservation in relation to wave exposure for this type of site. The model could be
extended by adding data from other areas where similar sites are found at different depths.
This sort of modeling process is fundamental to understanding the distribution of Lapita
across Near Oceania. The Reef-Santa-Cruz terrestrial Lapita sites at the near margin of
Remote Oceania and terrestrial sites in the Talasea region may mark a shift by some early
Lapita-era people onto land.
288

CHAPTER 7:
SITE/ASSEMBLAGE FORMATION PROCESSES
Introduction:
(SHERD TAPHONOMY).
Determination of formation processes (or at least thorough consideration of the
possibilities) is widely regarded as an essential stage in making archaeological inferences,
although,
“Many archaeologists simply ignore the subject, perhaps because they are
unconvinced of its importance or wary of its implications for the validity
of their interpretations of the past (Shott 1998).”
Interest in archaeological formation processes burgeoned during the 1970s (Schiffer
1995c), however, for archaeological sites located in the sea, development of formation
theory and method has lagged by comparison (Stewart 1999). A key question facing the
analyst is whether comparisons between samples, for example, seriation, are invariant
under the transformations brought about by site formation processes (Orton 2000:66). For
the Roviana intertidal sites, another fundamental behavioural question is whether
deposition was in the sea in the past, or whether submergence/coastal erosion has made
intertidal sites out of terrestrial sites. Although marine transgression is primarily a
destructive process (Waters 1992:275), it is clear that coastal occupation sites can survive
submergence relatively unscathed in ideal circumstances, these being where sites are
sheltered from wave energy (Flemming 1983) or where submergence (and sometimes
burial in sediments) is rapid (Stewart 1999:572, Waters 1992:277), as in the case of Port
Royal (Hamilton & Woodward 1984), and sufficiently deep to evade wave action (Waters
1992:250-251). Comparison of Roviana density of deposits with the quantities of sherdage
289

expected from estimates of the breakage population (Felgate 2002, Felgate & Bickler n.d),
as discussed in Chapter 5, suggests that Roviana sherd samples are far from pristine, and
that substantial sherd attrition has occurred during formation (see Chapter 6). The
possibility exists that the sites originated as terrestrial deposits and have been damaged in
the course of becoming intertidal deposits.
I will argue below against submergence or coastal erosion of terrestrial sites as an
explanation for the observed archaeological distribution. My arguments in doing so closely
parallel those of Reeves, Specht and Wickler for similar sites elsewhere in Near Oceania
(Reeve 1989, Specht 1991, Wickler 2001), and follows closely the argument to this effect
presented previously for Roviana underwater sites (Felgate 2002). Recent paleoshoreline
research is cited as additional support of this argument (Mann et al. 1998).
In this chapter taphonomic approaches to the identification of formation processes
are taken, seeking to determine, from the characteristics of artefact assemblages and
individual artefacts, the processes that have acted on these in forming the observed sites.
A series of models of cultural formation processes are presented below, followed by a
series of post-depositional formation process models. A range of sherd-based taphonomic
analyses contribute to discriminating between these various formation models. In Chapter
11 a spatial approach to the identification of postdepositional process is presented, while
in Chapter 6 wave exposure was modelled for sites, and compared with other taphonomic
information in this chapter.
Beyond the identification of formation processes, a second major aim in seeking
to understand taphonomic effects on assemblage composition is to assess the likely nature
of taphonomic bias in assemblage composition. Assemblage stylistic composition is an
important data source for chronological inference, through seriation analysis (see Chapter
12). While it does not seem possible to fully understand the exact nature of assemblage
biases arising through taphonomic processes, a number of observations can be made that
inform in this regard, and which aid in the analysis of decorative variability.
290

Review: Middle Range Theory for the Identification of Formation Processes:
A general review of suggested methods by which formation processes might be identified
is given here. The classic review is that of Schiffer (1995c), whose categories of evidence
are used here.
Size effects, density, shape, orientation, structure and context of the deposit:
There is a substantial literature on the size-sorting effects of cultural transformation
processes such as cleanup activities in the past. The McKellar principle, for example
states,
“ that small artefacts, especially microartefacts on occupation surfaces
often indicate primary refuse, whereas in activity areas not habitually
cleaned, such as some lithic workshops, abandoned structures... and vacant
lots..., larger items can accumulate as primary refuse (Schiffer
1995c:175).”
Perhaps more important than such behavioural correlates in the present context are non-
cultural size-sorting effects as governed by the laws of hydrology or hydraulic transport,
viewing archaeological items as sedimentary particles (e.g. Shackley 1978). There are
many studies dealing with stream and river environments from this perspective (Waters
1992:321), and also hydrological experimental sedimentological approaches such as flume
experiments (e.g. Day 1980). For coastal sediments, wave processes assume a greater
importance than currents, as waves entrain sediments in a zone extending from the wave
base to the landward margin of the swash zone, sediments which are then transported by
beach drift, longshore current, and rip currents (Waters 1992:249-251). These
erosive/transportation processes fluctuate in magnitude during changes in weather, and are
displaced landward and seaward with tidal fluctuations. When such coastal high-energy
processes come into contact with archaeological sediments, the fine fraction of the
artifact/sediment matrix becomes suspended in the water column, while a lag deposit of
291

larger/heavier items may be abraded by suspended sand or reworked by rolling (Waters
1992:270-271).
These processes can create a swash-zone lag deposit by eroding and transporting
terrestrial sediments and artefacts, or can rework materials initially deposited in the water
as primary refuse. Both terrestrial and in-water discard have similar consequences for the
size, shape, density and orientation of artefacts in some respects. Artefacts in such settings
should occur in a matrix of similarly large or dense sedimentary particles. Some shapes will
transport more easily by benthic rolling than others, while flatter shapes might be more
easily suspended in the water column by turbulence than blockier shapes. The same across-
shore size/shape/density-sorting evident in the sedimentary particles should show up in the
size patterning of artefacts. Armouring of graded sediments may also occur, where a
surface layer of larger, less easily transported items protects less-sorted underlying
sediments and artefacts.
To distinguish between erosion of a terrestrial deposit into a maritime lag deposit
and primary deposition in the sea, one must look not only at the lag deposit itself, but
landward, in search of traces of a source of the artefacts (remembering also that storm
wash can transport materials from sea to land). This is the line of reasoning used by others
in suggesting reef flat lithic/ceramic scatters represent deposition in the sea in the past
(Reeve 1989, Specht 1991, Wickler 2001: 241). Absence of adjacent terrestrial erosion
features that might supply a source of artefacts would suggest that the materials were
deposited in the sea directly by some means.
In deep water, deposition could occur as a result of a shipwreck or capsize, or
funerary/religious practices, or discard from anchored vessels (Stewart 1999). In the
intertidal zone or on the reef flat, one would look shoreward in search of an attractant that
would cause a concentration of activity in such shallow water, such as a local resource (for
example water or gardening land) that might cause intensive canoe traffic and an
accumulation of pottery breakage events and discard. If there is a widespread pattern
292

comprising many instances of reef-flat or intertidal site locations, without attendant
deposits on land, and without obvious localized attractant resources, this is more likely a
pattern resulting from settlement. Artifact accumulation quantity/density and assemblage
composition/diversity can allow some further idea of site function and occupation intensity
in such circumstances.
Artefact size/density/shape should give clues here, even if the shore deposits have
been reworked by storm wash. In the case of primary deposition on land, the terrestrial
artefact deposit should show less size/density sorting than the adjacent marine lag deposit,
as it cannot have been reworked to the same extent as the materials in the sea. In the
circumstance where the artefacts on land are rounded and small, while the artefacts in the
sea are of a greater variety of sizes and in good condition, one must conclude either that
some harsh taphonomic process such as gardening has reduced sherds on land, or that the
small sherds on land are a wave-transported size fraction that originated in the sea. Even
in the circumstance where sherds on land are not particularly well sorted by size, there is
the possibility that a more random size-selection has been cast up from the sea by large
waves, and that only some lucky sherds still in the sea have been sheltered from the force
of the water by microtopography. In such circumstances other geomorphological
characteristics of storm wash deposits aught to be identifiable (for example Gosden &
Webb 1994).
Processes affecting artefact orientation and dip are not particularly well understood
as yet. Dip may be particularly useful for identifying primary deposition of artefacts
tramped onto a flat surface, and orientation has been used to infer aeolian and fluvial
processes (Schiffer 1995c:178). Marine currents and some propwash effects from treasure-
hunting techniques may leave patterned artefact orientation (Stewart 1999). Strong
currents may occur at times in reef-flat locations near reef passages.
Use Life, Damage and Accretions:
293

Schiffer notes that burials, caches, and floors of structures may contain whole pots with
remaining use life (Schiffer 1995c:178). Conversely, refuse deposits will seldom contain
pristine articles. Postdepositional reworking can obscure remaining use life by breaking
pots, but other types of use life evidence, such as the degree to which stone adzes have
been re-sharpened in comparison to unused reject preforms, are more robust. Intuitively,
ovenstones are a robust item that may preserve indications of remaining use life better than
pottery, which could suggest site abandonment or lack of economising behaviour.
Damage such as edge-rounding, surface ablation, spalling and chemical weathering
of mineral constituents can give clues as to the use and taphonomic history of an artefact
or fragment (see Schiffer 1995c:178-181 for a review). Water transport can cause rapid
rounding of sherd edges (Skibo & Schiffer 1987) and abrasion by rolling can be
distinguished (on lithic implements at least) from edge-rounding by smaller waterborne
particles (Shackley 1974). Surface ablation and pedestalling of temper grains can result
from water transport of sherds, and wetting/drying cycles in a saline environment can be
expected to produce surface spalling or exfoliation (Schiffer 1987). In the extreme
weathering environment of the tropical Pacific, calcareous temper grains are known to
sometimes have dissolved after deposition, leaving voids in the fabric of sherds (see for
example Dickinson 2000b), but there is no evidence for this occurring in sherds deposited
in the sea. Sherds formerly on land and subsequently deposited in the sea as the result of
erosion might be expected to display this effect at some level, particularly if it is thought
they were eroded from acid soils. Lack of weathering of calcareous temper grains may
therefore be indicative in some settings of primary deposition in the sea.
It was observed through differential conservation during the project that complete
drying of sherds without desalination caused rapid degradation of the recovered
assemblage, whereas even rough methods of desalination showed a marked improvement
in the condition of analysed sherd surfaces and reduction of sherd crumbling. The absence
of salt crystallization damage on sherds might suggest they have spent most of their lives
294

in either a terrestrial or an underwater context, rather than in a cyclic wet/dry salty
environment.
Marine accretions have the potential to inform on taphonomic processes and
landscape change. Sherds found on land that have accretion of marine organisms (coral
growth for example) must have been in the sea at some time. Similarly, sherds with coral
growth found in shallow water above the coral growth line may be evidence for landward
transport, or for an emerging coastline. The converse is not true though, sherds found on
land without such accretions need not have been deposited on land through their entire
history. Abrasion and weathering can remove such accretions, which do not necessarily
form in the first place.
Artifact Diversity:
The range of types of artifact in deposits is sensitive to cultural formation processes,
especially the occupation span of settlements, but also to differences in the functions of
settlements and activity areas (Schiffer 1995c: 182-183). Schiffer’s discussion is weak on
taphonomic effects on artefact diversity, and sampling effects on diversity generally (e.g.
Kintigh 1984, Shott 1989a). Lag deposits, for example, may be size or density sorted,
reducing diversity by favouring chunky types, those that break into large pieces, and
different materials are differentially preserved in different depositional contexts, so there
is always potential for taphonomic variation in diversity. Surface or lag deposits are more
subject to collector effects, which can be expected to reduce diversity rapidly unless items
are highly fragmented (Felgate & Bickler n.d). While Schiffer suggests that great diversity
is found in secondary refuse deposits containing refuse streams from a settlement’s entire
range of activities, this assumes a well-preserved deposit. Deposits lacking such diversity
may do so as a result of taphonomic processes rather than cultural formation processes.
Shott notes that richness as a measure of diversity, as used by Kintigh, fails to
account for the evenness or equitability of item frequencies by class (Shott 1989a). Where
295

large numbers of some classes of artefact, for example ceramics, exist at a location, and
small numbers of another class, for example adzes, or where count data by class is sparse
(low average abundance within classes), both richness and other more sophisticated
measures of diversity such as the Shannon-Weaver information statistic can be sensitive
to sample size effects. Shott goes on to suggest ways of inferring use-lives of tools from
relative frequencies in diversity-saturated assemblages, like Schiffer, with minimal attention
to the recognition of taphonomic reduction of diversity. He seems not to expect a strong
taphonomic bias in the diversity of types within a single material class, e.g. lithics. It seems
that a fundamental question in making use of diversity measures to identify behavioural or
organizational (e.g. use-life) properties of assemblages is the effect of taphonomic biases
on the data.
Diversity therefore seems to me to be a poor route into the understanding of
formation processes for sites in a poor state of preservation, as indicated in the Roviana
case (see Chapter 5).
Representation of Parts:
Use of part representation in ceramic assemblage formation studies is in its infancy
(Schiffer 1995c:186) and preliminary ideas in this regard suffer from a convergence with
inference from part representation in faunal taphonomy. Schiffer, for example, envisages
part selection for recycling by potters as the sort of process that might be identifiable
through part representation, rather like the way animal scavenging often involves selection
of skeletal parts. Given that there is no evidence for use or re-use of sherds as grog temper
in the Roviana materials, and only minor possible evidence for re-use of bases as bowls or
frying pans (see Chapter 8), analysis along these lines would not seem to be a productive
exercise.
Other mechanisms, such as unrecorded collection by archaeologists and others, can
be viewed as taphonomic processes affecting the composition of assemblages of artefacts
296

(Ruig 1999). While there is as yet no generally accepted theory of the effects of collecting
on ceramic part representation, some expectations are developed below in relation to the
Roviana samples.
Archaeologists and other destructive collectors are expected to pick up larger items
first, and those with more information or aesthetic value. Whole adzes, more complete
sherd or whole pots, are likely to be the first things collected from sites like the Roviana
intertidal scatters. Next in order of preference are smaller items with obvious information
content: decorated sherds, sherds on which the vessel lip, neck or shoulder/carination are
represented, smaller lithic artefacts, or larger fragments of lithic artefacts. Some
archaeologists may also recognize the information value of exotic lithic manuports such
as sandstone fragments or ovenstones.
Non-archaeologists might selectively remove sherds or lithics for purposes other
than information gathering. Anecdotal evidence suggests two processes of this sort,
skipping of sherds into deeper water by children, which should affect mainly sherds
lacking complex curvature for hydrodynamic reasons, and collection of sherds and lithic
artefacts by villagers following collection by archaeologists, where a perception of
commercial value has arisen. In this instance collector effects can be expected to closely
mirror the information-value model of archaeological reconnaissance collection. A series
of models of sample characteristics in line with these collection theories will be
constructed below.
Sediments:
The sediments comprising the finer fraction of the matrix of sites may include
archaeosediments (Waters 1992:16-17). Textural properties, mineralogy and chemical
properties of sediments may inform on human behaviour (cultural formation processes) or
on postdepositional process that have resulted in the formation of sites. For the Roviana
sites, postdepositional sherd attrition as evidenced in Chapter 5 suggests the potential for
297

a disaggregated pottery-derived component within site sediments. This is discussed further
below in relation to models of site formation.
This inventory of formation theory was used to develop a series of models of
formation and the material correlates of these. Models for cultural and natural formation
processes are listed separately below, but there is substantial overlap in the correlates
identified for the models, so data will be considered together, followed by a synthesis and
conclusions.
Models of Cultural Formation Process:
The investigation of cultural formation processes is structured in terms of the following
competing models:
• Ceramic/lithic scatters formed as terrestrial discard from coastal settlements,
subsequently inundated by the sea. The correlates of this model are evidence for
coastal erosion features, poorly sorted artefact deposits on land and possibly sherd
damage consistent with a terrestrial weathering regime, such as the weathering of
calcite grains.
• Ceramic/lithic scatters formed as secondary accumulations in the sea when
people living in terrestrial settlements dumped sherds and ovenstones into the
sea. The correlates of this model are evidence for structures and midden on land
from former settlements, or erosional features that suggest obliteration of such
evidence.
• Ceramic/lithic scatters are discard from stilt settlements in the
intertidal/subtidal zone. Expected correlates are minimal contemporaneous
evidence for occupation on adjacent land, with any ceramic-lithic evidence on land
showing evidence of size or density sorting, and associated with geomorphological
298

stormwash features; we would not expect to see much in the way of stone or coral
building materials scattered about if timber posts were the primary means of
supporting post and thatch structures.
• Ceramic/lithic scatters formed as discard from now-defluved artificial islet
settlement located in the intertidal. This should look similar to stilt houses, and
we might also expect to see a scatter of coral stones or boulders that formerly
comprised the sea-walls of such structures, although mining for building materials
in later periods might remove such evidence.
Some additional cultural formation models that are not mutually exclusive with the above
list can be advanced:
A. Site abandonment might have a correlate of remaining use life of adzes and
ovenstones, where dicarded adzes show little sign of resharpening, and many large
unfractured ovenstones remain on site, but mining of ovenstones and collector
effect on adzes could remove such evidence.
B. Whether terrestrial or over water, permanent settlement should have the dual
correlates of diversity of artefact types and substantial breakage populations of a
diverse range of functional types of pottery; conversely, short-term and/or low-
intensity use should be evidenced by smaller accumulations and variable artefact
inventories.
C. Specialized site functions should result in low-diversity artifact assemblages, with
preponderance of particular vessel forms. Conversely, generalized settlement
should have a diversity of special-purpose forms alongside general-purpose forms.
Models of Post-depositional Formation Process:
299

Substantial sherd attrition through post-discard processes (as shown to have occurred in
the Roviana cases in Chapter 5) raises the prospect of taphonomic sampling bias in the
surviving sherd population. In addition to the cultural formation models outlined in the
previous section, a series of models of sherd attrition factors will be constructed and
considered in relation to the Roviana data. These, with their correlates, are outlined below.
A. The transport model, where the missing sherds were transported by wave action
into the higher-energy environment of the strandline and destroyed by mechanical
abrasion. Under this model sherds which were sheltered from wave energy by local
microtopography or sediment armouring form either a random selection of sherds,
or a selection structured by sherd transportability. Also, weak sherds should be
recovered, but extremely thin or light sherds should exist principally as small
fragments on the strandline. Sites with a greater wave exposure should show
greater evidence for size-sorting.
B. The dis-aggregation model, where poorly-fired sherds or sherds otherwise having
poor fabric strength have disintegrated in place regardless of local
microtopography. Under this model, soft or weak sherds should not be present in
assemblages, although porous sherds suffering salt damage (Schiffer 1987:277) on
retrieval may be present. An assessment of fabric strength variability (as expressed
by sherd size) in relation to temper types of the recovered sherds will be presented
in this chapter to try to determine whether much strength variability has survived.
If it has, this favours a transportation model, and if not, a disaggregation model is
favoured. Sherd size in relation to thickness and fabric will be used as a proxy for
sherd strength. No experimental strength tests on sherds were performed. (As will
be argued below in relation to signs of weathering, post-depositional reduction in
sherd strength does not seem to have occurred in any of the fabric groups during
their time in the sea.)
300

C. The collector model, where the ratio of plain to decorated diagnostic sherds is
substantially different between sites. One assemblage is known to have been
depleted in the past through collection by archaeologists and villagers, and this site
will be used as a control in identifying the effects of sherd collection by humans.
The number of sherds for which vessel part representation is high can be expected
to vary with assemblage brokenness, but also with the level of collecting by humans
in the past. Assemblage brokenness can be compared using body sherd size as a
control, and the ratio of eye-catching diagnostic sherds to body sherds can be
developed as an indicator of collecting effects in the past. One might however
expect some larger body sherds to be uplifted, which might render this model
equifinal with brokenness through wave exposure or trampling.
D. The sherd-skipping model, where anecdotal evidence for sherd skipping by
children (throwing a sherd or flat stone with a spinning action across the water
surface to cause it to perform a series of bounces) is tested against assemblage
composition. Assemblages that have been subject to a high degree of sherd
skipping should have relatively fewer flattish medium-sized body sherds than would
otherwise be the case.
E. The collection intensity model: in this model sherd size varies in relation to
collection intensity. The more intensively a site is collected, the more small
inconspicuous or undiagnostic fragments are retained into the sample assemblage.
Summary of Models:
The consequences of the actions of these various processes for archaeological sequence-
building objectives, as laid out at the beginning of the chapter, vary: the transport model
is liable to create bias in favour of heavier, thicker pottery, with light calcareous fabrics
301

possibly under-represented, and thin-walled pottery particularly likely to be under-
represented. The dis-aggregation model is likely to result in the under-representation of
non-cooking vessels, which may be lower-fired, and/or less tough through choice of
temper during manufacture.
This possibility is of particular concern for Lapita, since some have suggested that
highly decorated flat-based unrestricted vessels might have a serving function, or that
highly decorated Lapita vessels in general might have a ritual function rather than
utilitarian. If this were the case a harsh taphonomic regime might turn Lapita sites into
post-Lapita sites from the point of view of the ceramic chronologist. Wickler’s
documentation of dentate-stamped pottery in sites of this general type on the reef flat at
DAF and DES are reassuring in this regard, suggesting that there is a possibility of such
sherds being represented in un-buried intertidal/shallow-water sites (Wickler 2001:Chapter
5). Collector bias might render an assemblage plainer overall than it might have been, but
degradation of the overall sample of diagnostic sherds rather than analytical bias might be
expected, as the analyst can control for this effect by controlling for sherd
“diagnosticness”. The sherd skipping model does not seem to involve any particular
stylistic taphonomic bias, except if decoration is concentrated on the upper portions of the
potsherd skipping could affect the overall ratio of plain:decorated sherds, but many other
factors could also affect this.
Qualitative Evidence From Samples:
Damage:
Wetting/drying salt crystallization damage was not quantified during analysis, but sherd
condition and the spatial distribution of sherd condition was used to infer drying regime
limits for major portions of sites in the past. Most sherds at the deeper margins of sites
exhibit no sign of salt erosion damage or damage arising as a result of water transport
302

(surface ablation, pedestalling of temper grains, and severe edge rounding). It was inferred
that these have never been exposed to severe, sustained swash zone turbulence (rolling)
or drying. Also, there was no sign of postdepositional weathering of carbonate temper
grains suggesting the sherds had never been exposed to an acid environment. Some
carbonate grains were partially sintered, with a rind of carbonate still lining the sides of the
void around a shrunken grain, thought to be an artefact of manufacturing firing
temperature, but many sherds had un-sintered carbonate grains which aught to be liable to
solution if the sherds had been in an acid environment. Most sherds had slight edge
rounding, consistent with abrasion by waterborne sand.
Accretions:
No quantitative analysis was conducted, but some general statements are made. Accretions
of marine growth were almost universal, except for rolled sherds found on the strandline,
and at the Gharanga stream-mouth site where salinity was probably low. Accretions
included dead hard coral in some cases, live sponges and coralline/green algae in most
cases. In the deeper margins of some sites dead coral growth was common on sherds
(Honiavasa site for example) in spite of the lack of any evidence for active coral growth
nowadays. These accretions may be evidence for a high stand at or post deposition, which
is consistent with current data on sea-level history (Mann et al. 1998). Dating of accretions
might be a worthwhile future exercise for palaeo-sea-level studies.
Had wave action transported sherds up the shore in significant quantity one would
expect to find coral accretions on land as evidence. It appeared as though there was a
significant correlation between sherd condition and accretion presence; sherds from deeper
water tended to have coral and sponge accretions and were also larger and in better
condition. No coral accretions were seen on land, but this may be a sample effect, as the
number of sherds located on adjacent land was small (one sherd from Honiavasa, two or
three small sherds from Gharanga, none from Hoghoi). It could also be an abrasion effect.
Sherds were seen in the roots of a fallen tree a short distance inland from the Honiavasa
site but extensive testpitting in the area did not reveal substantial pottery deposits
(Sheppard, pers. com.).
Artifact Diversity:
Artefact diversity/inventory of material is discussed in Chapter 13.
303

Sediments:
In regard to the model of cultural formation of settlements located on artificial islets in the
past, now reworked, some sites were characterized by a scatter of easily transportable
coral blocks of the type used locally these days for building wharves and islets. Although
no quantified data were collected that bear directly on this issue, it was noted during
fieldwork that the gravels and coral cobbles of some site sediments could be the reworked
remains of such structures (especially at Hoghoi). No specific patterned features were
noted. As it is, only the speculative suggestion can be made, and this is a question that
must be left to future research. Cultural formation processes, like sea levels, should not be
expected to have been static over time, and the changing conformation of the shoreline
profile with falling sea-levels may have enabled or encouraged such cobble structures later
in time, where earlier the depth of water over the current reef flat may have precluded islet
construction as too labour intensive.
Micro-examination of two sediment samples from Hoghoi suggest that substantial
quantities of ferromagnesisn opaque minerals and clinopyroxene were present in one of the
samples, and minor quantities of fine ferromagnesian minerals and occasional small lathes
of pyroxene were present in the other. The first sample was collected by the author from
the vicinity of the densest accumulations of pottery at the site, while the second was
collected from an unknown location in the general vicinity of the site (Sheppard pers com).
These results have yet to be confirmed or quantified by thin-section petrography.
Remaining Use Life:
All pottery seen by the team was broken, although anecdotes of “whole” pots in deeper
water were heard. Two complete adzes and a number of fragments were recovered, and
an unground adze preform in the possession of a local landholder was photographed (See
Chapter 10). Comparison of the recovered whole adzes and the preform suggested that
both whole adzes found had been heavily re-sharpened, and were too short to have much
remaining use life in their intended function. Ovenstones were systematically collected at
Hoghoi. Many large fragments and some complete water-rounded cobbles were found,
enough to use for cooking, suggesting remaining use life existed for these. The presence
of these may have resulted from abandonment, or, more likely in view of the lack of adze
remaining use life, the presence of these at Hoghoi may simply indicate a lack of
economising behaviour, or perhaps discard of insufficiently tough stones liable to explode.
The stones may have had other, non-cooking functions such as canoe or net anchors.
304

Quantitative Evidence: Size/Density Effects and the Structure and Context of the
Deposit:
Sherd Size and Sherd Thickness (Thickness Strength):
To investigate the relationship between thickness and sherd size, while controlling for
fabric variation and form variation, body sherds of the most numerous temper class
(placered volcanic temper) were used. It was initially expected that strength, and thus area
of recovered sherds would show a positive correlation with thickness, which proved to be
the case (Figure 76).
Control for gross variation in form strength was achieved by selecting out only
those sherds without complex form, that is, body sherds. This means that many of the
larger sherds were omitted, due to representation of more complex forms likely to confer
sherd strength. Those too small to assign to vessel part were also omitted. Sherd area
seems to plateau above 9mm thickness, with the greater area of the 13mm and 15mm
categories possibly being a sample size effect. Thin sherds (less than 8mm thick) decline
in median size roughly proportional to thickness, and sherds less than 4mm thick are rare.
Sherds in the 4-8mm thickness range form the bulk of the total sample, while sherds in the
8-12mm thickness range yield consistently larger median sherd sizes. Thickness greater
than 12mm is rare, and seems to yield more variable sherd size, although this might be a
function of small sample size. These data seems to support the hypothesis that there may
be a bias against very thin wares in the recovered sample, which are liable to exist only as
small sherds fortuitously preserved rather than transported to the strandline and rolled to
destruction. Large thin sherds are either overly fragile and became broken into small, easily
transported thin sherds, or large thin sherds are more easily transported by water in the first
place, and thus tended to be destroyed by transport to and abrasion in the swash zone.
Temper Class and Sherd Size:
Temper classes are detailed in Chapter 4. Without controlling for thickness, but looking
at just body sherds, sherd size seems only weakly related to temper class (Figure 77),
except in the cases of Groups 4 and 7, where sample size was inadequate. Group 5
(Quartz-calcite) median size is slightly smaller.
305

Figure 76: Effect of sherd thickness on sherd strength
(controlled for temper variation by using placered volcanic
tempered sherds only).
Figure 77: Body sherd size for the various temper classes.
306

Body Sherd Size Frequency by Site:
The sherd skipping model predicts depletion in mid-sized (throw-sized,10-30cm 2 ) body
sherds, but utility of this model was limited by the potential for assemblage brokenness to
be equi-finally structured by other postdepositional factors. Analysis was limited to those
sites which retain a reasonable corpus in their collected assemblages of body sherds larger
than 30cm 2 . The dataset provided 1053 sherds comprising body part of the vessel only,
from eight sites, but insufficient larger sherds remained to test the sherd skipping model
using this method. Body sherds in the 20cm 2 size class predominate in the Miho site
(Figure 78), located at a modern village from where anecdotes of sherd skipping were
heard, and at Honiavasa, across a narrow channel from the same village, so either there is
something wrong with my theory of the correlates of this behaviour, or other processes
have structured the assemblages to a greater degree.
Differences between sites in these histograms are largely attributed to collection
intensity: for some sites sherds in the smallest size category (4cm 2 ) are numerous, while
absent in others, a pattern which dominates variability between assemblages. Low-intensity
collections like Honiavasa, Gharanga and Miho have few sherds in the 4mm size class,
while these predominate in the intensively collected sites Zangana, Paniavile and Hoghoi
This is clearly not the whole story though, as despite low collection intensity Honiavasa
yielded many more large (>56cm 2 ) sherds than either Hoghoi, Miho, Gharanga or Zangana.
Representation of Parts in the Site Samples:
The collector model predicts that assemblages previously collected by archaeologists or
other interested people will be depleted of conspicuous large or highly decorated sherds,
particularly those with information pertaining to upper vessel form (lip, rim, neck and
shoulder). Lip/Rim-only sherds and body-only sherds were counted by site. These counts
together with the ratios between them may provide an enumerator of previous collection
intensity, operating on the assumption that body sherds have a greater probability of being
ignored by a collector (Table 23). Only sherds greater than 15cm 2 were selected, to
control to some extent for assemblage brokenness.
307

Figure 78: Histogram of body sherd size classes by site.
308

Table 23: Ratio of lip-rim sherds to body sherds as a measure of collector effect.
SITE Total >15cm 2 Lip+Rim
count >15cm 2
309
Body count
>15cm 2
Paniavile 184 17 71 0.239
Hoghoi 155 31 66 0.47
Miho 205 39 58 0.672
Honiavasa 319 75 70 1.071
Gharanga 135 20 41 0.479
Nusa Roviana 40 5 9 0.555
Kopo 22 3 10 0.3
Zangana 222 37 56 0.661
ratio of LR:B
It seems the ratio of lip-rim sherds larger than 15cm 2 to body sherds larger than 10cm 2 may
well be a good measure of collector effect, as the most collected site (prior to the present
study) was Paniavile (previously yielding 84 rim sherds, 41 “decorated body sherds” and
an uncounted number of plain body sherds in an initial collection event reported by Reeve
1989:50).
What then of the other sites? Nusa Roviana and Kopo yielded small samples of
sherds, and their “collector” ratios are best ignored. Hoghoi and Gharanga both have ratios
nearing 0.5, which provide a weak suggestion that some of the more complete sherds may
have been removed from these in the past, prior to initial collection in the course of the
present study. Miho and Zangana both yield ratios approaching 0.7, suggesting this is a
reasonable figure for newly-discovered sites collected in a moderately intense fashion (the
bulk of these collections being obtained in single days by two and five collectors
respectively). Honiavasa, collected by a lone archaeologist on a rising tide over about four
hours, looks most like a first-hit, low intensity collection of large complex-shape sherds,
analogous, perhaps, to the original Reeve/Spriggs Paniavile surface collection in vessel

part representation. The major difference between the Paniavile ratio (0.24) and that of
the most similar sample (Hoghoi, 0.47), reassures to some extent that the major collections
other than Paniavile are not severely biased by previous unreported collection, and alerts
us to the degraded information content of the Paniavile sample as used in the current
research.
Frequency of Decoration:
This is potentially another way of measuring collector effects. While variation in
production in the past might account for different frequency of decoration in assemblages,
collector effects could create the same result, so this is something to be aware of in
analyzing surface collections. It might be simplest to avoid using frequency of decoration
in seriations etc., but as others do this (see Chapters 1&2) it was thought to be worth a
short detour to investigate this topic. For each vessel part, the number of plain instances
and the number of instances of decoration were counted. The ratio of undecorated to
decorated was calculated (Table 24).
For all vessel parts except shoulders the previously collected Paniavile site scored
within the range of the other sites. These data are interpreted as showing that the ratio of
decorated to undecorated sherds can remain stable through collection events. It should be
Table 24: Ratios of plain to decorated sherds, controlled by vessel part (lips, rims,
necks and shoulders.
SITE Plain L:Dec L Plain R:Dec R Plain N:Dec N Plain S:Dec S
1 (Paniavile) 20:59 0.34 110:36 3.1 98:31 3.2 56:26 2.2
2 (Hoghoi) 34:51 0.64 113:25 4.5 134:18 7.4 121:22 5.5
3 (Miho) 14:52 0.24 80:35 2.3 77:30 2.6 84:14 6.0
4 (Honiavasa) 23:80 0.29 154:17 9.1 153:20 7.7 125:23 5.4
5 (Gharanga) 10:24 0.42 36:15 2.4 62:12 5.7 49:21 2.3
8 (Zangana) 36:44 0.82 95:39 2.4 137:36 3.8 114:30 3.8
310

noted in this regard that initial surface collection of a full cover sample of sherds from
Zangana targeted only decorated sherds, with subsequent 1/5 and full cover collections
targeting all sherds and large sherds respectively. Zangana may thus be slightly enriched
decoratively through the initial collection strategy, but this effect is not large.
Other patterns worthy of note in Table 24 are that the Honiavasa site gains a marginal
score in two out of four vessel part categories (rim and neck), while lips are more likely
to be decorated than in other sites and rims and necks are more likely to be plain, showing
that overall ratio of decorated to plain sherds, without control of location on the vessel,
can be something of a lottery rather than a salient measure of assemblage decorative
variability. Miho, also, for example, has the most decorated lip, rim and neck scores, but
the plainest shoulders. Breakdowns of the specifics of decoration contributing to these
broad categories of data are given in detail in Chapter 9.
Another index of sherd skipping was considered, the ratio of neck sherds (not a
good skipping shape from personal experience) to body sherds, the latter of size range
>12cm 2 ,

epresented would be too sensitive to assemblage brokenness. This index is problematic
as it may relate to overall assemblage vessel form, and/or to collection bias rather than to
postdepositional sherd skipping. These data are presented in Table 25.
The weakest of these “Skipping indices” is for the Paniavile site, heavily collected
prior to the present study, suggesting again that previous collection may have been biased
against body sherds. The lowest (strongest) index was that of Honiavasa, despite a
tendency for sherds to break at the neck in this site, probably as a result of slab
construction. Zangana is not far separated, suggesting Zangana, at least, might have been
subject to selective removal of mid-sized body sherds, particularly since, the intensity of
collection at Zangana was quite high. An alternative, and more likely possibility is that the
high neck count at Zangana relates to an overall pattern in vessel form where the necked
vessel is more common in production, or more robust and more commonly preserved.
Miho is next strongest in this index, and as Miho is located at the anecdotal skipping
village, and Honiavasa within a short distance, there may be something to it.
Sea Level History:
Radiocarbon ages of uplifted corals, combined with an assumed high stand of +1m at
5500BP generate mean uplift rates for the New Georgia tectonic block of between
0.2mm/yr and 0.9mm/yr, with the three data locations in Roviana Lagoon area suggesting
mean tectonic uplift rates of 0.7mm per year (Rereghana Island and Araroso passage) to
0.9mm per year (Nusa Roviana Island) (Mann et al. 1998: Figure 8, Table 1, Table 2.). If
sea levels have been falling an average of 0.2mm per year since the late-Holocene high
stand, and the land has being going up at an average 0.7 to 0.9 mm per year in the Roviana
area, then the combined effect over the 2,800 years may be somewhere in the vicinity of
an effective fall in sea level of between 2.5 and 3.0 metres. It is unlikely that anything in
312

nature happens this smoothly, but there is currently no local data for the period between
about 6000BP and 200BP, leaving little alternative but to extrapolate. Data from Ruvi Bay
on Kolombangara is reassuring in this respect, as coral dated there to 2250BP ±60 is one
metre above sea level, and as uplift rates there seem substantially lower (0.1 mm/yr,
calculated from a total Holocene emergence of 1.5m) suggesting that if a correction is
applied for the greater uplift at Roviana Lagoon of 0.6-0.8mm per year, we could add
1.35-1.8 metres on to this (if the average uplift rates are useful) and that Roviana sea level
(relative) might have been about 2.35-2.8m higher up the shoreline at around 2250BP.
More data from the Roviana area is needed before this figure can be accepted with any
certainty. All paleo-shoreline evidence, however, points to an emergent coastline, with the
artefact scatters in deeper water (possibly buried in soft lagoonal sediments) until fairly
recently. This is consistent with cultural formation processes for the artefact scatters as
discard from anchored vessels, fishing platforms, or more substantial stilt houses. This
conclusion would need to be revised, though, should the gradual sea level model used to
extrapolate between insufficient data points turn out to be incorrect.
Chapter Summary and Conclusions:
This chapter was written from the perspective suggested by the results of the analysis in
Chapter 5 that most of the sherdage discarded in the past has not made it into our sample,
and of the missing sherds, most are no longer extant. It has not been possible to provide
a definitive account of where the missing pottery went, but a brief synthesis of the
analytical results in hand is given below, to clarify which of the models set up at the outset
of this chapter are consistent with the properties of the collections.
Cultural Formation Processes:
The absence of adjacent settlement evidence on land for all intertidal sites argues against
313

models of terrestrial settlement. The good state of preservation of Acropora corals on the
adjacent shore platform, and the evidence from sea level studies (Mann et al. 1998)
suggest that this feature is an emerged reef flat which would have been the upper limit of
coral growth at the time the sites were formed. Any ceramic deposits on this flat would be
highly visible, and would have been deposited into the sea in the Lapita period. These
factors strongly support the conclusion of a maritime rather than terrestrial origin.
Additional evidence against terrestrial models is the complete absence of
weathering of carbonate grains, which sometimes occurs in terrestrial deposits in the
tropical Pacific. Reefal carbonate detritus used as tempering material is invariably in
pristine condition except where slight sintering has occurred during firing.
Accretions provide an additional suggestion of a high stand of sea level at-or-post-
deposition. Lack of salt drying damage in the deeper margins of sites, and severe damage
of this sort to many sherds post-recovery, suggest that gradual submergence of terrestrial
sites is unlikely to account for the presence of the scatters in the sea.
Some sites could conceivably have been located on piled coral wharves or artificial
islets, now eroded or otherwise reworked. The shoreline and intertidal at Hoghoi has many
coral blocks of the size used in this sort of structure today, but these are common along
extensive stretches of the raised-reef barrier island where Hoghoi is located, even where
there is no pottery. Better-preserved buried sites from the same general period in the
Bismarck archipelago (Gosden 1989, 1991b, Gosden et al. 1989, Gosden & Webb 1994,
Kirch 2001), seemed to have post structures, and there is no strong evidence yet for any
different pattern at Roviana Lagoon.
Postdepositional processes are likely to have removed any ceramic evidence for site
abandonment. While there was some remaining use life in “ovenstones” collected from the
Hoghoi site, the two complete adzes recovered were heavily re-sharpened if the
proportions of an unused reject preform in a local collection were anything to go by. The
ovenstones could indicate lack of economizing behaviour rather than abandonment.
314

Post-deposition Taphonomic/Formation Processes:
In the disaggregation model outlined above, it was predicted that sherd strength variability
would be low. Analysis of sherd size in relation to temper suggests that strength variability
between temper types as reflected by sherd size is low for the assemblages. Whether this
is due to congruence in technical competence using differing materials, or whether the
weaker sherds have already gone is uncertain. Favouring the latter view was the presence
among the stronger sherds of infrequent soft sherds or (very rarely) partially dissolved
sherds with solution scarring, suggesting that disaggregation has been a factor in sherd
attrition. Additional support for a disaggregation model was found when sediments from
the Hoghoi site were examined microscopically, and found on examination under low-
power magnification to contain abundant volcanic mineral grains exotic to the calcite
raised coral barrier island local geology. These grains are thought to be temper sands
deriving from disaggregated pottery, but this conclusion has yet to be confirmed
petrographically.
The data on sherd thickness indicates that thinner sherds are weaker (break more
easily) which makes the surviving fragments more easily transported by wave action, and
possibly also that large thin sherds have been transported to the strandline and destroyed
(weak fabrics and large thin shapes are selected against by a transport model). This model
was consistent with the frequency distribution by thickness in which small thin sherds were
rare. Large, thick sherds were rare also, which could relate to behavioural factors such as
use life, but may also have resulted from inadequate firing of thicker sherds, and thus less
resistance to breakage/transport and to disaggregation.
There is some support in the part representation data for the collector model and
the collection intensity model. The collector model is supported by the ratio of lip/rim
sherds to body sherds, which showed Paniavile to have been more heavily collected in the
past than other sites (corresponding with events reported in Reeve 1989). Body sherd size
histograms showed correlation with collection intensity for assemblages in the present
study.
315

Sherd skipping behaviour has not left any obvious hole in histograms of body sherd
size-frequency, although there was some support for the model in the ratio of body sherds
to neck sherds, but underwater archaeology might be a better way to ascertain the extent
of recent sherd skipping or sherd throwing into deeper water.
Interpretation:
It seems likely that different processes have predominated at different times. During the
period of occupation, stilt houses would have been constructed in water possibly too deep
to stand in at low tide, with sharpened posts driven into lagoonal mud until they reached
the coral underneath. At this stage there was probably a zone of shallow-water Acropora
corals to landward of some sites, backed by a low cliffed shoreline (Hoghoi, Miho,
Zangana, and Paniavile at least, although this is not immediately apparent at Zangana today
due to landscape alterations during WWII, and possibly Nusa Roviana also), making canoe
landing and access difficult. Early in the history of stilt settlements, the pottery is likely to
have been semi-buried in soft lagoon sediments, with disaggregation due to dissolution,
and turbation by plants, grazing animals and burrowing animals and
construction/maintenance being major factors contributing large quantities of exotic sands
to the local sediments. Falling sea levels post-abandonment are likely to have brought the
ceramics into the lower margins of the swash zone fairly recently, removing organics and
fine sediments to a varying extent depending on the site location, and leaving lag deposits
of stronger sherds. Wave transport and anthropogenic processes such as sherd skipping
may have caused gradual or intermittent sherd attrition to a greater extent than other
mechanisms in more recent years. It seems from the Rereghana oyster evidence presented
by Mann et al. that the current sea levels are very recent, and have been at their present
level no more than 400 years, so these latter processes are unlikely to have affected the
sites over the millennia. In the modern era, particularly over the last three decades, interest
in the sites by archaeologists and others has had a major impact on one site (Paniavile),
selectively removing large, decorated, and rim sherds initially, with more intensive
316

collection of a number of sites during the present research.
Some of the most significant taphonomic biases may have occurred during the early
disaggregation phase of sherd attrition, where particular functional variants or production
groups may have disappeared almost entirely, due to low firing or incomplete firing of the
vessel wall interior. These types of biases would have to be taken into account in
comparing intertidal sites with terrestrial sites, as overall effect on assemblage composition
could be large. Evidence will be presented elsewhere (Chapters 8 and 9) that at least one
production style had a fragile tall, thin excurvate lip, and that as a result the style is known
largely through rim/neck/shoulder sherds rather than lip/rim/neck/shoulder sherds.
Transport processes are likely to have biassed preservation towards chunkier types which
broke into large pieces, like Honiavasa carinated sherds, and biases between intertidal sites
can be expected due to slight differences in wave exposure.
Of the taphonomic models constructed in the introduction, it is the collector and
collection intensity models that are most clearly supported by the size data and the part
representation data in this chapter. Collector processes seem to have impacted severely on
the information content of sites, leaving the Paniavile sample stylistically depauperate in
information on vessel form and decoration.
A methodological conclusion arising from this result is that reconnaissance survey
and recording of surface intertidal sites in the future could be formulated in the light of
these findings, i.e. that the effects of casual unprovenanced surface collections on the
information content of intertidal surface scatters is drastic. Where possible, in-situ non-
destructive recording is advocated unless detailed spatial recording (point provenancing
of all items, including accurate heights in relation to a datum) and field materials
conservation can be carried out. One potential problem with in-situ recording is accretions
of marine organisms, which occasionally can hide finer decoration until cleaned in the
laboratory. Partial cleaning in the field might be sufficient. Also, temper studies would
require that snapped samples of some or all sherds would need to be collected for analysis.
317

318

CHAPTER 8:
VESSEL FORM VARIABILITY AND VESSEL
Introduction:
FUNCTION
The moderate degree of vessel brokenness of the Roviana assemblages (Chapter 5), in
which many sherds are identifiable to one or more vessel parts, permits an analysis of
assemblage vessel form characteristics, which was carried out using a metric vessel contour
approach (introduced in Chapter 1). The classification in this chapter is materialist in that
it is a measurement device created by the analyst from observed variability, for the
purposes of controlling to some extent for functional variation in the seriation analysis in
Chapter 12.
What are the sources of form variability? Are there constraints imposed by the
material and method of construction? What aspects of observed form variability are related
to vessel function? Is temporal variation in form constrained or directed by function, or
might it fluctuate in response to some natural or social environmental condition? Which
aspects of form variability might be seen as stylistic, or free to drift in a stochastic manner
(Dunnell 1978, 2001)? This last question is key for the chronological aims of the analysis,
while the other questions are the reverse side of the same coin. By seeking general-purpose
run-of the mill vessel forms for use in seriation, some gross functional variation is removed
from the mix, reducing the possibility of incorrectly seriating functional variation.
Some characteristics of Roviana vessel forms are inferred from negative evidence.
For example, the lack of any flat base sherds will be discussed and interpreted in this
319

chapter. The main positive evidence for vessel forms is upper vessel contour variation,
especially the measurements of rim angle, rim depth, rim Vcurve, neck angle, neck Vcurve,
shoulder depth/Vcurve and carination angle/Vcurve (see Chapter 4 for definitions). Body
sherds provided an additional line of evidence which was largely experimental, but which
showed up some interesting differences between site assemblages. The precision with
which these body form data are replicable has not been assessed, and there are difficulties
in associating body forms with rim/neck/shoulder forms in assemblages comprising a
mixture of production styles, so primary emphasis is laid on upper vessel contour for
classification of vessel forms.
In accordance with the theoretical discussion in Chapter 1, functional classes are
kept broad, and multi-functional or general purpose forms (e.g. cooking/serving/liquid
transport) are expected to dominate assemblages, on the basis of use-life theory.
Subordinate (long-use-life) functional classes, for example large storage jars or ritual
vessels, were also expected to be present in lower frequencies in the assemblages.
Substantial form variation was expected within such broad functional classes.
Sample size comes to the fore when dealing with long-use life or low-frequency
functional classes, and there is no attempt here to characterize these in any detail. Analysis
focuses on classification of and variability within the largest general purpose functional
class. Comparison of site sample characteristics is also undertaken, using a measure of
sherd restriction factor, the ratio between total sherd area for a site, and rim EVE for
restricted vessels. Covariation between thickness, vessel size and vessel form is examined,
as it relates to the functional structure of assemblages and the characterization of vessel
function. Sherd strength was not tested, but some qualitative remarks on this subject
relating to assemblage functional structure are included.
Methods for Inferring Vessel Form from Sherds:
Rim forms alone were not regarded as a reliable guide to vessel function (this is a
320

departure from the analyses of others in the general region, e.g. Summerhayes and Wickler,
who in some instances regard rim forms as proxies for vessel forms (see Chapter 2). If a
fundamental distinction is made between jars and bowls, it is essential that only those lip
sherds which are sufficiently large to discriminate between these categories of vessel
contour are taken as evidence. For very broken assemblages, neck sherds by contrast can
be used to demonstrate the presence of restricted jars, as this contour is diagnostic of
restriction if the sherd is large enough to orient using the neck horizon. Similarly,
carination sherds and sherds from the junction between flat bases and vessel sides can be
used to infer carinated and flat-based vessel forms respectively. Neck/shoulder sherds can
be used also to infer generally carinated form, although the form of the carination itself
cannot be specified. One would expect to find carination sherds in the assemblage too in
such a case. Small lip sherds cannot be inferred to represent bowls, unless independent,
convincing evidence is presented to demonstrate a correlation between the particular
morphology of lips and overall vessel morphology.
Figure 79: Initial classification of vessel forms.
321

Function and Temper:
Much of the Roviana pottery is relatively strong, compared with some samples of
terrestrial origin (For example Reef-Santa-Cruz materials or modern Mailu pottery in
storage at Auckland University). This is reflected in sherd size in some sites. A lot of the
tall thick everted rims from Honiavasa for example are usually very strong, highly
tempered, difficult to snap with pliers, and are quite elastic (can be dropped on the
linoleum-covered concrete of the laboratory and will bounce). This seems to be evidence
for either unusually strong manufacture for some production units, or preservational bias
towards strong fabric (see Chapter 7). If the former were the case this would tend to
support an understanding of the assemblages as predominantly cooking and/or general
purpose vessels although this may relate to drying technology during pre-firing stages of
manufacture (drying in a torrid-zone tropical environment may require rapid fire-drying or
drying in direct sunlight, necessitating increased temper concentration to reduce shrinkage-
cracking (see Chapter 1). Alternately, terrestrial deposition might lead to a substantial
weakening of fabric through chemical weathering, an effect not present in the sea, or
perhaps immersion strengthens some sherds. Many sherds from the submerged Mulifanua
site in Samoa were,
“...dense and largely unaffected by age or wear. Under the petrographic
microscope it was apparent that zeolites had filled in some pores. This is
a phenomenon caused by sea water immersion....(Petchey 1995:57).”
Some white mineral veins in some of the Roviana pottery may be a similar phenomenon.
Form: Initial Nominal Classification:
The following classifications of vessel form are combinations of the major nominal form
attributes as described in Chapter 4. These are preliminary classifications that are
recombined into functional classes following a metric analysis of form variability within
322

classes. Some are mutually exclusive, others (e.g. Form 1 and Form 6a) are
complementary, in that similar vessel forms are being inferred from independent lines of
evidence. Some of the initial descriptive categories turn out, on further analysis of
variability later in the chapter, to be quite arbitrary divisions of a continuum of variation,
and are lumped in a functional classification as a consequence.
Form 1: Conical-to-Hyperboloid Shoulder Form and Carination:
Hyperboloid form is similar to conical form but concave in vertical profile (Rice 1987:219)
with a corner point (Rice 1987: 218) or carination. Classification criteria in the table of
form data (see Chapter 4 and attached CD-ROM table “Form.db”) were “carination type”
with a value “H”and “shoulder type” with a value “L”. These forms were mainly found in
the Honiavasa site (see also Figure 7, Figure 8, Figure 9 ). They are interpreted as
representing carinated restricted jars with a conical/hyperboloid shoulder, everted rims and
round bases (round bases by negative evidence:- absence of flat bases in assemblages)
(Figure 79). One sherd from Hoghoi had this shoulder form, and was probably broken at
the carination (HG.99.061). There was also one hard carination sherd in the Punala sample,
which has not been analysed in detail. Most, though, were from Honiavasa.
Form 2: Unrestricted /Weakly Restricted Rimmed Jars or Deep Bowls:
This class was subdivided as follows:
Form 2a:
Form 2a (Figure 79, Figure 80, Figure 81) is an unrestricted jar or bowl form with a
slightly excurvate rim and a thickening of the upper body giving the external appearance
of a carination, without any obvious interior join, probably originating through a two-piece
construction method where the excurvate rim is joined to a rounded body/base.
(Classification criteria are “rim type” has value “K”, “shoulder type” does not have value
“L”). This is a small group, defined mainly on their construction method and usually
323

weakly restricted or unrestricted form, not easily differentiated from compound rimmed
vessels. Because there were only a few of these vessels, and because decoration was
confined to fingernail impression or excision of the outer rim (an attribute omitted from
seriations), these sherds had no effect on the seriations.
Form 2b:
Form 2b (Figure 79, Figure 80) is represented by a single large quartz-calcite tempered
sherd, and is thought to indicate a shallow unrestricted ellipsoidal base topped by a corner
point (hard carination) and a narrow conical restriction. At the top of the conical restriction
there is another corner point in the profile, with a tall excurvate (hyperboloid) attached rim.
Apart from the distinct corner points and the shallow base/body or lower rim, this sherd
bears some resemblance to Form 2a, but was given a separate class as the exotic quartz-
calcite temper in its fabric suggests a different manufacturing centre to the Form 2a sherds
(suggesting the differences between this sherd and the rest of Form 2a may turn out to be
more than just “noise” despite occurrence in the sample as a single sherd). Classification
criteria are “rim type” with a value of “K” and shoulder type “L”.
This unrestricted vessel form is sufficiently deep, rounded and well-tempered to
have had similar functions to more restricted vessel forms, but equally, due to a shallow
base and unrestricted access to the contents, could have fulfilled a variety of serving, non-
fluid cooking, drying, washing and other functions, and would have been a fairly stable
form not liable to complete capsize due to the shallowness of the base. Because it was
undecorated, it was automatically excluded from all seriation analyses.
Form 2c:
These were everted-rim to vertical-rim unrestricted jars or weakly restricted jars defined
reasonably securely from rim/neck attributes (especially neck angle), and forming the least-
restricted end of a continuum of rim/neck/shoulder variation (Figure 79, Figure 80),
which included everted rim restricted jars to straight or incurvate restricted/unrestricted
jars. Classification criteria were “rim type” with a value of “X”, “S” or “V” and “neck
type” with a value of “T”.
324

Figure 80: Examples of unrestricted Form 2 variants: Form 2a (top); Form 2b (upper
middle); Form 2c (lower middle); and Form 2d (bottom).
325

Figure 81: Additional Form 2a Sherds (top 6); the two decorated gambrelled vessels
from Nusa Roviana are a short-rim variant of Form 2a; the lower sherd is transitional
between Form 2a and Form 2d, being unrestricted.
326

These sherds do not display any morphological evidence of multi piece construction, and
are inferred to be modeled from a single lump of clay by paddle and anvil techniques (or
perhaps paddle and finger in some cases). Anvil impressions are common on the vessel
interior below the neck, but are usually absent from the interior of the rim, presumably as
a result of wiping of the rim or forming with a large anvil/tool of some sort.
Form 2d:
This class comprised unrestricted bowls and/or jars (Figure 79, Figure 80, Figure 81).
There were few examples, classified together by absence of any suggestion of a neck,
together with excurvate rim inflecting into a globular body. There are only two examples
within this category, from two sites (Hg.10.103 and HV.5.205). Classificatory criteria in
“Form.db” are: “rim type” has a value of “X” and “neck type” has a value of “N”.
Form 2e:
This class comprised a single sherd, suggestive of an unrestricted jar or deep bowl, with
a flat thickened lip and punctate/fingernail impressed decoration (sherdA32, see images
appended on CDROM). Classification criteria were “lip type” value “L” and “shoulder
type” value “T”.
Form 3: Inverted Rims:
Form 3 vessels (Figure 79, Figure 82) were thought to be small, inverted (direct) rimmed
vessels (classification as direct is tentative: there is a possibility that these are the heavily
flared everted rims of small-orifice restricted vessels such as the water jar illustrated in
Ross (Ross 1996:78, 2b water jar). Classification criteria were:
• the lip was definitely represented on the sherd (to avoid confusion with rounded
shoulders of everted-rimmed necked pots);
• rim depth was sufficient to suggest absence of a neck;
327

• rim form was inverted “rim type” had a value “T” in “Form.db”;
• no carination was recorded (“carination type” had a value of “9"), to differentiate
from one rim of this form which had a carination near the lip.
Judging by the horizontal curvature of the rim, some of these sherds were from tiny
vessels, about the size of an egg-cup. Vessel B.6/B.7 from Zangana was inferred to have
had loop handles (Figure 83), the only example of this handle form, suggestive of
suspension during use, and conceivably having a specialized cooking/heating/drying
function at some height above a fire which would have made support on three stones
impractical. Given the uncertainties surrounding the overall vessel forms and sizes and
functional implications of Form 3 these were omitted from seriations.
Form 4: Shallow Carinated Bowls:
There is one example of this form, in an exotic quartz-calcite fabric ( Figure 79,Figure
83).
Form 5: Reworked Body Sherd:
The single instance of Form 5 (Figure 79, Figure 83) was a large shallow spherical sherd
with a circular lip of unusually high lip angle, with small irregularities of the lip suggesting
possible reworking of a spheroidal body sherd. The sherd was found at Paniavile, and may
be a potting stand or mold of some sort, or alternatively fits a description of frying pan
sherds attested ethnographically in Oceania (Ross 1996:70). As this sherd was undecorated
it had no effect on seriations.
Form 6: Everted-rim Restricted Vessels:
These can be inferred with confidence from the numerous neck sherds indicating this form.
In many cases, but not all, neck sherds can be approximately oriented by latitudinal
decoration and corner point horizon to ascertain whether the vessel is clearly restricted
328

Figure 82: Form 3 inverted restricted vessels.
329

Figure 83: Form 3 or Form 4 (top left) and another example of a short-rim variant of
Form 2a (top right); the sole example of Form 3 with loop handle(s) (2 nd to top); the
only confirmed Form 4 carinated bowl, in exotic temper (2 nd to bottom) and a large
base sherd or frying pan, Form 5 (bottom).
330

(Form 6) or unrestricted/weakly restricted (Form 2c). An analysis of metric variability
within this large group of sherds was undertaken to investigate functional variation.
Metric Variability Within Form 6:
If orifice diameter is related to function, and fluid storage or fluid transport vessels are
expected to have had small orifices, and cooking vessels larger orifices, this might produce
a bimodal or multimodal distribution of orifice size (for which neck “Hcurve” measurement
is used as a proxy here). Neck Hcurve for all such vessels taken together, using size
intervals of 10mm, looked approximately normally distributed, but with a hint of multimodality
in the tails of the distribution (Figure 84).
Neck Hcurve seems largely constrained between 95mm and 155mm, although this
could be a sample size or measurement effect. This makes some sense from the point of
view of function, since Hcurve of 95 mm is about the smallest that will admit a hand
(remembering that Hcurve is measured external to the neck so neck thickness must be
subtracted from Hcurve to yield orifice radius). General purpose vessels that require
abrasive cleaning need to be large enough to admit a hand (this is the difference between
pots or jars on the one hand and flasks or bottles on the other).
To test whether this aspect of the distribution was perhaps being obscured by
measurement error on small sherds, either through curve irregularity or taphonomic
rounding of sherd edges, the same data were run with cases excluded where sherd area was
less than 25cm2 . In this bar chart (Figure 85) sample size was much reduced, which
Count
30
20
10
0
100.00 200.00 300.00
EXTERNAL NECK RADIUS
331
Bars show counts
Figure 84: External neck radius, all sites, size
intervals 10mm.

can be expected to introduce a greater level of jaggedness to the pattern. The pattern was
clearer rather than obscured, suggesting that either measurement error was a factor on
small sherds, or that the smaller vessel size class was represented principally by small
sherds. Although at least one vessel style within Form 6 is represented predominantly by
small sherds, as will be discussed below, this was not thought to be the explanation for the
discontinuity, as neck Hcurve was generally larger than 96mm radius for that group.
The relatively small number of necks measuring outside these apparent constraints
were of interest. Were those necks of less than 95mm Hcurve perhaps functional variants;
or evidence of multi-part construction; evidence that some potters (perhaps children) had
very small hands (there is a photograph of a mother and daughter making pottery in Specht
& Fields 1984 which illustrates this possibility):17); or were these merely irregularities in
curvature, unrepresentative of average vessel neck radius? Were necks measuring larger
than 150mm radius (the upper tail of the frequency distributions) perhaps other functional
variants; unrepresentative irregularities in curvature; or a large variant specific to a
particular period or production unit?
Count
20
15
10
5
0
100.00 200.00 300.00
EXTERNAL NECK RADIUS
332
Bars show counts
Figure 85: All sites, sherds >25cm 2 , neck Hcurve intervals
10mm.

Form 6a:
The 95mm lower limit of external radius of the bulk of necks approaches the minimum
needed for the hand of the potter to access the vessel. Only one excurvate rim sherd
provides secure evidence for an external neck horizontal radius below 95mm (Figure 86),
with a internal orifice radius measured as 47mm. This sherd EVE measured 25%, making
irregularity of manufacture an unlikely explanation for the anomalous neck size. This sherd
was from the Honiavasa site, where two-piece or multi-piece construction is evidenced by
a number of slab-built carination sherds (Form 1). No neck exteriors are measurable for
the Honiavasa carinated sherds, but measurement from drawn profiles for which the
location of the CVA had been established from measurement elsewhere on the sherd
suggest the following approximate data: 80mm, 90mm, 100mm and a single example of
less than 70mm. Together with the neck measurement of HV.04.133, these data provide
two examples of orifices too small to get a large hand through, indicating that at least some
vessels of this form were unlikely to have been used in any function requiring direct
abrasive cleaning. These two necks were interpreted as evidence for an everted-rim
carinated vessel class with a probable use of fluid transport/storage, poorly represented in
the Honiavasa assemblage, and not securely evidenced in samples from other sites
(remembering there were single sherds with hard carinations at Hoghoi, Paniavile and
Punala).
Whether all Honiavasa vessels represented by carination sherds were so used is
unknown. The effect of this vessel form on seriations is discussed in Chapter 12. Form 6a
and Form 1 are regarded as possibly functionally equivalent, although the larger orifice for
some Form 1 vessels could confer utility as multi purpose vessels including cooking
functions.
Form 6b:
Of the 32 sherds which measured above 160mm external neck radius, some
333

Figure 86: Form 6 “Neck Hcurve” variation; the two top sherds are too small to get a
hand into (Form 6a), while the two lower sherds have head-sized or larger orifices
(Form 6b).
334

seemed inexplicable by curve irregularity alone but larger sherds in this category confirm
the presence of a larger, less frequently discarded or preserved Form 6 vessels. Due to low
numbers these do not affect the outcome of seriations in Chapter 12. Counts of vessel
forms are given in Table 26.
Form 6c:
The general situation regarding the neck radius of everted rim restricted vessels, taking all
sites together, is that the vast majority of reliable Hcurve observations fall between 95mm
and 155mm radius. This range represents substantial variation in the physical size of vessel
orifice, from adult hand-sized to adult head-sized, and might encompass virtually any
function one might think of for a vessel of this general form, from wet/dry storage or
fermentation (possibly with a tied cover) to boiling/simmering to feed a variety of group
sizes, serving, drying/roasting (e.g. Canarium processing), ritual consumption or offering,
soaking, washing, transport of wet or dry materials or food processing (e.g. collection of
grated foodstuffs, settlement of starch grains, expression of coconut cream, processing of
cream into oil). The list is limited more by imagination than by vessel form. The necks tell
us that these are pots rather than flasks, bowls or platters, and that these occur in a variety
of neck sizes, within quite broad but strict size limits predominantly.
There is substantial form variability and decorative variability within the numerically
large Form 6c (Table 26), and undoubtedly (on the evidence of rim form and evidence
from body and base sherds presented below) functional variation also. There seems little
basis for separating Form 2c from Form 6c for the purposes of seriation, as assessment of
the degree of restriction relied on measurement of sherd orientation, which was inaccurate.
Moreover the amount of restriction seemed to form a continuum, and therefore
classification into Form 2c and Form 6c is arbitrary. Forms 2c and 6c combined seemed
to be the best candidate of all the vessel forms on which to perform form-controlled
seriations, due to large sample size, presence in all sites and decorative variability.
335

Table 26: Counts of vessel form classes.
Form Paniavile Hoghoi Miho Honiavasa Gharanga Nusa
Roviana
Not
Classified
336
Kopo Zangana Total
517 708 273 264 199 91 17 677 2746
Form 1 1 1 9 1 12
Form 2a 4 1 1 2 2 10
Form 2b 1 1
Form 2c 5 6 3 1 2 3 2 22
Form 2d 1 1 2
Form 2e 1 1
Form 2f 1 1 2
Form 3 2 1 1 6 10
Form 4 1 1
Form 5 1 1
Form 6a 1 1
Form 6b 1 1
Form 6c 116 142 103 165 73 21 5 170 801
Everted Rim Form Variation:
Seventy-seven examples of everted rims from restricted vessels were complete in profile
from lip to neck, for which rim angle and rim depth were measurable. Neck angle and neck
Vcurve were measurable for 60 of these sherds, and rim Vcurve for 54. These data
permitted a rough analysis of rim and neck form variability within this class (Forms 6 and
2c).
Rim eversion angle and rim depth for Honiavasa and Hoghoi sites illustrate the
extremes of assemblage difference in rim form for the everted-rim restricted vessel class
(Figure 87). Multiple models can be constructed to explain this difference: there may be
two or more contemporaneous functional classes of vessel represented by groups in these
data, or there may have been a gradual shift in production style over time, with overlap in
occupation span accounting for Hoghoi rims being made in Honiavasa forms. There

Figure 87: Relationship between rim angle and rim height for two sites, all
measurements shown.
may have been a sudden shift in trade-exchange patterning during the occupation span of
Hoghoi, after occupation ceased at Honiavasa, that brought a new style of pottery to the
area or perhaps there was a shift in economy, encouraging the adoption of a new rim form;
or alternatively a change in pottery diversity over time, where Hoghoi and Honiavasa
occupations are non-contemporaneous but Honiavasa had two functional classes of pottery
while Hoghoi had three.
Measurement error (particularly of rim angle, given the difficulty of rim orientation
measurement) is unlikely to explain the separation between these two sites, but might tend
to blur the overall relationship between rim depth and rim angle in all assemblages. Using
only those cases where EVE was greater than 9%, for which rim angle measurement error
can be expected to be smaller, the plot of rim depth against rim angle for all sites was more
expressive of a negative correlation (Figure 88).
337

Figure 88: Rim angle and rim depth, all sites, filtered so that EVE is greater than
9%, to reduce the effects of measurement error.
This clarification of the relationship between rim depth and rim angle suggests that
measurement error was creating some of the groupings in the data in Figure 87 (the
greater the EVE, the more confidence we can have that the sherd is correctly oriented, on
which the rim angle measurement depends). Re-examining the relationship between
Honiavasa and Hoghoi rim form using EVE>9% as a filter retains a separation of
Honiavasa and Hoghoi rim forms, but with more linearity. It seems likely that there are
technological or functional constraints on rim form, which might explain the linearity in
these data. Tall rims tend to be less everted. This may be due to technical difficulty in
forming a heavily-everted tall rim from a single lump of clay using paddle and anvil
technique, or alternately, may reflect a continuum of intended-use variation. These factors
may combine in a temporal phyletic series, placing Honiavasa and Hoghoi at opposing ends
of a temporal continuum, or may not.
Outliers above this negative correlation tendency are probably slab-constructed
rims, formed from an arc of clay cut from a flat slab, and the most extreme outlier of this
338

sort is from the Honiavasa site. There is other evidence for slab construction in that site,
where Form 1 shoulders are most common, and where many tall everted rims are snapped
from the vessel in the neck region, suggesting a weak join at the neck.
The results of this analysis suggest that rim form variation of everted rim restricted
jars may be constrained both by manufacturing processes and/or by functional variation,
and that the effect of this variation on seriations should be assessed or controlled for.
Inferring Vessel Form from Body Sherds:
Curve fitting using brass template curves allowed accurate estimation of the evenness of
curves, and the radii of even curves. Errors of parallax that arise when using curvature
charts were eliminated.
The observed arrangement of interior anvil marks in latitudinal circles on larger
sherds which could be oriented with confidence allowed approximate orientation of many
less complete body-only sherds to be inferred. Using this method it was possible to
measure the convexity of the exterior vertical profile (Vcurve), and to take a second
measurement normal to this curve (Hcurve). This was not necessarily a measure of
latitudinal curvature, and in most cases is better likened to a “great circle” curve in
navigation, a curve marking the intersection of the earth’s surface and a plane passing
through the earth’s centre. This curve was measured by choosing the tightest curve that
would sit conformably on the sherd surface at right angles to the Vcurve measurement line.
Curves which were uneven were noted but not entered as Vcurve or Hcurve.
Body sherds for which anvil marks were absent or not patterned, and for which
orientation could thus not be inferred, were measured in the dimension of their maximum
and minimum curvatures. Where these two measurements differed they were noted in the
“Notes” field of the “Form.db” table. Where they were equal (indicative of a convex
spheroidal sherd surface) they were entered as Vcurve and Hcurve measurements. There
is thus a higher probability of measurement for spheroidal sherds than for cone/can forms,
which is significant for the plots presented below. While all body sherds were measured
in this manner, a minimum size filter of 25cm 2 was applied to the data analyzed, to reduce
the possibility of sampling error due to an unknown level of curve evenness. These
339

methods are similar in overall concept to the two curvature method (Hagstrum &
Hildebrand 1990). Results for all sites combined are presented in Figure 89 (Honiavasa
is shown using a symbol different to the other site to show up a difference in the patterning
of observations for that site.
Figure 89:Body sherd curvature measurements, showing spheroidal sherds
(diagonal alignment) and other canister/conical forms (e.g. s-shaped pattern
of Honiavasa sherd measurements).
The truncation of Vcurve data in Figure 89 at 300mm radius represents the upper limit of
the curve template set, and the difference between a radius of 300mm and a straight line
is probably within the measurement error of the technique, given the sherd sizes under
study.
The other obvious patterning is concentration of the data along the line of
sphericity, where one curve fits the sherd in both dimensions of measurement. Most of
these measurements probably represent globular (spheroidal) bodied vessels of up to
200mm body radius. Spheroidal sherds smaller than 100mm radius may simply be the
340

smallest examples of spheroidal bodies, or may be the tightly curved bases of vessel bodies
that are semi-conical in shape, approximating an upside-down parabola in external base
profile (see Figure 35). Sherds at the upper end of the line of sphericity are fairly gently-
curving, near-flat, suggesting flattish base sherds from squat globular vessels or carinated
vessels with a flattish rounded base and built-up sides.
Sherds falling substantially below or above the spheroidal group are either the
flatter side portions of conical lower bodies, or tightly curving (in vertical profile) but
horizontally moderate soft shoulders or soft carinations.
By site, samples show some variation when plotted in this manner (Figure 90).
Most sites show a cluster of spheroidal sherds above 100mm radius, extending to
approximately 150-200mm. The spread in this cluster for Paniavile is probably
measurement bias in the first analyzed sample, created by choosing the 150mm curve
before others, and tending to accept this reading. This tendency was actively resisted by
more careful curve fitting in other site assemblages. Hoghoi, Miho and Honiavasa have
what are potentially pointy bases or small (less than 100mm radius) globular bodies. Miho
and Zangana had outliers of low curvature.
The Honiavasa sample showed a tighter clustering of spheroidal sherds than other
samples, between 100 and 150mm radius, suggesting smaller-bodied globular vessels were
the norm here. This sample also had a number of tight Vcurve measurements of around
50mm associated with Hcurves of about 120mm suggesting a soft carination body form,
which did not occur in other sites among the body-only sherd samples (there was a soft
carination in the Hoghoi site sample).
341

Figure 90: Comparison of body sherd curvature measurements by site.
342

Gharanga clearly has the lowest incidence of sphericity, which is potentially significant for
vessel function. A comment by the landowner that these sherds were probably indicative
of pots broken during water collection from the Gharanga stream provides an additional
hint that perhaps there could be some functional information in this plot. If water jars
tended to be wall-sided in body shape, or biconical (unrestricted spheroid base, restricted
cone or restricted hyperboloid shoulder) in shape, and the Gharanga site was a specialized
water collecting site, one would expect this patterning in the data. There are strong
decorative similarities between Hoghoi and Gharanga, and yet marked dissimilarity in this
plot, which may be functional difference. It should be noted that vertical-sided jars would
produce restricted tall ovaloid or restricted tall elipsoid body sides with low Vcurve (Rice
1987:219), but there should also be a minority of spherical sherd forms in an assemblage
resulting from the breakage of bases of these vessel forms, which is the pattern seen in the
Gharanga data although some of the spherical sherds that have made it through the filter
are rather large at up to 200mmx200mm, indicating globular rather than pointed bases.
Body Sherd Thickness: Form 7 Robust Low Curvature Body Sherds:
While many sherds yielded large-radius curvature measurements, some unusually thick
sherds within this category are of interest, as their large curves and robust construction
suggest substantial vessels (Figure 91). Of the thicker sherds, those which yielded even
curvature measurements are shown in Figure 92.
343

Figure 91: Robust base sherds and cannister-shaped large body sherd (the latter having
exotic quartz-calcite temper).
344

These sherds seem likely to represent large and heavy vessels. Three are cylinder-shaped
in body, while the other is so flat as to be difficult to measure, with one thick tightly-
Figure 92: Curvature of robust body sherds (thicker than 14mm).
curved spheroidal sherd with a conical to spherical transition, likely to be the slightly
pointed spherical base of a large vessel. Shott’s theory of vessel size and use-life predicts
that such vessels will be under-represented in assemblages due to infrequent use/breakage
(Shott 1996). In light of this well supported theory of assemblage formation, unusual
sherds of these dimensions should be viewed as rare glimpses of vessel forms likely to
have been a prominent feature of past economy, and of low archaeological visibility due
to low breakage rates combined with low vessel sample sizes in the case of the Roviana
sites.
One can speculate on the functions of such vessels: storage of fluids, oils, starches
or dried or otherwise preserved or fermented foods, cooking or serving for large groups
on unusual occasions, or perhaps even provisioning/watering of canoes (one of these
sherds was of exotic quartz-calcite temper). It seems clear that due to the paucity of
345

evidence for such vessels and the likelihood of functional divergence from the run-of the
mill or frequently broken vessel form, that these sherds should not be used in any temporal
characterization of assemblages. Absence of decoration automatically excluded them from
the seriation analysis in Chapter 12.
Vessel Function from Evidence for Use-Alteration:
The varying degree of surface ablation of sherds from Roviana intertidal sites, evidenced
by pedestalled temper grains or rounded sherd edges, suggests that surface sooting, if it
was present, has generally been removed by postdepositional abrasion by water-born sand.
Heat spalling was not recorded for the Roviana materials due to the difficulty of
distinguishing heat spalling from salt crystallisation spalling.
One Form 6 vessel from Hoghoi had an encrustation of soot on the exterior (see
Chapter 12), thought to have been preserved as a result of burial in soft sediments for
much of the past three millennia. One sherd from this vessel was recovered by shallow
excavation, the other was recovered from the mud surface. Both had sooting. No other
evidence of surface sooting was seen, which is not to say that it does not exist (soot may
well be retained in surface crevices or in decorative punctations, or possibly grown over
with coral). It is not assumed that sooting relates to the pot’s primary or intended use, as
secondary or impromptu use over a fire without cleaning cannot be ruled out.
One other sherd displayed near-surface carbon, which was dated by AMS. A plain
body sherd from Honiavasa site was submitted for radiocarbon dating of a seed inclusion.
The date turned out to be about 9000bp, and the laboratory also dated a thin subsurface
carboniferous rind or zone near the surface of the vessel. This also yielded an age of
around 9000bp, suggesting that the clay incorporated organic materials of this age which
had vaporised and carbonized near the surface of the vessel during firing, and that this was
not use alteration.
346

Twenty large body sherds were examined as a pilot study to test the incidence of
macro-cracking (visible to the naked eye), micro-cracking (visible under 10x
magnification), pitting and spalling. Exterior and interior surfaces of sherds were compared
(eleven from Miho, five from Hoghoi and four from Honiavasa). Five sherds showed
interior macrocracking without exterior macrocracking. Old macrocracking of the exterior
occurred in seven cases and in all but two cases this was matched by interior
macrocracking. Most instances of microcracking appeared to be post-recovery as interior
faces of the micro-cracks were unweathered and without marine growth, thought to be due
to shrinkage on drying. Microcracking of the exterior was noted on eight sherds, matched
on the interior in five cases. There were a further three cases of interior microcracking
unmatched on the exterior. Three cases of exterior spalling were noted, all from the sample
of eleven Miho sherds. Three cases of interior spalling were recorded, all unmatched by
exterior spalling. Two of these were from Miho and one was from Honiavasa. Exterior
pitting was recorded on thirteen sherds, matched on the interior in eleven cases. Three
further cases of interior pitting were recorded which were unmatched on the exterior. This
pitting could have been post-depositional solution pitting of organic inclusions or
something similarly soft.
These pilot data seemed to hold little promise that any strong pattern of differential
alteration of interior and exterior body sherds would be seen, and no further use-alteration
data was collected. It would have been worthwhile, had the time been available, to look
microscopically at indentations on body/shoulder sherds to try to find retained traces of
carbon. The virtual absence of evidence for use alteration should not be read as evidence
for absence in the past, due to the slight polishing effect on sherd surfaces of water-borne
calcite detritus in the swash zone.
Secondary Smoothing of Vessel Interiors, and Interior Neck Form:
347

It was noted that some interior profiles of restricted, everted rim vessels had a smooth
upper body interior and a hard angle at the point of interior restriction (Figure 36 and
Figure 93). It was assumed that this pattern was produced by secondary smoothing of the
interior using a larger anvil than that used for primary forming, in contrast to those vessels
having an uneven interior composed of multiple small anvil/finger impressions. Some
vessels had thin walls with an even interior, but without the hard line around the interior
neck. The possibilities are either that these variants represented temporal changes in
potting skill/technique or that these are style or functional variation within/between
assemblages. These neck/shoulder interior variants are identified in table “Form.db” and
in table “Flat.db” appended on CD-Rom.
Four hundred and thirty-one sherds were recorded as having a restricted neck with
both the neck and shoulder represented. Of these, 308 were of area greater than 15cm 2 ,
chosen as a reasonable lower size limit, below which classification of shoulder interior
evenness or neck interior form was considered insecure. Interior evenness data by site are
given in Table 27. Hoghoi and Miho have the greatest abundance of even interiors, while
Paniavile yielded only a small sample and Gharanga had a low abundance of even
interiors.
Table 27: The attribute “even/not even” by site.
Site Count E Count not
E
348
Total % E % not E
Paniavile 8 14 22 36 64
Hoghoi 17 19 36 47 53
Miho 22 26 48 46 54
Honiavasa 30 62 92 33 67
Gharanga 7 28 35 20 80
Zangana 23 39 62 37 63
The relative abundance of hard neck interior profiles to not hard grouped Gharanga,

Figure 93: Hard interior neck profile and even interior body/shoulder profile (top); a
softer neck interior profile (middle) and uneven interior profile (bottom).
349

Hoghoi and Paniavile together (Paniavile had the smallest sample size, and is therefore the
least reliable data), while Miho and Zangana yielded similar relative abundances. The
Honiavasa site was the odd one out, with intermediate relative abundance (Table 28).
The relative abundance of even interiors with hard profiles to even interiors
without hard profile (see Figure 36 for definition of these attributes) grouped sites Miho,
Honiavasa and Zangana together, as distinct from the Hoghoi samples. Hoghoi, and
Gharanga / Kopo, if sample size is ignored, tended to have thin-walled vessels with even
interiors and soft interior profile at the neck (Table 29). This tendency is highly correlated
Table 28: Relative abundance of “hard neck” interior profiles to “not hard”.
Site hard not hard total % hard % not hard
Paniavile 3 19 22 14 86
Hoghoi 5 31 36 14 86
Miho 15 33 48 31 69
Honiavasa 18 74 92 20 80
Gharanga 4 31 35 11 89
Nusa Rov. 5 4 9
Kopo 0 4 4
Zangana 18 44 62 29 71
Table 29: Relative abundance of even interiors with hard neck interior profiles to even
interiors without hard profile.
site even and hard even not hard total % e and s % e not s
1(Paniavile)) 2 6 8
2 (Hoghoi) 1 16 17 6 94
3 (Miho) 9 13 22 41 59
4 (Honiavasa) 14 16 30 47 53
5 (Gharanga) 2 5 7
6 (Nusa Rov.) 3 0 3 100
7 (Kopo) 0 2 2 100
8 (Zangana) 11 12 23 48 52
350

with rim form (this Chapter) and decoration (Chapter 9).
These three relative abundances serve to separate Hoghoi from Miho/Zangana, and
Honiavasa from all of these. Miho/Zangana have the highest relative abundance of hard
neck interiors, while those site assemblages which have lower numbers of hard necks can
be separated into two groups, one of which has even interiors highly correlated with hard
necks, the other of which has even interiors with a softer neck, indicative of a different
production style or technique. The low abundance of even interiors at Gharanga is of
interest, as decoratively Gharanga, Hoghoi and the northern end of Zangana bear many
similarities (explored further in Chapters 11 and 12), and this is reflected also in Table 27;
but in terms of their body sherd form these assemblages show some differences (this
chapter) and this is reflected in their neck angles and rim forms also (this chapter). The low
evenness count for Gharanga compared to Hoghoi may have more to do with a difference
in site function than a temporal difference indicated by other aspects of vessel form,
whereas the differences in evenness abundance and neck interior hard profile abundance
between other sites seem to correspond with stylistic differences, and are thought therefore
to be more safely viewed as a temporal variant within the Roviana Lagoon region.
Shoulder Form Classes of Form 6 Vessels:
Two shoulder variants were recorded for restricted everted-rimmed jars: hard shoulders
(see Figure 34 for a definition of these and for examples see Figure 94) and the much
more common soft shoulders (e.g. Z.11.572, Figure 94). These variants were reminiscent
of two forms recorded for the Kalinga of the Phillippines (Longacre 1991). In that case
they were contemporaneous geographic variation of no functional significance within a
local region. There seems no evidence at present to suggest that these are correlates of
functional differences, and the Kalinga case offers uniformitarian support for this.
351

Figure 94: Hard shoulder variants of Form 6 (top and middle) as distinct from the more
common soft shoulder form (bottom).
352

These differences most likely reflect either slight differences in manufacturing technique
or simply different production styles, either contemporaneous or temporal. Due to the
possibility that this represents contemporaneous variation, this attribute was not used in
seriations.
Sherd Restriction Factor:
Sherd restriction factor is a measure of the amount of surface area of sherds in an
assemblage in relation to the amount of orifice represented (Smith 1985:269). Calculation
of sherd restriction factor was made using EVE as a unit of rim circumference rather than
cm as used by Smith. This might create some orifice-size bias in the results, but as neck
Vcurve was not highly structured by site, this should not be too much of a problem. It
would have been preferable to measure circumferences of lips, necks and carinations in cm
as well as EVE.
Table 30: Sherd restriction factor comparison of site samples.
Site total sherd surface
area (cm 2 )
353
total rim circ. (%
EVE)
Paniavile 9438 375 2517
Hoghoi 9421 448 2103
Miho 8438 274 3080
Honiavasa 12487 814 1534
Gharanga 4976 215 2314
Zangana 11694 493 2372
Sherd restriction
factor (cm 2 per
100%)
Miho had the highest sum of sherd surface area per everted-rim vessel equivalent (Table
30). Honiavasa had by far the lowest, while the other sites formed a group (in those cases
where sample size permitted comparisons). The first question in relation to these data is
whether collection intensity has biased the ratio of rim sherds to body sherds, and here
Honiavasa has the lowest collection intensity, with Hoghoi and Zangana the highest. Miho,
in spite of having an intermediate intensity of collection, not much

Table 31: Preservational bias of vessel parts as an explanation for differences in sherd
restriction factor (data has the form average thickness (count) standard deviation).
Site Lip
Thickness
(mm)
Paniavile 8.21(66)
2.83
Hoghoi 6.32(84)
1.78
Miho 6.45(65)
1.60
Honiavasa 8.81(102)
1.96
Gharanga 7.11(34)
1.45
Zangana 7.76(80)
2.37
Rim
Thickness
(mm)
7.49(141)
2.12
5.96(135)
1.51
7.21(111)
1.5
8.33(171)
1.79
7.02(49)
1.56
7.25(126)
1.79
Neck
Thickness
(mm)
9.49(128)
2.12
7.72(152)
2.12
8.84(104)
2.24
10.28(173)
2.34
8.33(74)
1.68
8.93(169)
2.45
354
Shoulder
Thickness
(mm)
8.91(75)
2.49
7.14(137)
2.03
8.37(97)
2.42
9.50(146)
2.48
8.04(68)
1.65
7.84(135)
2.18
Body
Thickness
(mm)
8.21(278)
2.26
6.67(355)
1.93
7.84(112)
2.13
8.69(139)
2.06
7.24(80)
1.39
7.03(247)
1.87
L:N
Ratio
L:B
Ratio
0.87 1.00
0.81 0.95
0.73 0.82
0.86 1.01
0.85 0.98
0.87 1.10
greater than Honiavasa, had the highest restriction factor, where for Honiavasa the figure
was almost half. The intensively collected Hoghoi and Zangana sites had near average
restriction factor. There therefore seems to be almost no correlation between collection
intensity and sherd restriction factor.
When the ratio between average lip thickness and average body thickness was
compared for site assemblages (Table 31), Miho, the site with the highest sherd restriction
factor, had the lowest lip to body thickness ratio, and lip to neck thickness ratio, predicting
significant postdepositional bias against lip sherds in this assemblage. Honiavasa has the
thickest and most rugged lip sherds of all sites, slightly thicker on average than body
sherds, offering clear support for a taphonomic rather than vessel form / functional
interpretation of restriction factor.
Tall, weakly everted rim sherds may be particularly fragile and may be under-
represented in assemblage EVE measurements due to failure to survive. There is some
support for such an assessment of the Miho sample and the southern area of the Zangana
sample, which included some particularly tall but incomplete everted rims (Figure 95).

Chapter Summary:
The primary evidence used in the identification and interptetation of vessel forms was
classification of vertical contour of the upper vessel. Form variability was analysed as a
means of instituting some control for functional variation in trying to identify stylistic
variation, or variation subject to drift over time. As vessel form is inferred from sherds
rather than observation of whole pots, epistemological issues arise in the identification of
form, and throughout the analysis the aim was to depend as little as possible on the form
of lip/rim/sherds for information on body form.
Thus analysis of vessel parts was separated into various sections, with more
emphasis than is usual on body sherds, restriction factor, shoulder forms and neck/interior
characteristics. Also, negative evidence was used: the absence of flat bases, stands, etc was
noted. The paucity of bases identified in the Roviana data (only a few sherds were
identified as bases, and these were rounded) strongly suggests that rounded, uniform
thickness bases were overwhelmingly the norm. Six broad functional vessel form classes
were identified through examination of upper body contour variation, while a seventh was
defined on the basis of body sherd thickness and form.
Forms 1-6 were defined using combinations of nominal descriptive variables. All
assemblages were dominated by Form 6 (everted rim restricted vessels), identified securely
by neck sherds. This vertical contour information was supplemented with metric analysis
of neck and rim form, body sherd curvature and sherd restriction factor. Analysis of metric
variables recording lip, rim, neck, shoulder and body curvatures, angles and thickness in
some cases, resulted in re-evaluation of the nominal combinations of Forms 1-6.
355

Figure 95: Tall, fragile everted excurvate rims.
356

Form 6 was subdivided into 6a, 6b and 6c using neck horizontal curvature variability.
Analysis of neck and orifice sizes indicated that the bulk of vessels of these forms had
orifices of hand-size up to head-sized, suggesting substantial size variation. Form 6c was
merged with Form 2c due to a subjective distinction between these classes, which also
need not be salient to functional variation.
Forms 2c and 6c had general-purpose form characteristics consistent with cooking,
liquid transport, serving, storage, and processing functions. An analysis of rim variation
within Form 2c/6c identified a negative correlation between rim height and eversion angle,
which tied into ideas about form strength that came up in the analysis of restriction factor,
and may also indicate some functional differences regarding suitability for liquid transport,
stacking, and access for stirring or ladling. Short, heavily everted rims are at the
stirring/ladling end of this continuum, while tall weakly everted rims tend more towards
liquid transport, with the tradeoff being a tendency to have a fragile thin lip (under-
represented in the sample), except where slab construction is used, resulting in a more
robust tall rim. Forms 2c/6c are the subject of a form/decoration analysis in the next
chapter.
Analysis of body sherd variability found some differences between sites. The
Honiavasa site was characterized as having mainly spherical bodies less than 150mm in
radius, with relatively few conical or cylindrical sherds. Gharanga was at the opposite
extreme, dominated by cylindrical sherd forms. Sample sizes were small for these analyses
when smaller sherds were filtered out, but do illustrate the uses to which body sherds can
be put in comparing samples of relatively plain potsherds.
Surface ablation of sherds, while moderate in may cases, was sufficient to remove
sooting in most cases, making use-alteration study difficult. Sooting was noted on one
multi-purpose everted-rim Form 6c vessel (see Chapter 12). A pilot study of
microcracking, macrocracking and spalling found that most instances of these had
probably occurred post-recovery. Some potential for soot preservation in crevices was
357

noted but was not followed up due to time constraints.
Form 7 comprised a small number of robust body sherds. One of these seemed
likely to be the tightly rounded base of a sub-conical vessel body, while the others were
wall-sided cylinder forms of relatively large horizontal radius. These seemed likely to be
rare preserved examples of high-use life large robust vessels, possibly storage vessels. One
of three sherds was of exotic quartz-calcite hybrid temper.
Sherd restriction factor was calculated for site samples, but was found to be subject
to taphonomic bias. The part thickness data which were used to justify this conclusion
draw attention to differences in lip and rim strength between sites, which created
differences in the number of vessel equivalents represented by lip EVE, relative to total
sherd surface-area in site assemblages. This will be shown to be significant for the analysis
of decorative variation in the next chapter.
A concern with sampling errors and measurement errors was maintained
throughout the chapter. Various size filters were applied during analyses to systematically
exclude or reduce the impact of measurement errors or poor representations of vessel
characteristics.
Form 1 and Forms 2c/6c are analysed decoratively in the next chapter, where the
relationships between vessel form variability and decoration are investigated further. Here,
the question for seriation is whether decorative variation is strongly correlated with form
variation, or whether these vary independently. In the latter situation, we approach more
closely a set of attributes for analysis that are not strongly structured by vessel function,
but rather by style, and thus more likely to show drift with time.
358

CHAPTER 9:
CERAMIC DECORATIVE CLASSIFICATION AND
Introduction:
DECORATION/FORM VARIABILITY:
Having made relatively detailed attribute descriptions of sherds, the analyst is faced with
the task of refining which of the recorded variation is salient to the research questions. One
crude but widely used option is to use all descriptive variability in seriation. In the worst
scenario these data are used in a similarity matrix clustering approach, which will yield
assemblage groupings, the significance of which may be hard to assess. The approach
taken here is an exploratory analysis of covariance between variables in the data, with
regard to theories of ceramic variation.
Sherd decoration was recorded primarily as nominal or category data (with mark
dimensions and spacing measured but mostly unused). A primary task in this chapter is to
move from descriptive decorative information about sherds to information about variability
of decoration at the scale of pottery vessels, as the units of behaviour across which
decoration is patterned during manufacture. Some patterns are difficult to reconstruct from
sherds (larger incised patterns for example). The first section deals with classification of
these using zone markers, which are identifiable across a range of levels of brokenness due
to location at rugged corner points of the vessel profile (lip, neck or carination).
Simple bands of decoration, and the permutations of these across the vessel parts,
provide a window into decorative structure that is, like the linear zone-markers of some
of the larger patterns, insensitive to brokenness levels. The flat data structure in “Flat.db”
(appended on CD) permits an evaluation of part representation and assemblage
information content: the “diagnosticness” of sherds. Sherds that represent more of the
vessel profile tell us more about the structure of decoration across the vessel. Sample sizes
359

for various part combinations are given and discussed.
Covariation of decoration with vessel part is investigated using these samples, and
this exploratory data analysis is used to refine ideas about the vessel production styles
introduced mainly as illustrations in Chapter 3. A detailed analysis of lip
impression/notching metric variability is given, allowing assessment of the temporal
saliency of these decorative attributes in general, and of some widely used classificatory
schemes.
In a section integrating form variability with decorative variability, ideas are
developed about which vessel decorative styles might be strongly constrained by or
correlated with artefact form/function. Decorative types that crosscut form types can be
better argued to be temporal styles (in the Dunnellian sense) rather than functional types,
as there is some evidence for functional variation within the decorative class. Decorative
attributes strongly correlated with particular forms may be poor attributes for seriation as
there is a possibility that the analyst is seriating contemporaneous decorative variability
related to function.
Classification of Linear Motifs:
As discussed in Chapter 1, a structural formulaic analysis is suited to simple arrangements
of elements, but larger, more varied, or less structured patterns encourage the use of
broader classification criteria. Sample size is also an issue in making decisions about
classification. The more vessels there are in a sample the more the classification can split,
without running into sample size problems for counts of attributes. Conversely, where a
sample is smaller, classificatory units may need to be lumped. Faced with the need to boost
sample sizes of complex and highly varied linear motifs located on the rim, shoulder or
body, and avoid the sparse data that would result from a splitting approach, these were
categorized using two criteria
• whether or not the decorative zone was delimited by a linear marker (as opposed
to for example, a band of wavy deformation at the lip or a band of opposed pinch
360

fingernail impression at the neck, or simply absence of a zone marker band)
• whether the primary pattern was composed of single or double/triple lines)
These criteria divided the motifs into three groups, those with linear zone markers
comprising double or triple lines delimiting the primary pattern (Class 1), those without
linear zone markers, and with the primary pattern laid out in single lines, often but not
necessarily with filled motifs (Class2), and unclassified motifs (Class 3).
Class 1 comprised the following pattern codes: CH9, C10, C11, C12, C13, CL5,
GM3, GM4, RL1. (See Table 15)
Class 2 comprised the following pattern codes: CHV, CH2, CH3, CH4, CH5, CH6,
CH7, CH8, C15, C17, C18, C19, C20, C23, C24, C25, CT3, CTW, GM5, GM7, PBR,
RBR. (See Table 15)
These criteria were developed to express perceived differences between
Honiavasa/Nusa Roviana incised/stamped motifs and all other sites.
This classification-by-zone-marker owes a lot to Mead (Mead et al. 1975),
although it is a lumping of Mead’s zone marker categories, and is similar to Wickler’s
classification of incised motifs for the Buka Lapita-phase reef sites (Wickler 2001: 112-
113), although in that case vessel corner points were viewed as zone markers, so the
distinction here is slightly finer. The classification used here differs substantially from motif
inventory approaches (e.g. Anson 1983, Bedford 2000), being a lumping of heterogenous
motifs according to broad classificatory criteria. This has the benefit of increasing sample
size for Class1 and Class 2 decoration, where a motif-based splitting approach would leave
the data with almost as many classes as observations, i.e. sparse.
Part Representation in the Potsherd Sample:
The diagnostic value of sherds for the development of a classification utilizing decorative
structure across the vessel depends on how complete a representation of vessel form and
361

decoration the sherds provide. All sherds are not equal in the information they convey
about decorative structure across the vessel, so sherd count alone is not an accurate
representation of the sample size bearing on the analytical problem. Sample sizes for the
various permutations of vessel part representation are given in Table 32, Table 33 and
Table 33. These information are given to convey the overall level of fragmentation of the
sherd assemblage, particularly as it pertains to decoration and form analysis, and
complement information given elsewhere on sherd area and weight, and lip EVEs.
Table 32: Part representation, all sherds including necked sherds but excluding
carinated and/or inverted-rim sherds.
lowermost part
lip rim neck shoulder body
lip 22 266 91 60 14
topmost rim 118 68 167 37
part neck 223 210 62
shoulder 233 75
body 1053
Table 33: Part representation for inverted-rim sherds (no neck corner point expected
from morphology of sherd).
lowest part
rim shoulder/carination body
topmost part lip 9 2 3
362

Nine-hundred and fifty eight sherds were not identified by vessel part. The bulk of these
would be body sherds or shoulder sherds with a convex exterior surface. For smaller sherd
sizes shoulders and bodies are impossible to distinguish from each other. By site, there
were differences in part representation that are thought to relate to the different breakage
patterns of various pottery styles. This is illustrated by the data in Table 35.
The Honiavasa site has the highest individual part counts, but the lowest relative
Table 34: Carinated sherds (excluding carinated sherds with inverted rims).
rim 3
363
lowest part
carination body
neck 5 2
Topmost part shoulder 10 7
Site Lip
observation
count
carination 3 2
Table 35: Form strength variation for different pottery styles and the effect on part
representation.
Rim
observation
count
Neck
observation
count
Shoulder
observation
count
Paniavile 80 146 125 82 2
Hoghoi 85 138 151 143 23
Miho 66 115 107 98 9
Honiavasa 103 171 170 148 4
Gharanga 34 52 74 70 9
Nusa Roviana 9 31 25 20 2
Kopo 3 7 8 8 1
Zangana 80 134 172 144 10
total all sites 460 794 832 713 60
LRNS sherd
count

proportion of Lip-Rim-Neck-Shoulder sherds (LRNS). As sherd size is on average larger
for this site than others, this low LRNS count is not an artefact of relative harshness of
taphonomic regime, but is instead interpreted as a relative weakness of the construction
of necks in the site which caused many sherds to break at the neck rather than elsewhere.
Similarly, the high LRNS count at Hoghoi is thought to relate to high form strength of
short, heavily everted rims common in this site. The low LRNS count at Miho and
Paniavile have something to do with rim fragility also.
Covariation Between Vessel Part and Decoration:
Lips and Rims:
Unbounded incision of the rim was matched by deformation into a wave form more often
than expected from sample sizes, but this could have been sampling error as sample size
for this decorative permutation was low, due to very few unbounded incision occurrences
having the lip and rim in one piece (14 cases in total) (Table 36). Punctation of the rim
was matched by plain lips more often than expected from sample sizes (15 examples where
nine would have been expected from sample sizes). Plain rims showed no particular
patterning with lip decoration beyond what would be expected from sample sizes.
Lips and Necks:
Lips and necks, being non-adjacent vessel parts, suffer from smaller sample size due to
breakage (155 in total:Table 37). Plain necks do not seem to depart significantly from the
incidences expected from sample sizes. This suggests lip decoration is not highly correlated
with neck decoration for sherds that survive in this configuration. This does not mean that
such structure was absent in the past, but suggests if it was present, it occurred
364

Table 36: Lip decoration and rim decoration.
Lip Decoration
Class 2 punctation
band
365
Rim Decoration
unclassifie
d
plain Total
Count
unclassified 1 (7%) 3 (11%) 5 (11%) 27 (8%) 36 (8%)
plain 5 (36%) 15 (54%) 16 (36%) 94 (27%) 130
(30%)
discontinuous deformation 2 (14%) 3 (7%) 38 (11%) 43 (10%)
impression, “wav” pattern 2 (7%) 8 (2%) 10 (2%)
deformation, “wav” pattern 5 (36%) 1 (4%) 5 (11%) 69 (20%) 80 (18%)
fingernail pinching, single band 1 (7%) 3 (7%) 24 (7%) 28 (6%)
impression, band on top edge 2 (7%) 4 (9%) 28 (9%) 34 (8%)
impression, band on inner edge 3 (7%) 20 (6%) 23 (5%)
impression, bands on both edges 4 (14%) 1 (2%) 8 (2%) 13 (3%)
impression, bands on outer edges 1 (4%) 3 (7%) 31 (9%) 35 (8%)
applied nubbins, band 1 (2%) 1 (0%)
Count 14 (100%) 28
(100%)
on forms for which the lip and neck were easily separated.
44 (100% ) 347
(100%)
433
(100%)
The low number of classified linear motifs occurring on the neck suggests this
occurrence is an inaccuracy on the part of an individual potter, and that the mental
template to which the potter or potters were working did not include decorating the neck
in this manner (more on this below).
Sherds from Honiavasa site are commonly broken at the neck. It was suspected
that many of these Honiavasa necks were plain as usually there was no trace of decoration
near the break. Only two lip/neck sherds represent the numerous opposed-pinch
fingernail-impressed necks, suggesting that these necks are attached to a fragile rim-lip
form. This is an important point for taphonomic inference and supports the idea that many

of the Honiavasa vessels were slab-constructed (see Chapter 7). The common tall rim
form at Miho may have been an attempt to replicate slab-constructed tall everted rims as
seen at Honiavasa, but using the one-piece method.
Table 37: Lip decoration and neck decoration.
Neck Decoration
Lip Decoration
Class
2
Unclassified 1
(100%)
punctation
band unclassifie
366
plain band
pinchin
nubbins
band
Total
2 (15%) 3 (75%) 12 (9%) 18
(11%)
Plain 4 (31%) 1 (25%) 45 2
(34%) (100%)
Deformation,
discontinuous.
Impressed bands,
both edges, “wav”
pattern
Deformation
“wav” pattern
Pinching, single
band
Impression, band,
top face
Impression, band,
inner edge
Impression, band,
both edges
Impression, band,
outer edge
Column total 1
(100%)
1 (8%) 15
(11%)
2 (15%) 15
(11%)
1
(100%)
53
(34%)
11 (8%) 11
(7%)
2 (2%) 2 (1%)
16
(10%)
10 (7%) 10
(6%)
17
(11%)
3 (23%) 6 (4%) 9 (6%)
1 (8%) 8 (6%) 9 (6%)
13 (100%) 4 (100%) 134 2
(100%) (100%)
10 (7%) 10
(6%)
1
(100%)
155
(100%)

Lips and Shoulders:
Here sample size is even more of a problem (71 observations in total), and a severe bias
towards rugged LRNS forms is likely. No significant departures from the expected pattern
based on attribute sample sizes are evident (Table 38).
Table 38: Cross-tabulation of lip decoration with shoulder decoration.
Lip Decoration
Class
2
pinching,
multi-band
Shoulder Decoration
band
punctation
367
unclassified plain pinch
band
unclassified 2 1 1 1 1 6
plain 1 6 2 1 19 29
deformation,
discontinuous
deformation,
“wav” pattern
2 5 7
6 6
pinch band 3 3
impressions,
band, top face
impressions,
band, inner edge
impressions,
band, both edges
impressions,
band outer edge
1 1 5 7
3 1 4
1 1 2 4
1 4 5
TOTAL 1 14 5 4 46 1 71
TOTAL

Rims and Necks:
Here sample size is improved to 389 due to part propinquinity (Table 39). Class 1
bounded linear motifs on the rim are rare, and occur in association with single examples
of unclassified neck modification, plain neck and opposed pinch fingernail-impressed band
respectively. (See also sherd HV.2.297 in Figure 9 which has Class 1 decoration of the rim
and a band of single fingernail impressions at the neck.) Class 2 unbounded linear motifs
occur predominantly in association with fingernail-pinched necks, with a minority of
examples linked to plain necks also. Punctate rims are rare, and sample size is insufficient
to see any patterned association with necks. Plain rims are matched more often than
expected from sample sizes by punctate necks or by plain necks, and more rarely than
expected by necks with a band of fingernail pinching.
Table 39: Cross-tabulation of rim decoration and neck decoration.
Neck Decoration
Rim Decoration
Class 2 pinching band unclas- plain pinchnub- TOTA
multipunctasifiedingbins L
bandtion single
band
Class 1 linear 1 1 1
3
(5%) (0%) (2%)
(1%)
Class 2 linear 1
2 9 19
31
(20%)
(11%) (3%) (46%)
(8%)
pinching,
1
1
multi-band
(100%)
(1%)
punctation,
20 1 1 22
band
(7%) (2%) (33%) (6%)
unclassified 1
1 5 12 9
28
(20%)
(6%) (26%) (4%) (22%)
(7%)
plain 3
16 11 261 11 2 304
(60%)
(94%) (58%) (86%) (27%) (67%) (78%)
TOTAL 5 (100%) 1 17 19 303 41 3 389
(100%) (100%) (100%) (100%) (100%) (100%) (100%)
368

Rims and Shoulders:
Sherds representing both rims and shoulders numbered 243, a better sample size than lips
and necks, indicating the more rugged nature of these sherds (Table 40). There were fewer
plain rims than expected that had Class 2 shoulders, fewer plain rims with shoulders having
multiple bands of pinching, and fewer plain rims than expected with shoulders having a
single band of pinching. Class 2 shoulders tended to have unclassified decoration, usually
consisting unclassified or incomplete unbounded linear motifs (Figure 12). There were
more punctate rims with multiple bands of pinching on the shoulder than expected from
sample sizes. There were more plain rims with plain shoulders than expected from the total
column percentage plain.
Table 40: Cross-tabulation of rim decoration and shoulder decoration.
Shoulder Decoration
Rim Decoration
Class 1
linear
Class 2
linear
Class 1 linear 1
(10%)
Class 2 linear
pinching, multiband
punctation,
band
2
(20%)
unclassified 4
(40%)
plain 1 3
(100%) (30%)
TOTAL 1 10
(100%) (100%)
Necks and Shoulders:
punctati
on
multiband
369
punctat
ion
band
unclass. Plain pinching
band
1 1
(7%) (1%)
3 16
(21%) (8%)
1
(6%)
7
1 10
(41%)
(7%) (5%)
1
3 7
(6%)
(21%) (4%)
8 5 6 156
(47%) (100%) (43%) (82%)
17 5 14 190
(100%) (100%) (100%) (100%)
3
(50%)
3
(50%)
6
(100%)
Total
3
(1%)
24
(10%)
1
(0%)
18
(7%)
15
(6%)
182
(74%)
243
(100%)
Here part propinquinity benefitted sample sizes, and 449 observations pertain to this

elationship (Table 41). Of the 15 punctate necks, nearly half had multiple bands of
fingernail pinch on the shoulder, the rest were plain. Of the 348 plain necks, 87% were
matched by plain shoulders, but there were a number of examples of multiple bands of
fingernail impression (14) and single bands of punctation (8) also. Plain necks occurred less
often than expected for Class 1 and especially Class 2 linear motifs on the shoulder. A
single band of pinching at the neck occurred more often than expected with Class 2
shoulders. These were matched in ten cases by Class 2 unbounded incision. Lateral bands
of applied nubbins did not form a large sample in this analysis, but in four of seven cases
were matched by Class 1 bounded incision on the shoulder, while only three cases were
matched by plain shoulders.
Table 41: Cross-tabulation of neck decoration and shoulder decoration.
Shoulder Decoration
Class Class 2 pinching punc- plain pinchin nubbin TOTAL
1 linear multitationunclas- g s
linear band band sified band band
Class 2
5 2
7
linear
(1%) (33%) (2%)
Pinching,
1 1
2
multi-band
(6%) (5%)
(0%)
Punctation,
7
8
15
single band
(32%)
(2%)
(3%)
Unclassified 1 3
6 10 1
21
(12%) (18%)
(24%) (3%) (17%) (5%)
Plain 2 3 14 8 13 304 3 1 348
(25%) (18%) (64%) (100%) (52%) (84%) (50%) (100%) (78%)
Pinching, 1 10
6 32
49
single band (12%) (59%)
(24%) (9%)
(11%)
Band of 4
3
7
nubbins (50%)
(1%)
(2%)
TOTAL 8 17 22 8 25 362 6 1 449
(100%) (100%) (100%) (100% (100%) (100%) (100%) (100%) (100%)
Neck Decoration
370

Summary of Decorative Structure Across the Vessel:
Lips were more likely to be decorated than other parts (30% of lips were plain, while 78%
of rims were plain, 78% of necks were plain, and 81% of shoulders were plain. Although
the types of decoration present on lips was quite strongly structured by site (Table 42),
there was little information to be had from the sherds themselves as to which lip decorative
attributes went with which other decorative attributes, due to sherd breakage.
Lip/rim sherds showed a weak relationship between Class 2 unbounded incision
of the rim and wave deformation of the lip, although this could have been a sample-size
error and the vast majority of wave-deformed lips which had a rim attached had plain
rims. There were no examples of unbounded incised rims with impressed decoration, but
there were sample size problems with unbounded incised rims. The low number of
complete lip/rim sherds for this rim decorative category suggests a fragile lip/rim form.
Punctate rims, by contrast, were more likely to have plain lips or impressed lips, with only
one example of wave deformation. While there was one example of a punctate neck with
wave-deformation of the lip, like punctate rims, punctate necks had either plain lips or
impressed lips in general. In spite of small sample size, punctate shoulders showed a
similar lip patterning to punctate rims and necks, and multiple bands of fingernail
impression on the shoulder followed this tendency towards either plain or impressed lips
also. There was only one sherd relating unbounded incision of the shoulder to lip
decoration (plain), a further indication of the fragile lip-rim form associated with
unbounded incision.
A band of opposed pinch fingernail impression on the lip did not occur with any but
plain rims and necks, with the exception of a single sherd which had dentate-stamped
bounded linear motif on the vessel body. This may relate simply to sample size, but the
concentration of sherds with this lip decoration in a single site on particular vessel forms
will be discussed further in Chapter 12. (This lip decoration was almost exclusively found
in the Lapita-phase Honiavasa site, which is consistent with the occurrence of this
decorative motif on a single dentate sherd from the Nusa Roviana site.)
371

Unbounded incision of the rim was most commonly associated with a single band
of opposed pinch finger-nail impression at the neck, and less commonly with a plain neck,
while by contrast punctation of the rim was overwhelmingly associated with a plain neck
(20 cases), with one instance of a single band of fingernail opposed pinching and one
instance of a band of applied nubbins at the neck.
Sherds with both rim and shoulder were not common, due to breakage, but
suggested that where rims with unbounded incision were matched in the majority of cases
by plain shoulders or occasionally unbounded incision on the shoulder, rims with
punctation were less likely to be plain-shouldered, and where the shoulder was decorated
for a punctate rim, all examples (7) were multiple bands of opposed pinch fingernail
impression.
Necks with a single band of opposed-pinch fingernail impression were most
commonly plain shouldered, but where the shoulder was decorated in such cases, the
decoration was almost always unbounded incision (10 examples). Only in three cases were
plain necks associated with unbounded incision on the shoulder, and in a further three
cases plain necks were associated with a single band of opposed pinch fingernail
impression on the shoulder. These latter three cases could be regarded as decorative
imprecision, where the intention was to apply a band of fingernail impression to the neck,
but the execution was less precise than the academic definition of body part in the present
instance. Two cases where the neck had unbounded incision and the shoulder had a single
band of opposed pinching were regarded as further examples of the pattern having slipped
down a little.
Where plain necks were associated with decorated shoulders, the decoration was
most likely to be a band of punctation or multiple bands of fingernail impression. Punctate
necks were sometimes matched by undecorated shoulders, and sometimes by multiple
bands of fingernail impression.
While sample size was small, the majority of examples of a band of applied
nubbins on the neck were associated with bounded linear motifs on the shoulder, with the
remainder associated with plain shoulders.
372

Interpretation of Decorative Co-variation by Vessel Part:
While the data presented could never prove that there are emic types in the assemblages,
it is consistent with the idea that the etic decorative structure of a motif, its analytical
meaning, varies according to its position on the vessel. In this view, a single band of
fingernail impression on the lip has a different attribute value to a band of fingernail
impression at the vessel neck. Fingernail impression as multiple bands at neck or shoulder
have the same value, a value which differs from either a single band of pinching at the lip,
or a single band of pinching at the neck. Unbounded incision has a similar value whether
it is on rim or shoulder. A single band of punctation has the same decorative value whether
it occurs on rim, neck or shoulder, a case of the analyst being more specific about location
on the vessel than the potter of the past. These data supported the classification of sherd
assemblages in terms of analytical decorative vessel classes, formulated in terms of the
structure in the cross-tabulations given above. These were useful for a number of purposes:
to provide a univariate decorative description for spatial analysis and for easy comparison
with form and fabric information below.
Miho Decorative Class:
Fingernail pinching of the neck, for example, commonly occurred with a plain rim or
shoulder, but when decorated, almost always occurred with unbounded incision on those
parts. While it was not possible with this information to classify undecorated sherds, it
seemed reasonable to regard sherds with unbounded incision (neck or shoulder), and
sherds with a single band of fingernail pincging at the neck, as belonging to a single group
(Miho-class decoration after the Miho site where this seemed prevalent). It is possible, and
likely, that the attributes which make up this grouping have independent rates of change,
and a seriation using correspondence analysis in Chapter 12 uses the short list of attribute
counts given in Table 42 for this reason, rather than the more normative decorative
classes.
The assignment of lip sherds to such a grouping is problematic due to poor
373

preservation of lip-rims for this group (see Chapter 11 for a spatial approach to this
problem). Wave-deformation of lips seemed to be associated with unbounded incision, but
the predominance of plain rims with this lip form suggested caution, as did spatial
information (Chapter 11). It is conceivable that a class of plain vessels with wave-
deformation of the lip could hide within a classification that lumped all wave-deformed lips
with unbounded incised rims and single-band pinched necks.
Gharanga/Kopo Decorative Class:
Another set of attributes that seem to group to form a style are punctation of the rim, neck
or shoulder and multiple bands of fingernail pinching of the shoulder/upper body.
Punctation of the rim, neck or shoulder may occur with plain adjacent parts, but where
other decoration is present below this band of punctation, it was always multiple bands of
opposed pinch fingernail impression. This decorative association was termed
Gharanga/Kopo style decoration in Chapter 3 after the sites where it is common. Rare
instances of punctation of the lip occurred with a variety of decoration, and were omitted
from this analysis as not highly correlated with any other decorative attribute.
Attribute Frequencies:
Attributes as defined in relation to vessel part above and their frequencies are given in
Table 42. The data for Zangana site has been provided as a single sample, but also is split
into two areas, Zangana South and Zangana North, as a result of clear large-scale spatial
structure of decoration at Zangana (explained in detail in Chapter 11). The two Class 2
linear motifs on the rim at Zangana North were sherds A21 and A22, found just outside
the arbitrary boundary of Zangana North/Zangana South used in the preparation of Table
42 (the boundary used in preparation of the attribute frequencies being in the vicinity of
the modern wharf at the end of the Munda-Zangana road).
374

Table 42: Counts of occurrences of decorative attributes.
Attributes
Paniavile
Hoghoi
Miho
375
Honiavasa
Sites
nubbins, band at neck 2 8 1 11
pinching, band at lip 5 22 1 1 28
pinching, band at neck 20 4 21 1 1 2 20 1 19 69
pinching, multi-band 3 10 2 11 2 1 1 28
lip impression, band inner edge 2 5 1 22 3 1 1 34
lip impression, band outer edge 11 7 3 4 4 2 4 2 2 35
lip impression, band top face 1 10 10 6 1 6 2 4 34
lip impression, band both edges 1 4 3 6 2 2 16
lip impression, staggered both
edges to form “wav” pattern
1 5 2 2 8
lip deformation, “wav” pattern 25 11 22 2 4 3 20 10 10 87
lip deformation, discontinuous 8 8 12 2 1 1 1 6 2 4 39
punctation, single band
anywhere
4 27 4 19 1 1 12 11 1 68
Class 1 linear motif 16 1 17
Class 2 linear motif rim 14 1 22 14 2 12 51
Class 2 linear motif shoulder 10 2 3 1 8 8 24
stamped decoration (dentate or
wavy)
6 1 7
Total 104 92 100 93 56 11 2 98 34 64 555
Variability of Impressed Lips:
Lip notching and crenulate lips are often cited as a temporal markers, and in this section
a sample of sherds with lip impression are analysed to examine the discreteness of these
categories and to ask whether this type of simple decoration should be regarded as a
homologous trait or as a simple analogous trait. In the Roviana sample, bands of
Gharanga
Nusa Roviana
Kopo
Zangana combined
Zangana North
Zangana South
Total

impression on the inner or outer and top edges of lips co-occurred with single bands of
punctation and multiple bands of opposed pinching lower on the vessel, and failed to occur
in the single band pinched neck group. In the Honiavasa assemblage such lips were
common in the absence of any punctation or multiple-band pinching, suggesting that lip
impression may be a very generalized technique that may well be reinvented over time or
occur across production units, and should be avoided for the purposes of seriation. The
presence of lip impression on late-prehistoric terrestrial plainware from Site 25, dated to
1400AD provided support for this view of lip “notching” as a simple decorative analogue.
Review of Lip Notching as a Temporal Marker:
Specht developed a theory that Watom Lapita and Buka-Phase Lapita were characterized
by a broad, deep, widely spaced notching of the lip (which looks like a v-shaped notch in
Specht’s photograph), a decorative style that did not occur later, although “simple
notching” was found throughout the Buka sequence (Specht 1969:218). In a similar vein,
for post-Lapita pottery from New Ireland, Golson differentiated three classes of notching
or crenulate decoration (v-shaped incision, parallel-sided incision/indentation, and wide
stick or thumb indentations (scallops) (Golson 1991, 1992). Similarly, White and Downie
differentiated several lip treatments of this sort (White & Downie 1980). Like Specht,
Wickler noted rare examples of deep u-shaped impressed lip notches in site DAF (Wickler
2001:120), as well as a variety of less bold notching or impression, classified in a similar
way to the Roviana data above, according to position on the rim (see also Kirch et al.
1991, Reeve 1989). Given that various people have classified pottery using the dimensions
and form of notching or impression in various ways, it seemed that the sample of Roviana
notched/impressed rims could usefully be analysed using the metric data that had been
376

ecorded, to pose the question whether there were discrete classes in the Roviana data, or
whether these classes are better regarded as arbitrary measurement devices, in which case
they require careful metric definition. Metric data recorded for bands of lip impressions is
analysed below using bivariate plots. The relationship between form and decoration is also
examined, as one form attribute seems to be correlated with the location of impressions,
suggesting that decorative location was an analogous decorative similarity arising from
form similarity in a single attribute, rather than any general similarity in form.
Lip Notching /Impression/Crenation in the Roviana sample:
There is no particular pattern to the metric data concerning banded parallel impressions on
the outer lip (see Figure 96). Neither, significantly, is there any strong pattern discernable
in the distribution of mark section. Not only does there seem to be little correlation
between the size of impressions and the spacing of their arrangement around the rim, but
there is no obvious correlation between these measurements and the shape of the marks
in section, although there may be a slight tendency for narrower impressions to be more
often v-shaped. Under Specht’s proposed Buka-Watom model, we would expect to see
a positive correlation between mark width and mark spacing. This does not appear to be
the case for the Roviana data, which raises the broader question for Lapita and related
pottery styles whether the distinction drawn by both Specht and by Wickler is purely an
arbitrary one, where occasional outliers with coarse decoration are being classified as a
separate notching type, in which case early notching is simply more varied than later.
Metric Data on the size and spacing of single-band impression on the inner
(interior) edges of lips displays moderate positive correlation. Larger marks tend to be
377

more widely spaced, while smaller, more closely spaced marks are almost always v-shaped
in section (Figure 97). This latter pattern is likely to be a matter of tool choice: if the
potter wished to execute a fine, closely spaced pattern of notching, then the edge of a
fingernail or other fine object was used. This finer decoration was dominated by the
Honiavasa site, with occasional examples from other sites too. This prevalence of the
Honiavasa site in these data raises the question whether this attribute is a useful temporal
marker. An argument will be put forward in the section below that it is not, that location
of notching on the inner lip is related to lip orientation angle rather than directly relating
to time.
A similar scatterplot analysis of the size, shape and spacing of single-band
impression on the top/end face of lips is reported (Figure 98). Again a moderate positive
correlation can be seen between the variables, with a suggestion of a cluster of points from
various sites of small, closely spaced v-section marks. This scatterplot is the closest any
of the data come to showing a real distinction between fine edge-impressions and crenulate
lips, but again, the overall pattern is one of a moderate positive correlation between mark
width and mark spacing, with a variety of mark sections among the coarser expressions of
the pattern. These is no strong patterning by site, which would be expected if the pattern
was a reliable temporal marker. The virtual absence of this pattern from Gharanga and the
absence of Nusa Roviana from the plot are most likely sample size effects.
For sherds having double-band impression, on both edges of the lip (Figure 99)
metric variables seemed weakly correlated, and mark section was either v or u, with no
particular clustering in the plot. There was no significant grouping by site.
378

Figure 96: Lip impression, single band on outer edge of lip, labeled
by mark section: u=u-shaped, v=v-shaped, o=oblique v, w=w-shaped,
s=flat-bottomed groove.
Figure 97: Lip impression, single band on inner edge of lip, labeled by
mark section: u=u-shaped section, etc..
379

Figure 98: Lip impression, single band on top face of lip, labeled by
mark section: u=u-shaped, etc..
Figure 99: Lip impression, bands on both edges of the lip. Labeled
by mark section: u=u-shaped, etc..
380

Conclusions, Lip Notching/Impression/Crenation:
There is a hint in the Paniavile data in Figure 96 of the pattern of coarse impression on the
outer lip noted by Specht for Buka and Watom, and by Wickler for Buka, but the
Honiavasa site, which is so much more evocative of Lapita proper, for various reasons
discussed in Chapter 3 and Chapter 12, shows a pattern dominated by small impressions
along the inner lip, with a weak correlation between mark width and mark spacing.
Crenate lips, as a broad impression of the top of the lip, are not well supported as
being a distinct class of decoration. Taking broad impressions as a group, without
worrying too much about where on the lip they fall, there seems no pattern by site to the
shape of the impressions. Either v-shaped or u-shaped can be broad and widely spaced, and
crenation as a lip decoration grouping seems to have no strong structure in relation to site
in the Roviana data set. There is a continuum of variation between small impressions
closely spaced and broad impressions widely spaced for impressions located on the inner
lip and on the top of the lip, suggesting that a classificatory distinction between “crenate”
and fine, closely spaced impression would be arbitrary rather than natural.
Location of Lip Impression and Lip Form:
The Roviana data is more strongly site-structured by decorative location than by size or
spacing of marks. Impressions on the inner edge of the lip are common in the Honiavasa
sample, and occur in some numbers in the Hoghoi sample (Table 42). Is this homologous
similarity, which might place the Hoghoi punctate-neck/pinched shoulder pottery style (on
which the attribute occurs at Hoghoi) closer in time to the Honiavasa site than to the Miho
decorative style, or is this simple decorative technique merely an analogous similarity that
would incorrectly order a seriation?
381

Lip orientation angle is the sum of rim angle and lip angle, as shown in Figure 100.
While lip angle was easily measured for most sherds with a well-preserved lip, rim angle
was difficult to measure for all but the largest sherds, and the method used was not very
precise (precision varied with EVE, but rim angle was seldom measured with a precision
better than ±10 degrees. The difficulty of measuring rim angle must be borne in mind when
interpreting lip orientation data.
A comparison in terms of lip orientation between band of impression on the top
face of the lip, band of impression on the inner edge of the lip and band of impressions on
the outer edge of the lip suggests that lip orientation is influencing location of impressions
(Figure 101). Although sample sizes are small and rim orientation was measured with poor
precision, these data favour an explanation of the decorative similarity between Hoghoi and
Honiavasa samples (bands of impression on the inner edge of the lip) as analogous rather
than homologous. This interpretation is supported also by the occurrence of “bpi” pattern
on Gharanga-style sherds (discussed further below).
Figure 100: Calculation of lip orientation angle.
382

Decorative Attributes and Vessel Form:
Vessel form variability suggested subdivision of some decorative styles. Correlation
between vessel form and decoration was obvious in some cases (e.g. Form 1 carinated jars
and Class1 bounded linear motifs), but more difficult to approach quantitatively in others,
for example in the large group of everted-rim jars. This latter group was the focus of
investigation of the relationship between decorative variability and form variability. The
classes of sherd relevant to this vessel class were everted rims, neck sherds, hard shoulders
and soft shoulders. No instances of decoration on the body of such vessels was recorded.
Everted Rims:
Figure 101: Lip orientation and location of impressions.
In Chapter 8 a negative correlation between rim angle and rim height was suggested. Rim
depth is re-plotted below for everted rims of five decorative attribute combinations
(defined in detail below). Only two of the lip-rim-neck sherds with measurements
pertaining to this relationship had Class 2 linear motif decoration and/or a band of pinching
at the neck, a further indication of the fragility of lips associated with the Miho decorative
class (Figure 102).
383

Figure 102: Rim depth by decorative class.
Sherds with Gharanga/Kopo-class decoration, comprising a band of punctation in the
general region of the neck, and/or multiple bands of fingernail pinching on the shoulder
region, can be subdivided using differences in rim form, in that pinching of the shoulder is
restricted to a short-rim group, while punctation cross-cuts rim depth variability. If the
short-rim variants, with or without fingernail impression, are classed as a production style
on the basis of this attribute, many of the undecorated sherds would be incorporated into
the class, as would the three short-rim outliers in the wave-deformed lip class in Figure
102. This short-rim form, which commonly has the decorative attributes of
rim/neck/shoulder punctation and/or a matrix of pinching on the shoulder, was termed
“Gharanga style” in Chapter 3, after the site where it was first noted.
“Kopo style” is defined as those vessels with “Rimdepth” greater than 20mm, but
with a band of punctation in the neck region (here style is not used in the Dunellian sense,
but rather denotes a classification, as form is involved in its definition, which may have
384

a functional aspect). Many vessels of this style have weakly everted rims and only slightly
restricted neck, and may well be a contemporaneous functional variant of the Gharanga
style. The taller, more vertical rim of the Kopo style may have been more suited to liquid
transport, although access to the contents for stirring or ladling would have been less, and
the form would have been less suited to use of a torque or tongs for lifting, and less suited
to the use of a tied cover.
Wave-deformation of the lip (fourth from left in Figure 102) only overlaps in rim
depth with the Kopo style by three outliers. These sherds (from Gharanga site) may be
interpreted as homologous or phyletic links between the Gharanga/Kopo styles and wave
deformation of the lip (the latter attribute weakly associated with the Miho decorative
type). This is because Miho-style and Kopo-style vessels are functionally equivalent as far
as can be established from form, and are likely to be temporal stylistic variants of each
other. It follows that we might expect one to evolve into the other over time, and we might
thus expect crossover styles forming an evolutionary continuum. The rarity of such links
could result froms time-gaps in current sampling of variability.
The sample of sherds with wave deformation on which the neck is still represented
are typically tall, with one example of 100mmm “Rimdepth”. What cannot be seen from
Figure 102 are the lip-rim sherds that are incomplete, missing the neck, and thus not
measurable for rim depth.
Bearing in mind the relationship between rim angle and rim depth already
established in Chapter 8, the maximum measurable depths of incomplete decorated rims
was of interest given the small sample of complete Miho decorative type rims, those
measurable for rim angle and depth. (There are several indications of a fragile rim form
for this style, and any information pertaining to the form of what can be assumed to be
many missing rims is of interest). Incomplete rims of Miho decorative type (including any
385

Class 2 decorated rims that may have had wave deformation of the lip) yielded rim depths
>48mm, >60mm, >65mm, >79mm, >69mm, >81mm, >76mm, >45mm, and >62mm,
adding further support to the notion of a tall fragile everted rim form associated with this
decorative type.
Incomplete plain rims with wave deformation of the lip, which seem to have some
relationship to Miho decorative type in view of five occurrences of this lip decoration on
Class 2 incised lip/rim sherds, yielded the following minimum rim depth measurements: one
sherd had measurable rim angle of 20 degrees, with a rim depth >65mm, and a further six
tall incomplete rims yielded minimum depths of >26mm, >40mm, >47mm, >60mm,
>79mm and >65mm, consistent with the data from complete rims, and consistent with the
forms of Miho style decorated rims.
For lips decorated with a pinched band, found mainly at Honiavasa, rim depth
ranged from a minimum complete depth of 25mm (n=6) to a maximum of >65mm (n=11).
There was therefore no overlap in rim form with the Gharanga type, but overlap with the
Kopo and Miho decorative classes and associated plain-rimmed wave-deformed lips.
Forms of Plain Everted Rims:
Sherds comprising at least the lip, rim and neck for which rim angle and rim depth were
measured, and which were devoid of decoration, were plotted (Figure 103) and can be
seen to be a mixture of the production styles or types discussed above. The category of
plain everted rim cross-cut the distinction between Gharanga and other forms, suggesting
it was not a fine-grained category of stylistic or functional information, and should not be
used in seriations when the aim is high temporal resolution.
386

Figure 103: Rim depth and rim angle of undecorated liprim-neck
sherds.
Rim Form and Discontinuous Deformation of the Lip:
Discontinuous deformation of the lip was present on only ten everted rim sherds that were
complete from lip to neck, suggesting either a weakness at the rim/neck correlating with
this decorative attribute or an unrestricted vessel form. Rim depth for this sample ranged
from 21mm to 73mm (n=10), and rim angle for all measured sherds of this lip decoration
ranged from 15 degrees to 50 degrees (n=10). The small samples of observations suggest
that given an expanded sample there may be some overlap with the Gharanga form class,
but lack of association of this lip decoration with any classified decoration on other vessel
parts other than two examples of Class 2 unbounded linear motifs suggest that any such
overlap was probably minor. The tendency for these sherds to be thick, chunky sherds
suggests slab-construction in many cases.
Everted Rim Form and Bands of Impression on the Lip:
The four locational classes of lip impressed bands (bpi, bpo, bpt, bpb) were found in all
sites, and analysis of decorative covariation between parts above found that these variants
of lip impression were found in association with both plain sherds and with
387

punctate/multi-band opposed pinch decoration on rim, neck and shoulder. The presence
of these decoration classes in quantity in the Honiavasa site assemblage, from which
Gharanga/Kopo decoration was absent, suggested that rather than these otherwise plain
sherds being linked in some way to the Gharanga/Kopo decorative type, these were
probably a set of lip attributes with a broad functional-temporal spread, possibly of little
use for seriation. Sherds with this lip decoration were plotted labeled by lip decoration
class (Figure 104).
Although the pattern is not as marked for rim angle as it was for lip orientation
angle (Figure 101), bands of impression on more everted rims tend to be on the interior
edge of the lip, while on less everted rims impression tends to be on the outer edges.
Impression on the top face is found across the range of rim angles, while impression on
both edges ranges between 25 degrees and 55 degrees rim angle. On the rim depth scale
the data fall into two groups, the sub-20mm group, which is argued above to be the
Gharanga style of decoration, and the tall-rim group, in which a variety of decorative styles
can be found. Lip impression, because it cross-cuts such a broad range of other decorative
and form variability, is therefore regarded as likely to be analogous similarity, until proven
otherwise, and is tracked carefully in the seriations in Chapter 12.
Figure 104: Location of bands of lip impression in relation
to rim form variability.
388

Figure 105: Neck thickness comparison of Gharanga/Kopo
and Miho styles/types.
Neck Form and Neck Decoration:
The rim-neck-shoulder area of vessels is generally robust due to form strength and part
propinquinity. The Miho decorative class, most commonly identified from a single band
of fingernail pinching in the neck region, was often seen to have a thickened neck, as in
Figure 33. Gharanga-Kopo styles, on the other hand (defined as a decorative class by a
band of punctation in the general region of the neck and multiple bands, or a matrix of
fingernail pinching in the region of the shoulder), tended to maintain a fairly constant
thickness from lip to body. Metric data supporting this observation are given in Figure
105.
Neck thickness for Miho decorative type was on average about 2mm thicker than
for Gharanga/Kopo type. Median neck Vcurve for Gharanga/Kopo decoration style was
substantially smaller than for Miho decoration style (Figure 106). This information
suggests that while Miho-style rims are likely to be under represented due to taphonomic
389

ias, Miho-style necks are rugged as a result of form and thickness, and likely to be over-
represented relative to other vessel styles/parts. This is borne out by sample sizes of the
different form attributes. While only two Miho-style rims were complete from lip to neck,
the count of Miho-style necks in Figure 106 is 48.
Sherds which were plain on the rim, neck and shoulder, and thus not part of the
Miho style by definition (with the possible exception of some wave-deformed lips; but the
relationship between wave-deformation of the lip and Miho-style rims and necks is not
clear) were often of much larger neck Vcurve than sherds of Miho style. This difference
shows up in the median Vcurve of such sherds in Figure 106. Plain necks of large Vcurve
were especially common in the Honiavasa site.
Figure 106: Neck Vcurve for Gharanga/Kopo decoration,
Miho Decoration, and plain rim-plain neck-plain shoulder
sherds.
390

Chapter Summary:
Linear motifs were analytically classed as either those with linear zone markers or those
without (with an unclassified dustbin category also). This classification is robust across a
range of levels of brokenness, as the classificatory criteria are located at the edges of
design zones, these being the corner points of vessel profile, locations which preserve well
due to robust form. Patterns comprising formulaic expressions of repeated elements,
together with the two linear motif classes, formed the set of decorative patterns that were
used in exploratory analysis of decoration structure across vessels and in relation to vessel
form variability.
Class 1 motifs were largely confined to the Honiavasa assemblage, as was a lip
zone marker of a band of opposed fingernail pinches. Similarly, a band of applied nubbins
at the neck was virtually confined to the Honiavasa assemblage, as were sharply carinated
vessel forms (Form 1), which were almost exclusively associated with Class 1 linear motifs.
Deformation of the lip into a lateral wave occurred in two instances in the
Honiavasa assemblage (both of these were coarse finger-impression variants of the wave
form rather than lateral shear-deformation seen in most other sites). Lateral wave-
deformation was more prevalent in Nusa Roviana, Paniavile and Zangana sites, with a
small number of occurrences at Hoghoi and Gharanga. Wave deformation of the lip
generally does not appear with punctation, with rare exceptions. While punctation was
dominant at Zangana North, and Class 2 linear motifs dominated at Zangana South, wave
deformation did not show this spatial patterning, occurring in several instances in both
areas, suggesting possibly that wave deformation of the lip, while related to Class 2 linear
motifs, may have had an extended production span relative to Class 2 linear motifs.
Unlike the Honiavasa site, where the few observed instances of neck decoration are
predominantly a band of applied circular nubbins (and occasionally a band of single
fingernail impressions repeated), neck decoration of the other collections taken as a whole
is commonly a single band of opposed-pinch fingernail impression. Where this co-occurs
391

on sherds with classified decoration, the decoration is almost always unbounded linear
incised (Class 2 linear). The combination of wave-deformed lip, linear incised Class 2 rim
and shoulder (perhaps with some idiosyncratic applied decoration), and a band of opposed
pinch decoration at the neck, is the classic expression of what I have termed the Miho style
in Chapter 3 and elsewhere (Felgate 2002), but the attribute analysis in this chapter
suggests that a less normative approach as taken here reveals that attributes comprising
this style vary independently to some extent, and may, if this variation is temporal, each
have their own trajectories and rates of change.
Within the Miho style there is some pattern by site regarding whether Class 2
motifs occur on the rim or shoulder, with Class 2 motifs on the shoulder being almost as
common as on the rim at Paniavile and Zangana South, while they occur mainly on the rim
only at Miho (where Class 2 motifs occur on the shoulder it is usually motif C18 that is
present). This distributional information, where the spatially distant Paniavile and Zangana
South sites share a decorative similarity, while the relatively close Miho and Zangana
South sites differ, suggests there may be temporal variation within the Miho style, visible
in the details of attribute structure across the vessel. Accordingly, Class 2 decoration of
the rim and Class 2 decoration of the shoulder are treated separately in the seriations in
Chapter 12.
Miho-style attributes usually (but not always) occurred on excurvate rimmed
restricted pots (Form 6c), and a case was made for these having a rugged neck form but
a fragile rim/lip, leading to a taphonomic bias in part representation, where few complete
lip/rim/neck sherds have survived.
Punctation occurs predominantly as a single band in the general region of the neck
(sometimes on the adjacent rim, as though the vessel was decorated while upside-down,
and other times on the shoulder, clearly decorated while upright). Although a couple of
sherds of flared everted form had dual bands of punctation (these may have been
unrestricted bowls) the single punctate band motif generally occurred on medium-height
vertical-to short heavily everted rims, in the latter case often associated with multiple bands
of pinching on
392

the shoulder, forming a matrix of pinches, or a band of vertically oriented raised bars in
effect, rather like some Honiavasa applied decoration. The tall punctation-only vessels
were named Kopo-style, while the short-rim heavily-excurvate variants which sometimes
had the matrix of pinching were called Gharanga style. Gharanga/Kopo showed a tendency
for thinner necks than Miho style.
There are various applied decorations in all site assemblages, most of which seemed
idiosyncratic and therefore difficult to classify. This lack of clear patterning to the applied
decoration meant that in the main (with the exception of a band of circular nubbins at the
neck) it was omitted from consideration in seriation analysis in Chapter 12.
Homologous vs. Analogous Similarity:
Analysis of lip impression/notching variability suggested there was little basis for
regarding similarities between Honiavasa and Hoghoi in this regard as anything other than
analogous similarity in the absence of evidence to the contrary, although the effect of these
variables on seriations is considered in Chapter 12. By contrast, the Honiavasa pottery and
the Miho/Gharanga-Kopo styles all share what seems likely to be a homologous trait in
their neck decoration. In the Honiavasa site there were a number of instances of necks
being decorated with a single band of circular nubbins. The Miho style is characterized
as having a band of opposed pinching at the neck, which often has the appearance of a
band of raised lumps of clay. In the Gharanga and Kopo styles, these are replaced by a row
of punctations (almost always circular) made using a tool (access for the finger-techniques
of applied nubbin and opposed pinch is restricted by the Gharanga rim form, and this
could explain a change in tool used). Taken together, these three neck decorative attributes
account for almost all bands of decoration at the neck or in close proximity to the neck
393

(148 instances in total), and often occur on otherwise plain sherds, suggesting they are a
primary component of a Roviana intertidal-site pottery design system, and that the three
decorative techniques are three expressions of the same pattern, a row of circular elements.
The three expressions of this pattern are thus likely to be homologous similarity between
styles.
This design feature may be phyletically related to the eyes in the Lapita faces as
suggested by Spriggs (Spriggs 1990). Such an interpretation could be extended to suggest
that these are either the eyes of the pot personified, or the pot as representation of person
or persons, ancestral or mythical. The persistence of this design structure across the
variability of the various styles suggests a temporal persistence, or, in Dunnell’s terms, a
function, since the pattern is not drifting, only the technique of execution, at least at the
temporal scale of the sample. The rate of change of this feature of the design system must
be constrained, since it is so conservative compared to other design attributes.
Eyes might have a protective ideofunction; after all, the contents, which are in most
cases likely to be food, provide both sustenance and threat of poisoning, the latter either
through inadvertent contamination or through malice (the distinction between these may
be peculiarly modern; a result of modern microbiological knowledge). I note that previous
explanations for Lapita “faces” as reviewed in the opening chapters have tended to be
sociofunctional rather than ideofunctional in that there has been a strong emphasis on ritual
relating to the maintenance of Austronesian house societies and/or prestige exchange. Here
the suggestion is put forward that for the Roviana pottery at least, a more prosaic efficacy
may be being attempted, simply protecting the contents through ritualistic manufacture
including the application of decorative eyes. Where the house-ancestor and prestige
exchange theories suggest male manufacturers, an ideofunctional explanation of the sort
suggested here is more consistent with utilitarian production by women.
In this chapter a spatial division of style between Zangana North and Zangana
South has been introduced, which is examined in more detail in Chapter 11, prior to the
ceramic seriations in Chapter 12.
394

Introduction:
CHAPTER 10:
LITHICS
The main focus of this research has been ceramics, but the intertidal/shallow water lag
deposits of resistant materials have the potential to provide good samples of lithic artefacts
and manuports. The analysis of lithics had several practical applications in meeting the
overall methodological objectives of this research. Firstly, the form of lithic artefacts has
the potential to slot the Roviana materials into an existing Culture-Historical framework,
providing traditional and useful independent evidence for chronological corroboration,
however coarse.
Secondly, this study has a central aim of providing a methodological scoping
exercise for these intertidal sites: how might we proceed in seeking to write prehistory
from the materials at hand? The intertidal artefact/manuport scatters have a conspicuous
lithic component, due to the rugged nature of lithics. We have so little preserved to work
with that it is impossible to ignore the lithic component of assemblages in seeking to
develop applied archaeological method. The preserved lithic material can be used in
sourcing studies, in behavioural studies, and in taphonomic studies, and this chapter
provides pilot studies along each of these lines of enquiry. Substantive results are not earth
shattering, but the attempt is made in order to think about what sort of information might
usefully be acquired from these sites and how it might be used.
Although Reeve reports an obsidian point from Paniavile, and Wickler found
copious quantities of obsidian on the Lapita reef sites of Buka, no obsidian, and only two
small flakes of chert were found in the present study. This may be an observer bias, or a
different setting which alters the visibility of these materials, but the expectation that these
395

would be present was not met during extensive searching with the aim of locating obsidian.
Shell adzes have been numerous in some early ceramic sites (more than 200 from
AN-6 on Anuta) (Kirch & Rosendahl 1973:67). While there are many small Tridacna
adzes littering the surface of some late-prehistoric sites of the Roviana area, none have
been found so far from the intertidal sites, other than the three Tridacna adzes reportedly
recovered from Paniavile by villagers, illustrated by Reeve (Reeve 1989). The lack of these
in all recent collections may be either a preservation, observer, or visibility bias or a
behavioural difference between the occupants of the collection sites in the past (we cannot
distinguish between these possibilities on current evidence). In order to compare like with
like, in terms of raw material properties and manufacturing constraints, no comparisons are
made between the stone adzes recovered from the Roviana intertidal sites and the shell
adzes recovered from early ceramic sites elsewhere.
A section about water-rounded and fractured lithic manuports attempts a
petrographic classification based on a macroscopic type series. Petrographic results 1 allow
reclassification into broad source groupings. These groupings will be used in Chapter 11
to investigate whether manuport source classes are spatially structured at Hoghoi.
Additionally, manuport size frequency distribution is used to make some behavioural
inferences about procurement strategies and uses of these. The results are valuable in that
they raise a series of questions on these topics which show the value in focusing this sort
of attention on these materials. Also, many of the nondescript fragments that looked as
though they held little archaeological value individually turned out to be more exciting
when analyzed petrographically, and suggest that intensive analysis of all lithic materials
can pay major dividends for this type of site. A wealth of information content is there,
1 Petrographic descriptions were transcribed by the author from verbal descriptions
generously provided by Dr Robin Parker, of the Geology Department, University of
Auckland.
396

awaiting detailed analysis.
Formation processes are a major focus of the ceramic analysis. The lithics, like the
ceramics, can provide information on formation processes through brokenness,
completeness and remaining use life. For lithic artefacts, brokenness and completeness can
be used in the same way as ceramics, to estimate a breakage population, except that
artefacts like adzes have bilateral symmetry rather than circular symmetry about a central
axis. The difference between a breakage population and the extant deposit may inform on
formation processes if a sufficient sample is available to allow an estimate of breakage
population within useful limits. This was not the case with the sample of adzes obtained,
but an expanded sample would potentially allow this sort of approach. Remaining use life
can be assessed for those adzes that are recovered whole, by dimensional comparison with
unused preforms. Where the degree of resharpening is low, a substantial use life remains,
and if this is the dominant pattern, site abandonment or destruction by a sudden natural
disaster or attack is a possibility. Where adzes are all re-sharpened to stubby dimensions
with little remaining use life other than recycling to smaller tools, gradual discard through
unexceptional processes is more likely. Some rudimentary comments can be made on the
Roviana materials in this regard, as one unused preform from a local landowner/collector
was recorded.
Sourcing studies can only be taken so far within the limited resources of a thesis,
but petrographic descriptions given in this chapter illuminate a diversity of geological raw
materials. The geological origins from which these were sourced and the uses to which
these were put are limited by little more than the imagination at present, and it is hoped
that these data will demonstrate the desirability of making these materials the object of a
sustained program of research and cultural heritage resource management.
397

Review:
Formation theory suggests that unlike utilitarian pottery, stone axes may have long use-
lives and low discard rates (Shott & Sillitoe 2001:276), leading to an expectation that
sample size will be smaller for stone axes than for pots. It is assumed here that the same
holds for stone adzes. Turner’s discussion of adze use-life (Turner 2000: 231-296) does
not cover the relative representation of adzes and other materials in archaeological
samples, but rather focuses on the various use-life states of recovered adzes and
experimental adzes. The conclusion that, “Both experimental and archaeological data
support the probability that adzes had long use lives and underwent considerable
morphological changes as a result of intensive curation (Turner 2000:296).” supports that
assumption.
In line with this expectation, few stone adzes have been recovered from Lapita-age
sites in Near Oceania. Four are reported from site SAC at Watom (Green & Anson 2000b:
59), and “...a few flakes... (Golson 1991:255) are reported from Lasigi. There is no
mention of stone adzes in any of the Arawe Lapita site reports (Gosden 1989, 1991a, b,
Gosden et al. 1989, Gosden & Webb 1994). Few are reported from excavations in the
remote Oceanic islands of the southeast Solomons or Vanuatu, with the exception of the
Reef/Santa Cruz sample (much of which may have been surface collected, and excavated
area was large). On Anuta, for example, no stone adzes were found either from
excavations at AN-6 or from surface collections, while more than two hundred shell adzes
were recovered (Kirch & Rosendahl 1973:66). Thirteen stone adzes or flakes from adzes
were reported as analysed petrographically from the Lapita sites of the Reef/Santa Cruz
Islands (Green 1976a:259), although it is not known whether these are all from separate
adzes; the number of adzes represented by that sample may be less than thirteen. There are
also a number of other as-yet unreported stone flakes (most with signs of grinding) in the
398

Reef-Santa-Cruz collections, some of which are clearly fragments of adzes (Sheppard
2003: pers. comm.). Apparently no stone adzes were recovered during the extensive
program of survey and excavation of Lapita and post-Lapita sites on Mussau, as there
seems to be no reporting of any such items in the most recent volume on the subject,
although no conspicuous mention is made of this absence in the report (Kirch 2001).
In Vanuatu, the best Lapita samples at present are from surface sites on Malo.
Hedrick found two stone adzes here (one plano-lateral and one plano-convex) (Hedrick
n.d). Ten shell adzes were recovered from the Atanoasao site on Malo, seven from surface
collection and three from excavation, but no stone adzes were found (Galipaud 1998).
Recent intensive testpitting on a number of early Vanuatu sites (Bedford 2000: 205-206)
recovered one Tridacna adze from Bedford’s early period (3000-2800bp), from the
Arapus site, while a number of shell adzes have been recovered from later contexts,
together with a single rectangular-sectioned stone adze from Bedford’s 2800-2500bp
period.
In contrast to this dearth of evidence from excavated Lapita sites on land in island
Melanesia, there has been a relative flood of stone adzes and flakes from reef sites in Near
Oceania, which may say more about the nature of the sampling process than any behavioral
differences (large areas are being sampled by the surface collections, as done also by Green
in the extensive surface collections and area-excavations in the Reef-Santa Cruz sites).
Three complete and four fragmentary stone adzes were recovered from Paniavile by local
collectors (Reeve 1989:55), which were either quadrangular or Plano-convex in cross
section.
Green and Davidson’s Samoan Adze typology (Green & Davidson 1969) as
recently modified by Wickler for a relatively large Island Melanesian sample (Wickler
2001:181) provides the classificatory units for Wickler’s study of Buka stone adzes. Green
and Davidson based their Samoan typology on previous schemes of Buck’s and Poulsen’s,
399

ut with more emphasis on cross section attributes, rather than the emphasis on attributes
of the butt, lugs, or curvature of the back as in Duff’s scheme (Green & Davidson 1969).
Green and Davidson’s type IVa and type V were characteristic of early Samoan ceramic
sites, and were rare in surface collections. These are similar in cross section, being flat on
one surface and convex-to-trapezoidal on the other, but type IVa is sharpened with the
edge at the flat side, producing a flat cut, while type V is sharpened with a horsehoof -
shaped edge along the convex side, which produces a gouge-shaped cut.
Green and Davidson’s types VI, VII and VIII are all triangular-sectioned,
sharpened either way to produce either a narrow straight edge or a narrow gouge, and
some of these are large and heavy, but many are small and fragmentary. Type VI was
thought to be present throughout the Samoan sequence. Type VII was a narrow deeper
version of type V, with a prominent narrow gouge-shaped cutting edge, as though for
producing grooves or rebates, and was found only in early contexts. Type VIII is rare in
the Samoan sample, and although only known from surface contexts, was inferred to be
an early form from its rarity. This was a heterogeneous type, most subtypes of which had
a straight cutting edge and a triangular -sectioned back, and were a broad triangle in cross-
section (broader than type VI).
Wickler looked at 18 adzes and 19 triangular-sectioned flaked stone tools from the
Buka reef sites. Wickler departed from Green and Davidson’s typology in that fragments
of triangular-sectioned flaked stone tools were classified separate to the adzes (Wickler
2001:175, 177), when these may well have been broken fragments of small type VI adzes.
Wickler describes ten of these as flakes of dark green lithified mudstone, and nine as
andesite, both materials used in the manufacture of the adzes in that sample. Five of these
had ground or polished margins. Both factors suggest that at least some of these fragments
could be classified as small Green and Davidson type VI adze fragments. Whether the term
adze is appropriate to these small narrow forms or whether they were used as chisels
400

is unimportant in the current context, where we are dealing with small fragments only, and
are limited to discussion largely of differences in cross-section.
Roviana Lithic Artefacts:
Lithic artefacts from the intertidal included seven relatively complete adzes (Figure 107,
Figure 108, Figure 109, Figure 110), a number of small fragments of what were probably
adzes (not illustrated), two chert flakes, a worked sandstone slab, notched as though for
an anchor and a number of small cobbles with depressions on the edges and ends, semi-
cubic in form, that are interpreted as Canarium hammers (Figure 111). These differ from
an ethnographically known hafted nut hammer found widely on land in the region, and are
identified by the use-indents on the six faces of the cube. They are typically composed of
a relatively soft rock that is inferred to have indented easily with use, compared to the
hafted variety, which is not to say these were too soft to be functional.
Petrographic descriptions of lithic artefacts are given in Table 43. These data
suggest a diversity of lithic resources were exploited for the manufacture of artefacts,
including at least two sources of adze rock, one represented by A.12, A.31 and A.6, the
other, more diverse group represented by the remainder of the adze fragments. The second
petrographic grouping includes substantial differences in the surface appearance of the
stone, with one adze and one fragment from Hoghoi being glossy black in appearance,
while other fragments from Zangana and Miho within this broad petrographic grouping
were more weathered in appearance, in a variety of hues. These surface differences suggest
that if further analysis was carried out to characterize diagenesis within this grouping it
could be further subdivided on a systematic basis.
401

Figure 107: Planilateral sectioned adze from Hoghoi initial surface collection (top),
plano-convex-sectioned type V adze from Zangana (middle), and planilateral adze
fragment from Zangana (bottom).
402

Figure 108: Images of the adzes shown in the preceding illustration.
403

Figure 109: Green “type VI” or “type VIII” triangular-section adze fragment from
Miho (top); butt-end of a plano-lateral-sectioned adze from Zangana South (middle)
and a fragment of a trapezoidal-sectioned Green “type IV” adze from Zangana.
404

Figure 110: Images of adzes illustrated on previous page.
405

Figure 111: Canarium hammerstone from H5 ceramic findspot (top) (a similar artefact
was found at Hoghoi); waisted sandstone slab from Zangana (middle); and chert flakes
from Hoghoi (bottom).
406

Table 43: Petrographic descriptions of lithic artefacts.
Artefact ID Sit
e
B1.T50 10-
15
Unit Description
8 13 In hand specimen a pale green-grey fine-grained adze fragment. In thin section a fine-grained sedimentary or volcanic
rock. It is impossible to be more determinate without further analysis, as recrystallization of sedimentary clasts and
primary volcanic crystallization could have the same petrographic result.
HG.17.1 2 17 In hand specimen a glossy black stone, being a small fragment of an adze with polishing visible on two faces. In thin
section this shows a fine-grained texture with sedimentary banding, with a possibility also of recrystallization.
HG.99 2 99 A complete adze in a glossy black stone, similar in appearance to HG.17.1. This adze is reminiscent of Green and
Davidson’s type V, occurring only at the early end of the Samoan sequence [Green, 1969 #375:25] In thin section,
sedimentary or volcanic: comprising fine material in a very fine-grained matrix, with some evidence of feldspar
recrystallization, with clay-rich silty minerals perhaps present as well.
B3.T10 5-10 8 89 Mid-grey fine-grained appearance in hand specimen, weathered to a green-grey on the surface. In thin section coarser
grained with micaceous clay precursors, dominated by plagioclase, with some Pyroxenes.
A.12 8 48 In hand specimen a dark grey rock with a dull but generally unweatherd surface, being the sharpened end of a thin flat
adze. In thin section a sedimentary rock, coarser grained. diagenesis indications include feldspar recrystallization and
clay precursors. There is no obvious calcite, and the matrix is dominated by plagioclase feldspars. Noted for
recrystallized opaque pyrite.
A.31 8 94 In hand specimen a complete but heavily resharpened plano-convex “horsehoof”(Green and Davidson type 5) or
gouge-edged adze in a glossy brown-grey stone (mid-grey when sawn). In thin section a coarser-grained sedimentary
rock, similar to A.12., with feldspar dominant, diagenetic recrystallization of a variety of minerals, including opaque
pyrite.
A.6 8 23 In hand specimen the midsection of an adze, probably plano-convex, with surface weathered to a dark brown-grey,
and mid-grey when sawn. In thin section the same as A.31 and A.12.
B1.T50 0-
5m
8 12 A small grey rock fragment without obvious polished or ground surfaces. Mid-to-dark-grey when sawn. In thinsection
it is unclear whether this is sedimentary or volcanic in origin, due to an advanced stage of recrystallization,
making it difficult to see original textures. Dominated by plagioclase feldspar in various stages of recrystallization.
Fine material also looks feldspathic, but includes a pyroxene/Ferromagnesian component. Pale-green chlorite is
possibly present. Texture possibly indicates a plutonic origin, but this is tentative, and would require further analysis
and thought to characterize this with more confidence.
Miho 3 1 butt-end of a triangular-sectioned or fractured adze? Lightly ground or water rolled? Surfaces are dark-brown-grey,
with sawn surfaces being mid grey to dark grey. This rock is a lot fresher than “B1.T50 0-5", with only slight
recrystallization. Fine-grained plagioclase crystals are set in an even finer-grained matrix of recrystallized feldspathic
character. This might be an immature, poorly-sorted sedimentary rock, or a crystal-rich igneous rock with a low
carrying fluid content. XRF analysis would clarify whether chemistry indicates volcanic or sedimentary origin and
diagenesis. X-ray diffraction would show up any quartz, clay minerals or calcite.
P199 1 1 In hand specimen a thin small flake of fine-grained stone weathered to a golden-brown color, with light-grey sawn
color. Possibly reworking debitage? In thin section this has a very fine-grained volcanic matrix carrying scattered
plagioclase and pyroxene phenocrysts.
HV.04.141 4 4 A flat sandstone slab 10x7x1cm, weathered to a golden-brown color on the surface and light grey when sawn. In thin
section this is well supplied with pyroxenes, plagioclase and quartz, these being immature, angular, poorly sorted,
with a silty clay cement. No Calcite. This is a packed stone with larger grains forming a self-supporting framework.
Clinopyroxene/augite are present(a green pyroxene is probably augite). Plagioclase can show simple and multiple
twinning as well as zoning. Amphibole and hornblende are present as well as a fine-grained foraminiferous silt clast.
B(Zangana) 8 B In hand specimen this was a fin-shaped sherd of sandstone weathered to a golden brown, about 12cm maximum
dimension, with a light grey color when sawn. In thin section a predominantly calcite cement with occasional
unfragmented foraminiferal shell, suggesting calcite is a direct perecipitant, or a biogenetic calcite fine fraction, or
recrystallization. The matrix is dominated by silt-sized glass fragments and some crystal products( feldspars including
plagioclase, pyroxenes, and opaques). This was interpreted a s a cemented volcanic ash dominated by glassy
fragments, with crystals co-genetic with the glass, which is bubble-fractured by expanding gasses. Microprobing the
feldspars and glass or grinding/acid-washing/XRF would identify the type of volcanism involved.
Z76 8 76 A coarse grey sandstone slab with pedestalled black mineral grains, 16x12x2cm and subrectangular in form, with two
notches forming a waist, interpreted as a sandstone grindstone modified for secondary use as an anchor stone. Sawn
colour matches the surface colour. In thin section fill of immature, angular, unsorted pyroxene, plagioclase and augite
up to 1mm in length in a calcite cement. Fine-grained silty volcanogenic clasts are also present, which sometimes
display fragments of coarser pyroxene and plagioclase. No confirmed quartz.
It seems clear there are diverse resources being used over the period of deposition of these
materials. Additional samples, more concerted analysis, and discovery and
407

characterization of source quarries are needed to explicate patterns of lithic raw material
transport. Adzes/fragments from Zangana of recrystallized fine-grained rock showed a
variety of surface colours, including a pale greenish example. This suggests that raw
material procurement may have been influenced by surface color, as there is some
congruence here with the green adze rocks noted by Green for the Reef-Santa Cruz Lapita
sites, the green plano-convex example from Samoa (Fijian origin?), the adze in green
alteration dacitic welded tuff from Tonga, and green meta-sedimentary adzes from Naigani
(Green 1979, 1994).
A number of adzes and axes from Paniavile held in the private collection of Mr Sae
Oka were photographed during the course of the study (Figure 112). Some small waisted
axes and one unfinished adze preform were reportedly collected from Gharanga, by the late
Mr Phillip Lanni (Figure 113, Figure 114). The preform is much longer than the complete
adzes, while cross-section size is similar to the other adzes. If the preform is any indication,
the adzes collected are less than half their original length, suggesting a lengthy use life
involving a massive grinding resharpening labour or possibly more rapid
reflaking/resharpening to repair edge damage.
Sandstone artefacts/manuports similarly suggest a diversity of sources, as yet
poorly represented in the collections. There is clearly vast scope for artefact studies and
lithic sourcing research for this type of intertidal ceramic/lithic scatter. HV.04.141 (Table
43) is unlikely to be from the New Georgia group due to the abundance of quartz in the
sand-sized component, as sediments rich in quartz should not occur given the geological
characteristics of the area, although some forearc sediments have a plutonic component
which might be relevant here (the nearest such source would be southern Rendova or
Tetepare). The cemented volcanic ash (sample B-Zangana South, Table 43) might
originate within the New Georgia group, as rocks originating as volcanic ashes are present
on Ranongga, and possibly Southern Rendova or Tetepare.
408

Figure 112: Artefacts photographed from Oka collection and reported to be from the
Paniavile site: shell and stone adzes (top); stone adzes (middle) and waisted
tools/weapons and a pineapple club (bottom).
409

Figure 113: Waisted axes photographed from the Lanni collection, courtesy of the late
Mr. Phillip Lanni, found in the vicinity of Gharanga Stream.
410

Figure 114: Un-ground adze preform photographed from Lanni collection
courtesy of the late Mr. Phillip Lanni, reportedly found at the Gharanga
site.
Sample Z76, from Zangana (the waisted sandstone slab in Figure 111), could conceivably
be from Ranongga/Rendova/Tetepare, as the forearc region has had a complex history of
volcanism, uplift of the forearc region followed by subsidence, associated with erosional
features, and subsequent rapid and accelerating uplift in response to subduction of the
Coleman seamount (Mann et al. 1998). These processes have created a variety of
siltstones, mudstones and sandstones of the Tetepare Formation.
Chert Flakes/Fragments:
Two chert flakes were found at Hoghoi (Figure 111). One (HG.20.22) was thin sectioned,
411

and identified as microquartz 2 , but did not have any microforaminifera that might narrow
the range of options for a source region. It is likely to be exotic to the New Georgia group
as cherts are not expected there. The nearest known sources of chert are in Vella Lavella,
where it occurs in minor amounts (Sheppard, pers. comm.), Santa Ysabel, Nggela or
Malaita (Sheppard 1993:124).
Analysis of Hoghoi Water-rounded and Fractured Volcanic Manuports:
A transect-collected assemblage of 345 lithic manuports from the Hoghoi site was
described in the field by matching to a type series of 23 stones or stone fragments. The
characteristics of degree of fragmentation (water rounded, fragmented with rounding, or
fragment with no water-rounded cortex visible), type, and mass (to the nearest 50g) were
recorded for each stone. These data are given in table “Hoghoi_lithics.db” appended on
CD. The type series was retained and petrographic thin-sections were prepared for each
type. A combination of physical properties in hand-specimen and optical petrographic
properties were used to reclassify the type series as shown in Table 44.
Porphyritic olivine basalt is the most widespread rock type in the New Georgia
group and occurs on all of the main islands except Simbo (Dunkley 1986:13). Classes 1,
2, 10, 12, 16 and 18 fall into this category. The nearest source is Rendova, lying about
12km distant across the Blanche channel, and this is a major source of ovenstones for the
Roviana lagoon area in modern times. A slightly longer but more sheltered trip 15km east
of Hoghoi can locate similar stones in mainland riverbeds draining into the Lagoon.
Class 15 is a feldspar-phyric basalt or andesite, another important rock type that
covers large areas of New Georgia, Vangunu, and southern Rendova. The nearest source
of this rock type to the Hoghoi site is the Hura River (Dunkley 1986:22), about five km
distant across the Lagoon. Vesicular and scoracious lava flows are present in the Picritic
lavas eroding into the Viru Harbour area and Marovo Lagoon (Dunkley 1986:16), and
2 Cherts were examined petrographically by Dr Peter Sheppard of the
Anthropology Department, University of Auckland.
412

Classes 4 and 5 might be from such a source, which would involve transport from Viru
Harbour at a minimum, a distance of about 50km. Kolombangara, at a similar distance
from Hoghoi, is also noted by Dunkerly as having scoria cones. Closer sources are not
precluded on current evidence. Class 6 has similar mineral composition to Class 5, and may
be from a similar source.
Class 7 is an olivine basalt with olivines serpentinised rather than altered to
iddingsite or magnetite, suggesting a magnesium-rich magma. Sepentinised olivines are
noted in the gabbros of the Choe intrusive complex (Dunkley 1986:31), but whether this
means volcanic rocks in that area are more likely to show serpentine rather than iddingsite
as the alteration product of olivine is unknown, but this suggests a specific question for
source sampling in the future. If it does turn out that serpentine is characteristic of Viru
harbour rocks, then the Class 4 and 5 scoriacious manuports, Class 6, and Class 7 could
all be a pointer to procurement of a minority of manuports at a greater distance than is the
norm today. Scoracious rocks in particular might have a non-cooking function, as their
heat capacity is low. That function could relate to their relatively low specific gravity, and
surface roughness which might confer desirable properties as net weights or possibly
abrasives.
Looking only at those manuports that were water-rounded and unfractured in form,
a comparison of size (average weight) by class is given in Table 45. Some classes appear
mainly as small cobbles or pebbles of less than 120g average weight (for example the
large Class 18, and also classes 3, 5, 7, 10 and 16), while other classes average more than
240g (Classes 1 and 2). This is most likely a slight difference in the survival rate of larger
stones if used for cooking (i.e. Class 18 is more likely to shatter than Class 1) but the
larger number of complete water rounded stones in Class 18 is a bit of a mystery, unless
these were preferred for some other function, such as fishing or net weights. Perhaps this
is just sampling error, as the number of occurrences of complete water-rounded stones is
low for each class.
413

Table 44: Petrographic classification of manuports collected at Hoghoi.
Sampl
e #
count class Description
HG1 67 1 Medium grey color, phenocrysts visible to the naked eye. In thin section a groundmass of Plagioclase laths and fine opaques
with Pyroxene phenocrysts and a brown-rimmed alteration mineral (probably Iddingsite).
HG2 36 2 In hand specimen a pinkish rock with pale green phenocrysts. In thin section the groundmass was predominantly Plagioclase
laths with subordinate Olivines altered to Iddingsite. Phenocrysts were large Pyroxenes. No Feldspar phenocrysts.
HG3 21 3 Fine grained spherulitic texture, a brown glassy matrix carrying scattered large Feldspar phenocrysts as well as Feldspar
elongate quench crystals, giving stong indications of rapid cooling forming glass followed by recrystallization from a series
of points. A large crystal of Calcite is present, pseudomorphing pre existing Olivine or Pyroxene.
HG4 21 4 Pinkish in hand specimen with small vesicles. In thin section a groundmass of medium-textured Plagioclase and Pyroxene
crystals.
HG5 26 5 In hand specimen pinkish, fine-grained, soft to saw. A dense, fine-grained texture with some vesicularity. Reddish
groundmass, with mainly Plagioclase phenocrysts and some Pyroxene.
HG6 1 6 In hand specimen pale grey to pale green and fine-grained in hand specimen. In thin section this had a very fine-grained
Plagioclase groundmass with Plagioclase phenocrysts, similar to HG5 but without the vesicular texture.
HG7 10 7 In hand specimen grey, medium grained, even textured. In thin section a plagioclase groundmass and abundant pyroxene
phenocrysts were noted, and also an alteration product of Olivine, Serpentine, indicating a relatively magnesium-rich
magma.
HG8 2 1 A coarse-grained grey rock in hand specimen, and in thin section a coarse volcanic Plagioclase groundmass with abundant
Pyroxene phenocrysts and smaller brown alteration mineral, possibly Iddingsite.
HG9 13 9 Grey and medium-textured in hand specimen. In thin section a very fine groundmass of unidentified mineral grains with
large phenocrysts of Plagioclase and smaller Pyroxene phenocrysts.
HG10 18 10 In hand specimen a dark-grey rock with greenish phenocrysts. In thin section a fine grained volcanic matrix, some small
crystals of an alteration mineral, possibly Iddingsite, with phenocrysts of Pyroxene and Plagioclase.
HG11 15 5 In hand specimen a soft pinkish fine-grained rock, and in thin section a dense, fine-grained texture carrying large phenocrysts
of Plagioclase and Pyroxenes, with a possible Olivine alteration product in the groundmass.
HG12 12 12 In thin section a coarse-grained phenocrystic volcanic carrying Clinopyroxene, Augite, occasional Orthopyroxene
(Hypersthene), red Iddingsite. (This alteration product of Olivines occurs in Iron-rich magmas and could result from use of
heat as oven-stones).
HG13 6 5 In hand specimen soft, pinkish and fine grained. In thin section a very fine-grained volcanic groundmass with mostly
Plagioclase phenocrysts.
HG14 4 14 A fine-grained light-grey water-rounded pebble. In thin section there is dense accumulation of Plagioclase and Pyroxenes in
layers, could be plutonic. No Olivine. Fine-grained matrix suggests volcanic.
HG15 10 15 Huge Feldspar/Pyroxene phenocrysts (some 1cm) in a matrix of fine Plagioclase. Some brown-rimmed smaller phenocrysts
are possibly an altered pre-existing Olivine.
HG16 14 16 In hand specimen a dark grey medium-grained rock similar to HG7. In thin section a matrix of very fine Plagioclase contains
subordinate medium-sized Plagioclase phenocrysts and dominant large Pyroxene phenocrysts.
HG17 7 17 In hand specimen a very-fine-grained dark grey angular fragment, in thin section a very fine grained volcanic matrix carrying
scattered phenocrysts of Plagioclase and a dark brown/red alteration mineral, with occasional Pyroxene phenocrysts (see also
GW8 in ?), interpreted as a fragment of adze rock.
HG18 57 18 Dark grey in hand specimen with greenish phenocrysts, similar to HG 7 and HG 16. In thin section a medium-textured
Plagioclase groundmass with Pyroxene phenocrysts and a lot of medium-sized opaques. Some brownish minerals that are
possibly an alteration mineral of Olivine or Calcite.
HG19 6 19 Vesicular in hand specimen with pink-to grey color zonation (heat alteration rom use?). In thin section a very fine
groundmass (mineral not identified) with rounded red-to-brown phenocrysts filling some vesicles.
HG20 4 2 In hand specimen dark grey and coarse-grained, and in thin section a coarse-grained texture, predominantly plagioclase
groundmass with large Pyroxene phenocrysts. Large plagioclase is absent. Iddingsite phenocrysts present
HG
“incise
d”
pebble
HG
40-45
pound
er
1 3 As below, except texture is slightly coarser and includes pyroxene phenocrysts.
1 3 In thin section a fine grained sperulitic texture, a glassy matrix carrying scattered large Feldspar phenocrysts as well as
Feldspar elongate quench crystals, giving strong indications of rapid cooling forming glass followed by recrystallization from
a series of points.
414

Table 45: Size comparisons between petrographic classes of lithic manuports.
Class Count of class Sum of mass (g) Mean(St.D.) mass (g)
1 15. 3600. 240(252)
2 10. 2600. 260(353)
3 17 1300 76(26)
4 2. 300. 150(71)
5 8. 825. 103(123)
7 4. 400. 100(71)
10 7. 550. 79(27)
14 3. 575. 192(267)
16 9. 1075. 119(146)
18 31. 3650. 118(188)
Taking water-rounded (complete) cobbles of all classes together (Figure 115), there is a
suggestion in a histogram of sizes that stone collecting strategies operated between the
limits of roughly 50g ( the scale used was not very accurate) and 1150g, although the
lower limit may reflect an archaeological collection threshold, and the upper limit is
imprecise due to small sample size of larger complete stones (heavier stones than this
were present among the fractured cobbles, with the largest stone being 1900g, with mean
mass for all stones collected being 200g, and Standard deviation of 212g). Most of the
larger stones within these limits would have been fractured by heat, but the smaller stones,
from 50g to 150g, seem to have been either not subjected to heating or have survived
better. These smaller stones are an inconvenient size for domestic cooking from the
perspective of the modern local practice of handling stones individually with bamboo
tongs, and may have had another function, such as net weights or fishing weights, for large
nets such as turtle or dugong nets, which required heavier weights due to their added
buoyancy.
415

Figure 115: Water-rounded lithic manuports, cortex
complete.
Figure 116: Size distribution of fractured lithic manuports,
either with some cortex or without.
416

Ethnographic cooking practice need not reflect those of 3000 years ago though.
If the weight frequency distribution of complete stones (Figure 115) is contrasted
with the weights of fractured stones (Figure 116) it seems that the complete recovered
examples underestimate the upper weight limit of procurement, as some fractured stones
approach 2kg in weight.
The spatial distribution at Hoghoi of the petrographic classes defined in this section
is discussed in Chapter 11.
Chapter Summary:
Lithic manuports systematically surface collected at Hoghoi included a diversity of volcanic
rocks, with coarse porphyritic textures and cortex suggestive of origins as transported
water-rounded cobbles. Sizes in excess of 2kg were targeted in some cases, but there was
no evidence for any rocks larger than this. These coarse textures, and the fractured forms
of most rocks suggests a cooking function. Some fine-grained sperulitic-textured
fragments with quench-textures suggest rocks poorly suited to oven-stone use (liable to
explode?), and some of these may be fragments of flaked artefacts. At least one fine-
grained recrystallysed sedimentary or volcanic fragment in the manuport collection is likely
to be a fragment of an adze, as petrographic characteristics matched those of some of the
adzes collected.
A number of smaller porphyritic water-rounded pebbles were collected, which do
not fit with local ethnographically observed oven-stone size selection. These may be
evidence of other non-cooking functions, such as net weights or fishing sinkers, or possibly
slingstones even, or may indicate a shift in cooking practices over time. The presence of
scoracious basalts is inconsistent with a cooking function also, and these may originate
from Viru or Kolombangara, although more local sources may exist, as yet
417

unreported. While most olivine basalts are most likely from Rendova, serpentinised olivines
may indicate different sources, but this has yet to be tested.
A diversity of adze materials indicated the potential for sourcing studies to
illuminate raw material procurement patterns for this site type. At least one plano-convex
horsehoof-edged adze form was recovered from Zangana, regarded by Green as indicating
a homologous link with Lapita at this site, as these forms are characteristic of the Lapita
horizon and derivatives (Green, pers. comm.). Comparison of the complete adzes with an
unused preform suggests the complete used examples have been heavily resharpened to
less than half their original length prior to discard, suggesting little remaining use life.
Although the sample of such items is small, it does suggest accumulation from gradual
refuse discard rather than discard as a result of sudden or catastrophic abandonment of
sites or extinction of inhabitants. The presence of a number of small fragments of adzes,
identifiable only through raw material characteristics, suggests either a taphonomic regime
capable of smashing these rugged artefacts, or reworking of adzes into smaller artifacts,
damning any notion that the observed quantities of even rugged lithic artefacts can be read
as directly measuring the intensity/duration of discard.
Other lithic items such as sandstone fragments and artefacts are not always easy to
find among coral gravels, but the diversity of sources indicated by petrographic analysis
suggests these can be a major component of sourcing studies, and are worth putting more
effort into finding and analysing.
This chapter demonstrates that archaeological values of this type of site lie as
much in the tiny nondescript lithic fragments as in the complete artefacts. One is of little
value without the other. The complete artefacts alone would mislead if read as the
quantities and diversity of adzes or whatever discarded in the past. Similarly, the
fragments alone would be difficult to recognize as artefact fragments without petrographic
analysis of both these and the complete or identifiable adzes. Intensive investigation of
lithic variability can unlock the cultural and postdepositional formation processes of these
418

sites. It is clear from this lithic analysis, as it was from the various ceramic chapters, that
the information content of these sites is fragile as a result of this interdependence: take the
fancy bits away and it gets harder, if not impossible to understand the fragments, and the
fancy pieces alone do not give an accurate picture of site formation.
The raw materials and forms of the Roviana flaked and polished stone adzes
recovered from the intertidal sites are very similar to the materials recovered from the
Buka Lapita sites on the reef flat. This similarity is so marked as to suggest that detailed
petrographic and morphological comparisons would be justified, although it was not
possible to do this within the time-constraints of the present study.
Comparisons between the Buka sample and the Roviana sample of adzes suggest
these are expressions of a very similar range of adze forms and selection of raw materials,
most economically explained as homologous similarity. This supports the notion of a
Lapita/post-Lapita stone adze repertoire, if not an adze kit, which is not to say that there
was no change over time or space, but which acknowledges that given the limited current
sample, particularly for near-Oceania, a coarse stone adze horizon can be seen, which
probably occupies a broader temporal span than elaborate dentate-stamped pottery, and
serves to link the “post-Lapita” Roviana intertidal sites to the Lapita ceramic series by
homologous similarity of the lithic adze component of sites.
The near-Oceania Lapita-horizon stone adze sample is now dominated by the items
recovered from the Buka/Nissan reef sites and the Roviana intertidal sites. This suggests
that further sampling of this settlement type holds the prospect of major dividends for this
rare class of artefact, and that a large sample attained by this means can be expected to
enable detailed sourcing and variability studies to proceed in the future, to an extent never
before possible in Near-Oceania due to sample size constraints.
419

420

Introduction:
CHAPTER 11:
INTRASITE SPATIAL STRUCTURE OF
CERAMIC AND LITHIC VARIABILITY:
Intrasite spatial structure is here defined as the patterned distribution of potsherds,
artefacts, and/or manuports across sites (Wandsnider 1996). Intrasite spatial structure in
the Roviana intertidal scatters is assumed to result from processes such as patterned
discard (which may or may not have included contemporaneous differentiated activity
areas), wave sorting of archaeological materials, selective trapping and preservation of
archaeological materials by site micro-topography and/or postdepositional pattern-
alteration by humans.
Site occupation span can be expected to have varied from site to site, and sites in
favourable locations may have been occupied more than once. Even within a continuous
occupation, it seems likely that the extent and locus of settlement may have changed over
time, the latter either randomly, or possibly through unknown socially-determined
mechanisms such as the avoidance of places formerly used by the deceased. Such processes
should be expected to create, over the generations, chrono/stylistic spatial structure within
sites. Ultimately the process of distinguishing whether stylistic spatial structure represents
chronological change in style or contemporaneous diversity is one that occurs in dialogue
with one’s ideas on ceramic classification and the nature of the stylistic series (Baxter
1994:23). For a poorly-known sequence like the Roviana one, with no well-established
fine-grained stylistic chronology, any conclusions must be regarded as preliminary working
hypotheses.
421

Review of Methods of Spatial Analysis:
While intersite distributional mapping of ceramic styles or types within a geographic
approach was common in European archaeology (e.g. Hodder & Orton 1976)and has been
applied in the Pacific (Irwin 1985), ceramic intrasite spatial studies have been relatively
rare in archaeology (Fontana 1998) compared to intersite or distributional studies, many
examples being student dissertations ( e.g. Henderson 1998, Henderson 1992, Robertson
2001). There are notable exceptions to this tendency, some local (Best 1984:594-617,
Cowgill et al. 1984, Fontana 1998, Sheppard & Green 1991).
During the 1980s, debate in intrasite spatial analysis centred around spatial
clustering and the methods by which archaeologists could identify spatially clustered
patterns (Baxter 1994:23-24). LuAnn Wandsnider noted that emphasis has shifted in recent
studies from a functional view of site structure in which activity patterning ought to be
easily read and toolkits easily identified spatially (e.g Blankholm 1991), to a broader and
more complex formational view of spatial structure, where the interaction of both cultural
and natural formation processes are acknowledged as responsible for the structure of
archaeological deposits, and the identification of specific formational processes has become
the goal (Wandsnider 1996). Congruent with this shift has been a shift in method from
partitive typological approaches seeking to characterized artifact clusters within sites as
functionally discrete to a more inclusive set of methods that examine distributions of a suite
of formationally sensitive attributes of artefacts across sites. Although Wandsnider does
not specifically make this point, it appears from the historical data presented that a shift in
technique of analysis accompanied this trend, from computer identification of spatial
clusters, to a recent state of practice in which it is more common to rely on visual
inspection of density maps of formationally sensitive indicators (Wandsnider 1996:Table
1).
An analysis of intrasite stylistic spatial structure can be used to make inferences
about changes in the extent/location of settlements over time. Most open sites are
palimpsests of formation events and processes: in relation to intrasite patterning generally,
422

Kintigh describes the problem of overlap or mixing of deposits as “perhaps the outstanding
issue before (spatial archaeology)” (Kintigh 1990), and I suggest that the prospects for
advance in this respect are best for the special case of ceramic archaeology, where
relatively rich stylistic information can be expected to be a sensitive chronological
indicator.
Selection of Sites for Intrasite Spatial Analysis:
Entire-sites were collected as single units during initial intertidal surface collections in early
1996. By 1997, when intertidal survey was concentrated in the Kaliquongu region of
Roviana Lagoon, the aim was to try to document intrasite spatial structure for sites. Hence,
the sites Zangana, Gharanga (2 nd collection), Honiavasa and Hoghoi (2 nd collection) were
surface-collected in multiple transects. The exception was the Miho site, which was of
more limited extent, where significant along-shore spatial structure seemed unlikely.
Intertidal ceramic/lithic scatters at H5, H6 and H7 were not surface collected due to time
constraints and the small samples of sherds present. Largely aceramic intertidal lithic sites
at Mbolave Island, H4, H7 and at Elelo/Mbanga were left undisturbed for the present.
Objectives:
Analysis of vessel completeness (Chapter 5) showed that only a small percentage of the
deposited pottery was recovered, suggesting a harsh taphonomic regime for most sherds,
resulting in the complete destruction of most vessels, and lucky survival in good condition
for a minority of sherds from a minority of vessels. It seemed unlikely therefore that
spatial distribution of sherds corresponding directly with activity areas (Schiffer 1995b)
would have been preserved, although the possibility was not ruled out. Of more direct
interest though, is the question of whether there were any broad-scale differences within
423

sites in the distribution of fabric or style. Such differences might be expected to survive
the taphonomic regime, and would be significant to chronological analysis of style, as sites
could potentially be split into smaller seriation units on the basis of such distributional
patterning, potentially lessening the problem of chronological mixing of styles within site
assemblages.
If wave-sorting of sherds has been a primary factor in creating spatial structure,
small sherds should be found along the strandline, the upper intertidal, the swash zone,
should be relatively free of sherds, and larger sherds should be found at the deeper margins
of sites, in shallow water at low tide. The Zangana sherd size data was analysed in these
terms. Furthermore, if wave-sorting of sites has been a primary factor in creating spatial
structure, along-shore movement of smaller sherds would create a spread distribution of
small sherds in relation to large sherds, which latter should more nearly reflect the location
and extent of primary discard. The along-shore linear nature of the Hoghoi ceramic scatter
was considered an ideal vehicle for an analysis of this sort, where the narrow across-shore
extent of the site precluded much in the way of analysis of across-shore sorting. For the
Hoghoi site, lithic manuports were used in addition to ceramics to address these questions.
The Honiavasa and Gharanga sites were collected in unequal-area transects. In the
case of the Honiavasa site this was due to collection during a rising tide in deeper water
and windy weather, where collection proceeded along five pace-measured transects
between an anchored canoe and the shore, arranged radially in a fan formation around
Honiavasa Point. The accuracy of measurement of area of the resulting transects is
unknown, so relative measures of assemblage composition by transect are used wherever
possible.
One aim was to seek, through spatial structure, to untangle the sources of
decorative variability, and another was to take a distributional approach to sherd size,
treating artefact fragments as sedimentary particles. While there may be some lag deposits
of larger sherds that preserve behavioural or activity patterning, these also may be highly
424

correlated with site microtopography and minimally correlated with specific activity areas.
Method:
A statistical approach was eschewed in favour of visual inspection of graduated symbol
maps representing sherd counts or other quantification of cell compositional attributes,
similar to the display used for ground visibility data in the Hvar Archaeological Survey in
Yugoslavia (Gaffney et al. 1991). A difference from the Hvar displays that should be noted
is that in the present study the dot diameter is related to the square root of the raw transect
data, whereas in the Hvar survey a limited number of dot sizes was used to display data
that had been converted to ranges or categories.
An advantage of this graduated dot approach over isopleth displays is that the
displayed data (symbol sizes) are derived fairly directly from the raw data (sherd counts
or sums of sherd areas by cell), making a visual assessment of sample size effects by
transect possible. An advantage of this type of display over choropleth displays was that
the elongated transect shapes of the field collection strategy would have been distracting
to the eye had these displays been used, whereas size-graduated circular symbols created
a directionally-unbiased visual effect. Were frequency data or ratio data displayed instead,
significant sample size information would be less accessible.
Sherds were coded by numbered collection unit for intra-site spatial analysis in
table “Flat.db” as follows:
425

Paniavile (site 1): 1=cave assemblage collected prior to ‘89?
2=Jan 1996 colln.
Gharanga: (site 5) 1=Jan 96
3=August 96 colln.
2=G west of stream 98
3=G East of stream 98
Hoghoi (Site 2) 1-21=5m wide transects
99=initial surface collection by Kenneth Roga August 97
130=hg13t(subsurface test)
150=hg15a(additional)
151=hg15s(subsurface test)
152=hg15t (another test)
Honiavasa (Site 4) 1-5=radial transects 1-5
Kopo (Site 7) 1=single collection unit
Miho (Site 3) 1=single collection unit
Nusa Roviana (Site 6) 1=single collection unit
Zangana (Site 8) 1-105= transect units as per Figure 117(collected August97)
110=initial surface collection (June97) of Zangana
111=last collection (August 1997) of Zangana
Digital site maps were drawn for Sites 2, 4 and 8 (Hoghoi, Honiavasa and Zangana) from
field notes and tape/compass maps using AUTOCAD drafting software. These maps were
linked to ceramic thematic data tables using ArcInfo and MapInfo programs. MapInfo was
used to create site graduated dot thematic maps.
426

Figure 117: Intertidal collection units at Zangana: numbers are values in “unit” column
in table “Flat.db” appended on CD. Units without numbers are those which yielded no
sherds.
427

Zangana:
Natural Formation Processes at Zangana:
Across shore size sorting. The question: can I demonstrate that appreciable numbers of
sherds have moved, and if some sherds may not have moved much, can this be shown?
It would be unsatisfactory to analyse the Zangana materials purely in term of distance from
shore in relation to sherd size, for two reasons: firstly, the shoreline has undergone
substantial modification during WWII as the site of a major American landing on New
Georgia, and more recently during refurbishment of the road and wharf circa 1988;
secondly, the topography of the intertidal zone varies significantly between the North and
South of the site.
Zangana Sherd Count
All temper classes
80
40
8
New Georgia Mainland Roviana Lagoon
428
35m
�
Figure 118: Spatial distribution of the sherd sample at
Zangana.

The dense cluster of ceramic sherds at the north end of the site (Figure 118) occur
where the intertidal is characterized by relative microtopographic roughness, and overall
narrowness of the intertidal and shallow subtidal (around 15m from the high water mark
to the lower limit of the intertidal). This microtopographic roughness is thought to have
acted as a sherd trap and shield, and supplies the primary explanation for the preservation
of sherds in this area. By contrast, the intertidal zone south of the wharf is generally about
30m wide during the low tides, with sherds tending to be found in the recesses of a more
gentle microterrain.
North of the Zangana wharf there are two clusters of transects with reasonable
sample sizes, interspersed by an area of soft sediment most likely deposited as a result of
the sheltered conditions from the SE tradewind and southerly sea breezes created by the
earthen wharf. South of the wharf there is a band of transects along the shoreline which
have marginal samples of around eight sherds each, and a zone or band towards the lower
limit of the intertidal zone which includes some slightly larger samples, comprising three
groups of transects.
Zangana Average Sherd Area
(cm2)
110
55
11
New Georgia Mainland Roviana Lagoon
429
35m
�
Figure 119: Across-shore size sorting at Zangana.

The larger sherds are along the seaward margins of the intertidal zone in most cases
(Figure 119), suggesting selective removal of smaller sherds and probably also of sherds
not fortuitously sheltered/trapped by microtopography. This pattern suggests the seaward
margin of Zangana South is a lag deposit, while the landward margin is a secondary
accumulation of wave-transported sherds. The large sherd counts from transects at the
north end of the site comprise mainly small sherds, with a couple of larger averages
occurring only in transects with small samples. There is a suggestion though of occasional
larger sherds around the seaward margins of this area also.
Sherds immediately to the north of the wharf are all small, and it is here that WWII
and subsequent effects are probably greatest, with trampling and recent sherd breakage
likely to be extreme due to modern foot traffic; also, accumulation of soft sediments
following rebuilding of the wharf in the 1990s has reduced archaeological visibility here.
The more extensive tidal flat to the south of the wharf displays a more coherent
pattern of size sorting across the shoreline, with all of the strandline sherd samples having
average sherd sizes of less than 8cm 2 . All this supports the intuitive impression formed
during field collection that the strandline comprised almost exclusively small, water
rounded sherds, while the seaward limits of the intertidal zone yielded larger sherds with
little evidence of rolling damage, and only minor sand abrasion in many cases, marine
growth probably being a factor in prevention of the latter.
A further question: was there a smaller locus or multiple loci of settlement in the
past that has been spread into the present 3000m 2 distribution by wave-transport of
sherds? It seems clear from the above that wave processes have created a pattern of small
sherds along the strandline which are relatively mobile and larger sherds further seaward
that are less so. Is there a spread of small sherds and a lag deposit of larger sherds? Or
perhaps multiple spaced lag deposits of larger sherds interspersed with small sherds
representing spread from multiple occupation loci? This latter possibility seems borne out
to some extent by the distribution in Figure 119, particularly to the south of the wharf. A
band of larger sherds is distributed sub-parallel to the shoreline, possibly following a pre-
bulldozer shoreline around what would have been the less prominent prehistoric Zangana
point. Whether clusters in this distribution are a random sample size effect or a
taphonomic effect relating to microtopographic variation are questions for future research
430

on these sites. A third explanation has not been ruled out, that these clusters of larger
sherds forming a line along the shore are a settlement pattern, a line of discard zones
surrounding house locations.
If sherd size and quantity are combined by summing sherd areas by transect, giving
weight to those transects having the greatest total surface area of pottery, a picture that
could be interpreted in this way emerges (Figure 120). South of the wharf, there are a line
of four holes in the distribution of potsherds which might result from a line of structures
surrounded by throw zones for broken pottery. A test of this idea would be to examine
whether these gaps could be random, and whether they correlate with microtopographic
variation, but these tests will not be attempted here.
Zangana Sum of Sherd Areas
(cm2)
620
310
62
New Georgia Mainland Roviana Lagoon
431
35m
�
Figure 120: Possible linear settlement patterning in the
distribution of large sherds at Zangana south.

If this distribution does represent a settlement pattern, this would imply either long-
duration stability in settlement layout, or short duration of occupation, as, were one of
these conditions not the case, such patterning would be obscured by overprinting.
Spatial Structure in Ceramic Style at Zangana:
The above discussion established that post-depositional taphonomic effects have probably
obscured most small-scale settlement patterning, with the possible exception of the
Figure 121: Initial point-provenanced collection of
decorated sherds at Zangana (subsequent collection
transects shown for spatial reference).
southern-seaward part of the site. Initial surface collection of a small number of sherds at
Zangana targeted only decorated sherds. Point locations of sherds were recorded.(Figure
121)
432

These data seemed to suggest Miho styles (Class 2 decoration and pinched band at the
neck were found mainly to the south of the wharf, and Gharanga/Kopo styles (punctate
band) were found to the north of the wharf. A single sherd with a Honiavasa style Class
1 bounded linear incised motif was found to the north of the wharf. The subsequent, more
comprehensive, 1/5 sample transect surface collection enabled stylistic spatial structure to
be established with more confidence. The distribution of ceramic styles at Zangana was
compared to the overall quantity distribution. Wavy-lipped sherds and Gharanga/Kopo
multi-band fingernail-pinched/punctate styles were distributed as expected from the
distribution of the overall sherd sample (Figure 122, Figure 123).
Deformation of the lip into a wave form
(Sherd count)
10
5
1
New Georgia Mainland Roviana Lagoon
Class 1 linear motif incised/fingernail-pinched sherds showed a different distribution to
433
35m
�
Figure 122: Lip deformation into a wave present in both
Zangana-North and Zangana-South.

Band of punctation/multiple bands pinching
(sherd count)
10
5
1
New Georgia Mainland Roviana Lagoon
434
35m
�
Figure 123: Bands of punctation were restricted to Zangana-North.
the one expected from the overall sherd distribution, with sherds of this style being found
mainly to the south of the wharf (Figure 124).
The pattern expected from the overall distribution of all sherds would be for the
bulk of Miho-style sherds to be found at the north end of the site, with only a few sherds
found south of the wharf. Instead, 24 Miho sherds were found either near the wharf or on
the reef flat to the south, with only four found at the north end if the site, where the
majority of sherds were found. The chances of this happening randomly have not been
calculated, but would be very slim.

Figure 124: Unbounded linear incision or necks banded with
pinching at Zangana.
Two alternate explanations for this patterning were considered, firstly that the Miho
Class1/pinch style, the Miho “wav” lip deformation style, and the Gharanga/Kopo
punctate/multi-band-pinch styles form a time-series, and are spatially segregated in this
instance by a shift or expansion/contraction of the locus of settlement over time (possibly
in the reverse order to that listed); secondly, that this patterning represents
contemporaneous, spatially segregated stylistic diversity, interpretable as diversity of
manufacturing traditions or diversity of activity patterning. This second explanation seems
unlikely, as the scale of the spatial strucure is too large to be activity patterning, and even
if one end of a village were producing different pottery discard is unlikely to be so neatly
structured.
435

Zangana Temper Classes
(sherd Count)
70
35
7
Quartz-calcite hybrid
Placered volcanic
Unplacered feldspathic
New Georgia Mainland Roviana Lagoon
436
35m
�
Figure 125: Distribution of temper classes at Zangana.
Distribution of Zangana Pottery Tempers:
Temper class sherd counts did not show strong patterning across the site (Figure 125).
Sherds tempered with unplacered volcanic/feldspathic/hornblendic stream sands are more
common in the northern part of the site (see an enlarged detail of Figure 125 in Figure
126). Quartz-calcite tempered sherds do not seem to have any particular departure from
the pattern expected from overall sherd quantities, suggesting that this anomalous temper
group is not highly spatially correlated with any of the decorative attributes. This suggests
that, if the Zangana stylistic spatial structure represents a time series of styles, the quartz-
calcite temper transcends this series, occurring throughout.

Zangana Temper Classes
(sherd Count)
100
50
10
Unplacered feldspathic
Placered volcanic
Quartz-calcite hybrid
Figure 126: Detail of distribution of temper classes at Zangana-
North.
Zangana Summary:
In summary, at Zangana the pottery distribution shows evidence of wave size-sorting,
where lag deposits of larger sherds are mainly found at the seaward margin of Zangana
South. There may be some settlement patterning in this part of the site, but the possibility
of correlation with mictotopographic variation has yet to be examined, and random
distribution might look something like the pattern observed. Stylistically, there is spatial
structure at Zangana, with the Miho style having a different distribution to other styles.
This may indicate a shifting locus of settlement across a ceramic time-series, or alternately
a contemporaneous expression of manufacturing diversity across a settlement. None of the
major temper types exhibit any clear spatial structure, although there is a (statistically
untested) suggestion of a negative correlation between Miho style ceramics and unplacered
volcanic temper.
437

Hoghoi:
Questions of taphonomy and other site formation questions at Hoghoi were addressed by
mapping data from the second surface collection, which comprised a linear along-shore
arrangement of 22 contiguous intertidal transects, each 5x12m, oriented with the long axis
at right angles to the shoreline (the first surface collection the previous year had been a
grab sample of 105 items, mostly decorated pottery).
Natural Formation Processes at Hoghoi: Along-shore Size Sorting:
The question: was there a smaller site in the past that has been spread into the present
110m distribution by sherd/manuport transport? This was investigated by looking at the
average size of sherds. Is there a spread of small sherds and a lag deposit of larger sherds?
Or perhaps multiple spaced lag deposits of larger sherds interspersed with small sherds
representing spread from multiple occupation loci? The overall sherd count per transect
identifies those transects where sherd sample size is sufficient to address this question
(Figure 127).
Transect sample sizes
(sherd count)
150
75
15
0 0 12 24 12 5 3 4 64 149 95 67 39 23 66 9 28 23 46 31 10 0
0-5m 15-20m 30-35m 45-50m 60-65m 75-80m 90-95m 105-110
Figure 127: Distribution of the sherd sample at Hoghoi.
Transects 0 through 10m, 25 through 40m, 75-80m and 100 through 110m were excluded
as having too small a sample of sherds to be reliable indicators of along-shore size-sorting
processes.
438
�
The fairly large sherd count at 70-75m is immediately adjacent to a transect with

a very low sherd count thought during field collection to have been the subject of coral
boulder clearance for a copra-loading canoe-landing (the 70-75m transect appears to have
been the recipient of the coral cleared from the 75-80m transect. It can be seen in figure
12 that the average sherd area for these 66 sherds is noticeable smaller than for
neighbouring transects, suggesting many of these sherds may have resulted from trampling
and loss of wave-shelter in the “cleared” transect immediately to the west.
Average sherd area
(cm2, showing sample sizes)
0-5m
30
15
3
0 12 24 12 5 3 4 64 149 95 67 39 23 66 9 28 23 46 31 10 0
15-20m 30-35m 45-50m 60-65m 75-80m 90-95m 105-110
Figure 128: Lack of sherd size variation at Hoghoi, except at 35-40m (n=4).
439
�
There is little evidence in Figure 128 that the site comprises concentrated lag-deposits of
large sherds flanked by transects containing only smaller sherds. The 45-50m sample of
149 sherds, which is the most dense scatter of sherds, is not composed of unusually large
sherds, but the 50-60m transects have very slightly larger than average sherd area and
reasonably high density of deposits.
Lithics at Hoghoi:
Data on lithic manuports, many of which appear to be ovenstones, was mapped to further
investigate the evidence for along-shore size sorting. Lithic manuports tended to be larger
on average between the 25m mark and 60m, in the region where sherd density is highest
(Figure 129).

Average weight of Hoghoi Manuports (g)
Sample count shown
200
100
20
3 3 3 11 0 8 20 9 44 53 11 29 21 12 14 8 30 12 19 23 7 4
0-5m 20-25m 40-45m 60-65m 80-85m 100-105m
Figure 129: Larger average manuport mass from 25m to 60m at Hoghoi.
Of the smaller lithic manuports, many were unfractured water rounded cobbles, prompting
speculation that these might have some non-cooking function such as net weights or sling-
stones. Distribution by form and size classes is mapped in Figure 130.
Hoghoi Manuport Forms/Sizes
(Quantified by Count)
40
20
4
Rounded stones

showing little evidence of the strong clustering at 45-50m in evidence in the ceramic
distribution, or the distribution of larger stones.
It seemed reasonable to conclude that these smaller rounded stones were more
transportable than ceramic sherds, and thus had been spread by wave action, obliterating
any activity patterning. A lithic manuport type series, established through petrographic and
macroscopic classification, identified several geological source groups within the
ovenstone assemblage (Chapter 10). When manuports were mapped according to source
group (Figure 131), Classes 3, 5 and 18 dominate at the eastern end of the site.
Hoghoi Manuport Classes
Quantified By Count
30
15
Class 1
Class 2
Class 3
Class 5
Class 18
3
0-5m 20-25m 40-45m 60-65m 80-85m 100-105m
Figure 131: Size-sorting of manuport petrographic classes at Hoghoi, suggestive
of different size-procurement patterns by source.
In addition to the size sorting effect there is also a possible difference in procurement
pattern by source, with smaller stones being selected from some sources than others.
Differences in the size of stones available or selected from different sources is likely to
have produced the patterning in Figure 131. The only large Class 18 stone (1100g) was
found in unit 10 (45-50m), offering further support for a size-sorting wave-transport
explanation for spatial structure at Hoghoi, although re-use of materials with remaining use
life from an earlier, more easterly occupation cannot be ruled out.
441

Distribution of Hoghoi Ceramic Tempers:
Ceramic tempers offer further evidence relating to the question of size/density sorting by
waves as opposed to along-shore behavioural spatial structure (Figure 132). Quart-calcite
and unplacered tempers are slightly more common from 65-80m than elsewhere, but since
these tempers may have lower specific gravity than placered volcanic tempers with a high
ferro-magnesian component, density sorting by along-shore wave processes could explain
this patterning (but analysis showed no difference in specific gravity by temper).
Distribution of pottery tempers
(Quantified by sherd count)
130
65
13
Quartz-calcite hybrid temper
Unplacered lithic-feldspathic
Placered volcanic
0-5m 20-25m 40-45m 60-65m 80-85m 100-105m
Figure 132: Distribution of potsherd temper classes at Hoghoi.
This analysis of the distribution of temper types suffers from sample size problems,
and highlights the need to make the best possible use of the samples that have survived the
last two-to-three millennia. The hundred or so larger sherds uplifted during the initial
collection, together with the adzes collected at the same time, would have alleviated this
problem to some extent.
Distribution of Ceramic Styles at Hoghoi:
The collection of a hundred decorated sherds or lithic artefacts from this site on first
discovery has had a greater adverse effect on the stylistic component of this spatial analysis
than on the manuport or temper analyses. This highlights the need for careful spatial
recording on these fragile sites. There is so little spatial stylistic information from the
second collection at Hoghoi that the information was not mapped, and some general
comments only are made.
There seems to be proportionately more deformation of the lip into a wave form
442

and Class 2 linear motif on the rim or shoulder between 80m and 95m than between 40
and 65m, especially at 80-85m,although numbers are so small that this could easily be
sampling error or quantification error, but punctate band/multiple bands of pinching are
found right along the site, albeit in small numbers ( peaking at six instances of punctate
band at 45-50m).
Hoghoi Summary:
Sherd quantities suggest a strongly clustered distribution of potsherds in the vicinity of the
45-50m transect, with a variable density across other transects. The largest mean sherd
sizes, by a very small margin, were around the 45-50m transect, suggesting along shore
size-sorting is operative to a slight extent, and that there is a slight lag effect in the vicinity
of these larger (on average) sherds. The spatial distribution of lithic manuports also
suggested along-shore size sorting (somewhat surprisingly in view of the higher specific
gravity of the lithic manuports, but perhaps the flat shape of sherds makes them less subject
to transport by rolling). Compositional differences and possible functional differences
between manuport samples from the main 45-50m sherd cluster and the western sherd
distribution are most likely a result of size-sorting by waves, added onto a difference by
source in procurement strategies (small stones were taken from the Class 18 source for
some reason).
Initial collection of a hundred or so larger sherds from the site prior to more
detailed transect collection made a spatial analysis of stylistic structure impossible, as these
were the bulk of the sherds diagnostic of the styles defined in Chapter 9. This highlights
the need to record the locations of items carefully for this fragile site type, which is easily
rendered uninformative by casual surface collection. This finding is important for future
resource management and research.
Honiavasa:
443

Sherds were collected in five pace-measured transects at Honiavasa, so the site maps on
which the spatial analysis which follows are based are only roughly to scale, and transect
areas may in practice have been unequal. Consequently, sherd counts cannot be translated
into sherd densities, but relative frequencies of sherd attributes and taphonomic measures
such as average sherd sizes can be used with confidence. Overall sherd sample sizes are
reasonable for all five transects when looking at the distribution of sherd size and
composition (Figure 133).
Sherd Count
110
55
11
Honiavasa Passage
� 35m
444
113
98
45
76
110
Honiavasa Point
Figure 133: Distribution of the sherd sample at Honiavasa.
Taphonomy:
Honiavasa has the largest mean sherd size of all sites in the study. The site had not
previously been discovered by archaeologists. Honiavasa is located opposite the modern
village of Sasavele, across the reef passage, and is not subject to intensive foot traffic or
intensive children’s play activities. The site as collected was largely subtidal in water up
to a metre deep at low water on 20 September 1997, and may extend into deeper water.
This means that for most of the year the site is well submerged and sherds are seldom
subject to high-energy wave transport, despite the reef-passage location. Sherd size
increases slightly with increasing depth of water from east to west (Figure 134). The
centre transect is that most exposed to wave -induced sherd damage, as it is here the reef
flat proper begins, and here that refracted swell curling around from the passage to the east
breaks at low tide. This is reflected in sherd size, although low collection intensity in this
transect due to a rising tide may have been a factor.

Average sherd area (cm2)
33
16.5
3.3
Honiavasa Passage
� 35m
445
Honiavasa Point
Figure 134: Effects of wave refraction (and collection intensity?) on sherd
size at Honiavasa: the western margin is exposed to waves from Honiavasa
channel, which expend their energy in a swash zone at about the centre of
the site at low tide.
Distribution of Temper Types:
Transect 2 (2 nd from right, n = 76) shows an increased frequency of placered volcanic
/unplacered pyroxenic tempers (Figure 135), otherwise the relative proportions of the
three major temper classes are reasonably consistent across the site. (Placered volcanic
with volcanic rock fragments is a subgroup of the placered volcanic group, and difficult
to distinguish reliably without petrographic analysis. Variations in the relative frequencies
in which this temper was identified may have more to do with variations in surface etching
resulting from cleaning than with behaviour in the past, hence shown in the same shade as
placered volcanic temper.)
Temper Types
110
55
11
Placered volcanic temper
Pv with V.rock fragments
Unplacered lithic/feldspathic
Quartz-calcite
Roviana Lagoon
�
Honiavasa Point
0 30m
Figure 135: Distribution of pottery tempers at Honiavasa.

Figure 136: Co-joining sherds from deeper western margin of Honiavasa..
There is a pattern of slightly increased incidence of quartz-calcite tempers at the eastern
and western margins of the site. Analysis of vessel completeness did show, however, that
the western margins of the site yielded larger sherds, and one particularly complete vessel
including the only two co-joining sherds recovered during the entire study (Figure 136),
so perhaps taphonomic variation should not be hastily discounted in explaining variation
in the distribution of Quartz-calcite tempered sherds. The same could hold for unplacered
lithic/feldspathic sherds, which might be slightly weaker, hence smaller and more easily
transported than placered volcanic sherds.
Distribution of Decorative Attributes at Honiavasa:
Due to the relative plain-ness of the Honiavasa pottery, analysis of the spatial distribution
of styles at Honiavasa is faced with severe sample size problems (Figure 137). The
relatively high frequency of notched lips of two types, and relatively low frequency of
carinations/Class 1 linear motifs at the western end of the site may be a sampling error,
446

Distribution of decorative attributes
(Quantified by sherd count)
30
15 3
Band of impression on the top face of the lip, or on the outer edge
Carinated vessels or Class 1 linear motifs
Band of fingernail pinching on the outer edge of the lip
Roviana Lagoon
�
Honiavasa Point
0 30m
Figure 137: Distribution of pottery decorative attributes at
Honiavasa.
especially in view of higher vessel completeness in transect 5.
Honiavasa Summary:
While sherd size distribution suggests some taphonomic variation across the site consistent
with field observation of site morphology and wave exposure, the data on Honiavasa
temper distribution suggests that mean sherd size as a taphonomic indicator may mask
additional significant taphonomic variation. Recovery of two large cojoining sherds from
transect five at the deeper western margin of the site alongside Honiavasa passage provides
a clue that taphonomic degradation has been less severe in this part of the site. If
preservation is better at transect 5 this may indicate that the relatively large sample of
sherds from that deeper location actually represents a relatively small number of vessels
in a more complete state, a biased sample from which several diagnostic pieces may derive
from a single vessel.
Chapter Summary and Conclusions.
447

In drawing together the information from analyses of these three sites, the first issue is
whether an inter-site comparison of any of the themes of these analyses, size distribution,
temper distribution, or stylistic distribution, can be informative. A question in this regard
mentioned above in relation to Zangana stylistic distribution was whether a comparison of
the distribution of styles between sites would show up regularities in such patterning, or
whether these might be considered as random variation from site to site. A further question
for intersite comparisons that arose in relation to a structured distribution of tempers at
Hoghoi was whether the anomalous quartz-calcite temper group separated out spatially
at all in other sites. The Honiavasa site analysis raised questions about the methods used
to assess intrasite taphonomic variation, the significance of which can now be discussed
in relation to other sites.
Class 2 Linear Motifs and Neck Pinching:
There was a clear pattern of spatial segregation of Class 2 linear motifs or neck pinching
(Miho style) from punctate bands/multi-band pinching (Gharanga/Kopo style) at Zangana,
which justifies Zangana South being treated separately to Zangana North in seriation
(Chapter 12). At Hoghoi, no firm conclusion of this nature could be drawn, and these
styles do not occur at Honiavasa. Of the two other sites where the Miho style occurs there
is no repetition of the pattern observed at Zangana, so the hypothesis that this spatial
structure represents a contemporaneous socio-ceramic spatial structure is rejected. This
leaves the competing alternative hypothesis, that the Miho style and the Gharanga/Kopo
styles form a time-series, and are spatially segregated in this instance by a shift or
expansion/contraction in the locus of settlement over time. This conclusion is significant
for seriations in Chapter 12.
Spatial Separation of Quartz-Calcite Temper:
At Hoghoi, the distribution of quartz-calcite temper showed some structure, but this did
not seem to be the case at Zangana. At Honiavasa there was a hint of similar structure, and
448

in neither case Honiavasa or Hoghoi) can taphonomic sorting or simply small-sample error
be ruled out as an explanation.
Implications for Resource Management and Future Research:
Spatial structure in collection sites has been shown to be key to interpreting these sites.
Wave sorting processes and swash-zone taphonomic effects can be seen in all three sites
analysed in this chapter. Also, the difficulty in analysing stylistic spatial structure at Hoghoi
was increased by removal of an initial grab sample. Similar sampling strategies at Paniavile
by a series of archaeological teams and by locals meant that no spatial analysis could be
attempted there.
It is clear that future research and resource management of such surface sites can
usefully apply detailed spatial recording, to the level of point-provenancing of artefacts and
recording of micro-topography, otherwise formation processes of these scatters of pottery
are unlikely to be understood. It seems unlikely to me that some of the fundamental
difficulties in seriating surface collections, such as unraveling temporal mixing from
different periods, and distinguishing activity patterning from taphonomic spatial sorting or
microtopographic trapping, can be resolved satisfactorarily unless such recording is done
carefully.
449

450

Introduction:
CHAPTER 12:
CHRONOLOGY
In this chapter information from thermoluminescence, radiocarbon determinations, and
ceramic seriation is presented, followed by a discussion which attempts to draw this
diverse information into a provisional chronology. Implications for future research and
resource management are also discussed.
Seriation measures time as a continuum of gradual change, in that rates of change
cannot be measured by seriation. In seriation chronological types or attributes are treated
as arbitrary units of measurement of the process of change rather than as essential classes
in the data, allowing change over time between one type and another (O'Brien & Lyman
2000:52-56). The seriator is not interested in establishing fixed boundaries of a type, for
example, whether something is Lapita or not, but rather in discovering which aspects of
artefact variability can be used to construct a coherent evolutionary series with which to
tell time. By contrast, in using absolute dates one is usually (but not always) concerned to
know which chunk of time is being dated (which event). Paradoxically, although
radiocarbon age is a continuous variable, applications of the technique seek to date
occupations, layers, horizons, and thus seek a discontinuous measure of time (for an
extended discussion of this see O'Brien & Lyman 2000). “Is this Lapita? How old is type
X? When was this stratigraphic unit deposited? When does Lapita end? Absolute dating
and seriation thus work best when combined in a dialogue in which the flow of time within
and between events can be calibrated by the age of the events and inform on the properties
of the events, such as relative order.
451

In this chapter the radiocarbon dates and thermoluminescence dates used are direct
dates associated with the manufacture or use of particular pots. Sites are temporal
mixtures, but the direct dates give the age of pots, not sites, and thus partially escape the
discontinuous view of time of dating by stratigraphic association.
Patchy sampling of a ceramic series might look like natural groupings in the
ceramic data, which could be misinterpreted as culture-historical phases. Distinguishing
between these alternatives, patchy sampling and discrete phases of occupation, requires
that other lines of evidence besides ceramic variability be taken into account to assess
whether there is historical continuity and heritable continuity in the sample.
If, for example, at some period the action moved offstage, out of the sampling
frame, and then descendants of the same people returned later, while there might be a
discontinuity in the sampled ceramic evolutionary record, other factors such as type of
settlement or use of raw materials and adze maufacturing techniques might be more
conservative, and provide evidence for heritable continuity.
The Roviana ceramic data falls quite easily into styles or types (as discussed in
Chapter 9). This tendency of decorative variability to come in types is regarded below as
more likely to be telling us something about the patchy view of time presented by the
available sample, than indicating any sort of cultural replacement. The argument is made
below that there is evidence for heritable continuity across the entire sample, from
settlement location type and from adzes, as well as from occasional sherds that are
intermediate between the styles or types that most decorated sherds can be easily assigned
to. The tendency for sherds to group neatly most likely results from historical
discontinuities in the sample rather than any lack of heritable continuity. This does not
automatically mean that there were hiatuses in occupation, just that the sample available
to the analyst is historically incomplete, in that some periods are not well represented.
452

Possibly settlements from these periods have been destroyed by waves due to exposed
location, are hidden under soft sediments, or simply have yet to be found, or alternatively
people may have indeed have been largely absent from the sampling frame at some times.
We cannot discriminate between these possibilities on current evidence.
Radiocarbon data from a series of AMS dates is presented and discussed in the
following section, followed in turn by a section on TL dating, and a section on seriation.
TL dates do not give any absolute chronology, but provide useful data on the relative ages
of units used in the seriation, despite problems with anomalous fading of the TL signal
from the feldspar grains used in the analysis. A series of CA seriations are performed,
examining the effect of controlling for vessel form in the selection of attributes used.
AMS Radiocarbon Dates on Potsherds:
As pottery was collected from the lower margins of the swash zone predominantly, spatial
associations between organic materials and the pottery were unlikely to be temporal
associations. Four AMS radiocarbon dates were obtained from sherds (Table 46). These
involve three different modes of association between pot and dated carbon,
• the formation of the clay deposit from which the pot was made,
• mixing of the clay with environmental impurities during manufacture of the vessel
• accumulation of smoke-derived carbon on the surface of the vessel during use.
Also, there is evidence from two of the dates to suggest that distinguishing between all of
these modes of association is possible, suggesting direct dating of pottery styles is now a
practical reality in Oceanic archaeology.
453

Table 46: Radiocarbon determinations from pottery (calibrated using OxCal 3.5,
Stuiver et al. 1998 atmospheric data).
Sample ID Fraction
Dated
Lab
Number
HV.4.143 “seed” NZA-
12345
HV.4.143 blackened
edge
P.68 charcoal
inclusion
HG.15s.1 fine soot on
sherd
surface
NZA-
12346
Years BP Cal. Age
2sigma
8250+/-370 8300BC-
6400BC
9742+/-80 9350BC-
8800BC
AA33504 2130+/-90 390BC-
30AD
NZA-1253 2619=+/-
45
454
900BC-
550BC
Site Unit
Honiavasa 4
Honiavasa 4
Paniavile Jan 1996
Hoghoi 15s
The dates from sherd HV.4.143 (Figure 138) most likely date the geomorphological
formation of the clay deposit from which the pottery was eventually made. The round
“seed” (identity as a seed is unconfirmed) dated to within 500 years in age of a dark
carbon-rich zone near the surface of the sherd, visible in the break. The carboniferous zone
can be expected to contain small amounts carbon expelled from the clay on firing,
potentially a mix of old clay-derived carbon and new carbon mixed in at manufacture,
while the round “seed” could have been included either at the time of manufacture of the
pot or at the time of geological formation of the clay. As both ages are substantially older
than the expected age of the pottery from style and provenance, we can be reasonably
confident that these do not date the manufacture of the pot.
The charcoal inclusion in sherd P.68 from Paniavile must predate firing of the
vessel and is of an age broadly consistent with our expectations of the approximate
antiquity of the intertidal pottery series (circa-or-post-Lapita) (Figure 139).

Figure 138: A method of identifying sub-fossil organic inclusions?
(Image supplied by Rafter Radiocarbon Lab)
Radiocarbon determination
2600BP
2400BP
2200BP
2000BP
1800BP
1600BP
Atmospheric data from Stuiver et al. (1998); OxCal v3.5 Bronk Ramsey (2000); cub r:4 sd:12 prob usp[chron]
AA33504 : 2130±90BP
68.2% probability
360BC (15.3%) 290BC
240BC (52.9%) 40BC
95.4% probability
390BC (95.4%) 30AD
1000CalBC 500CalBC CalBC/CalAD
Calibrated date
500CalAD
Figure 139: Calibration of radiocarbon determination from a charcoal inclusion in
a sherd from Paniavile.
455

Sherd HG.15s.1 was dated using smoke-derived carbon which had formed a fine soot on
the surface of the sherd, preserved as a result of shallow burial in soft mud (Figure 140,
Figure 141, Figure 142). The sherd had been treated by the archaeologist with methyl
methacrylate (B72). Pretreatment at the Rafter lab included soaking in acetone, scraping
off of a soot sample, followed by a series of organic solvent washes (two cycles of hexane,
two cycles of isopropanol and four cycles of acetone, prior to the usual acid/alkali/acid
sequence) (Prior 2000). This allows a reasonable level of confidence that the methyl
methacrylate was removed and the 14C age is reliable. The extent to which the soot dates
the use of the pot is dependent on the average age of the fuel used in the fires, so there is
some potential for an old wood effect.
Radiocarbon determination
2900BP
2800BP
2700BP
2600BP
2500BP
2400BP
2300BP
2200BP
Atmospheric data from Stuiver et al. (1998); OxCal v3.5 Bronk Ramsey (2000); cub r:4 sd:12 prob usp[chron]
NZA12353 : 2619±45BP
68.2% probability
830BC (65.2%) 780BC
775BC ( 3.0%) 765BC
95.4% probability
900BC (86.7%) 750BC
690BC ( 3.5%) 660BC
640BC ( 3.6%) 590BC
580BC ( 1.6%) 550BC
1000CalBC 800CalBC 600CalBC 400CalBC
Calibrated date
Figure 140: Calibration of a radiocarbon determination from smoke-derived
carbon on a sherd from Hoghoi.
The dates from Paniavile and Hoghoi do not overlap at 95% confidence interval, and
suggest that the total duration of the intertidal/stilt house settlement pattern was at least
160 years and possibly much longer. These dates suggest there is likely to be significant
456

temporal variability within the total intertidal-zone pottery sample, which is good news for
seriation, as the possibility that the decorative variation which forms the data of the
seriation is contemporaneous non-temporal variability of some sort is reduced.
The older of these two dates is from Hoghoi, while the younger is from Paniavile,
but as each of these sites contains a mixture of styles, and as neither of the dated sherds
is clearly identifiable to particular styles, these dates don’t tell us much about the relative
ages of styles. Although the Hoghoi site has a lot of Gharanga-Kopo style sherds, sherds
of the dated sample are from an area of the site (transects 12 and 15) where attributes
identified with the Miho style occur also, and the form of the vessel, with tallish thin
Figure 141: AMS sample taken from blackened sherd at far left, found on
the surface of unit 12 at Hoghoi. The sherd to the right, found in a
subsurface test of unit 15 at Hoghoi, may be from the same vessel, and also
has a sooted surface.
excurvate rim, is more like the Miho style than Kopo style (Figure 141, Figure 142). The
only decoration, if it can be called that, is a deep impressed groove around the neck. The
vessel is unusually small in horizontal dimensions, which may have been a factor
contributing to its preservation through increased form strength.
457

Figure 142: Vessel with surface sooting from Hoghoi dated by AMS
radiocarbon.
Another sherd in the illustrations thought to be from this vessel was located by test
excavation within 20m of the found location, on the surface in square 12. These were
attributed to the same vessel by virtue of similar form, the grooving at the neck, and similar
sooting on the body surface, not seen on any other sherds.
The dated sample from Paniavile was recovered from a plain sherd comprising a
soft shoulder and globular body, which cannot be identified to any stylistic grouping or
period.
Thermoluminescence (TL) Dates from Quartz-Calcite Sherds:
Applicability of TL:
Thermoluminescence dates the time a pot was last heated to a sufficient temperature to re-
set the luminescence clock (about 450/C) (Feathers 1997). In most cases this will date
either the manufacture of the pot or its last use. This assumption is particularly secure for
materials discarded into the sea. This makes TL particularly useful for surface
archaeological distributions in the intertidal zone and in shallow water.
458

Uncertainties arising from factors affecting the precision of radiocarbon dates such as old
carbon and contamination from geologic carbon do not apply to thermoluminescence.
Calibration of radiocarbon dates sometimes involves a reduction in precision, when the
calibration curve is flat, as in the period 760BC-250BC (Figure 138 and Figure 140). This
flatness makes charcoal-dating the transformations of the Lapita cultural complex
following its initial spread across Oceania extremely difficult.
The disadvantage of thermoluminescence dating lies in “its dependence on
numerous complex variables that can be difficult to estimate in any given situation”
(Feathers 1997:4). TL dating can be used as an ordinal measure of age, an interval measure
or as a ratio scale measurement, with a reduction in precision as one moves up the scale
between these. Long-duration events such as an occupation can be dated with more
precision by dating more of the objects involved (Feathers 1997:7).
Luminescence Physics:
The following is a synthesis from a recent TL review (Feathers 1997:11-54) with
comments relevant to the Roviana case added.
A material that has been exposed to natural radiation will emit a faint light when
heated, proportional to the amount of energy absorbed, and proportional to time elapsed
since such accumulation began. This relationship breaks down when enough time has
elapsed for the material dated (the dosimeter) to become saturated with radiation, to the
point where it can trap no more energy. The most common natural radioactive isotopes are
40 K, 238 U and 232 Th, which have long half-lives and effectively constant global levels,
allowing estimation of dose rates impinging on archaeological samples. There are other
minor contributors to natural radiation flux. Terrestrial natural radiation can be divided into
three kinds (Alpha, Beta, and Gamma rays) of which Beta and Gamma rays are effective
in inducing luminescence, by virtue of their penetrating power (Alpha rays do affect the
surfaces of grains and small grains).
Successful dating requires a source of radiation in the environment of the item to
be dated, or even within the item itself, and a natural dosimeter within the item (which
459

may also be a source of radiation, as in the cases of potassium feldspar, calcite or zircon).
Quartz and sodic feldspars are good dosimeters, but lack an internal radiation source.
None of the usual dosimeters will become saturated in less than about 100Ka (thousand
years) so saturation is not a serious problem for Lapita archaeology.
Fading is said to occur when the accumulated latent signal is not stable through
time, but leaks away at ambient temperatures, making the date too young, either by
mechanisms for which the probability of fading is known (thermal fading) or by poorly-
understood and difficult-to-predict mechanisms (anomalous fading). Thermal fading can
be more of a problem in hot climates, and for older samples, but in the case of the Roviana
pottery thermal fading is less of a factor as the pottery has been underwater for most of its
relatively short depositional history and has not been subject to high ambient temperatures.
Anomalous fading can render a sample undatable, and is thought to have affected the
Roviana samples, being a major obstacle to accurate calendrical dating in this case.
Anomalous fading is more common with feldspars, and seldom affects quartz, unless the
latter has feldspar impurities.
To derive the age of the sample two measurements are required, equivalent dose
and annual dose. The age is the equivalent dose (the total amount of radiation absorbed
since last zeroed) divided by the annual dose. Equivalent dose is measured either by the
additive dose technique, whereby the material is irradiated in increments and the growth
curve of the emitted luminescence is compared to that of a standard, or by the regeneration
technique, whereby the luminescence of the sample is measured, then zeroed by heat or
light, then rebuilt by incremental dosing, producing a regeneration curve. The shapes of
these response curves (the sensitivity of the sample) are used in both techniques to
estimate the age of the sample. Sensitivity to dose can change during measurement.
Linearity in the growth curve cannot be assumed. The causes of sensitivity changes during
measurement are not all well understood. After a zeroing event, the luminescence signal
can build at a different rate, depending on the nature of the zeroing event.
Measuring emission is more difficult for young samples like pottery, as the
amount of light emitted is less. Selection of an appropriate spectral filter may be required
460

to maximise the signal from such samples. Detection equipment tends to be designed for
older samples than typically analysed by archaeologists, and may lack sufficient resolution
for the task in some labs.
Measuring dose rate requires correction for moisture content, correction for
uncertainty regarding the external dose rate, and the possibility of changes in the dose rate
through time. Water absorbs some of the radiation, and correction for this factor is easiest
where the environment is known to have been either saturated or fully dry during the time
since zeroing (most likely fully saturated in the Roviana case). Alpha and beta radiation,
with little penetrating ability, are relatively easy to estimate from the properties of the
sample itself, but gamma radiation, with a range of up to 30cm in soil, comes mostly from
outside the sample, requiring some idea of the homogeneity of the environment in this
respect. The Roviana sites with a scatter of volcanic ovenstones and volcanic-tempered
pottery in a predominantly calcareous environment, are potentially inhomogeneous in this
respect.
“Particularly troublesome is the presence within the 30cm sphere of large
clasts with a radioactivity different from that of the surrounding sediment”
(Feathers 1997:31).
Placement of zeroed dosimeters in the sample location, or use of portable gamma-
ray scintillometers can calculate the modern dose to which the sample is exposed, but
where there is a history of turbation or movement of the sample, calculation of
environmental gamma dose will not be as clear-cut, but may be less variable between
samples from a site due to an averaging effect over time. Perhaps the issues for the
Roviana materials is how long have these been in lag deposits as opposed to a stable
matrix, and whether the dose rate was lower or more variable between sherds when these
were deeply immersed, compared to the present dose rate. A detailed sea-level history
combined with placement of dosimeters in experimental deposits could partially answer
these questions.
Weathering can affect internal and external dose rate through postdepositional
movement of radionuclides. For example water can remove U but leave Th intact. Clay
461

minerals in ceramics are end products of weathering, and in the case of the Roviana
ceramics, have clearly been exposed to salt water. Low fired ceramics (less than 1000/C,
as in the Roviana case as calcite reef detritus is unsintered), are particularly prone to these
effects, but if such a disequilibrium is detected for radionuclides with short half-lives a
correction can sometimes be applied, especially where the leaching effect is more or less
constant. If radionuclides of longer half-life are involved this is more difficult, and
uncertainty about dose rate can be the result. Detection of such disequilibrium is largely
a matter of cost: a variety of measurement techniques can be applied, the cheaper of which
measure alpha radiation only, and estimate other radiation from elemental abundance
combined with natural isotopic abundance, or by Neutron Activation Analysis. High-
resolution gamma spectroscopy is expensive, requires long counting times and large
samples, but measures the gamma activities of several radionuclides, and this supplies a
means to assess equilibrium.
Luminescence signal can be measured from individual grains (inclusion dating),
usually from either quartz or feldspar, or, where possible, zircon, which has abundant
radioactive impurities that negate the need for estimation of external dose. Use of
individual grains facilitates choice of dating procedures and eliminates influence from clay
minerals. Fine grain analysis, making no attempt to separate particular minerals, is required
where large grains are unavailable, and despite some problems, can be useful where the
environmental dose is complicated because the influence of gamma radiation is less with
the inclusion of alpha-irradiated fine grains.
Low signal output is the commonest cause of low precision when dating young
ceramics. The best results come from samples in which there are abundant quartz grains
of optimum size. Rocks which have been sufficiently heated often produce reliable dates,
using either TL or OSL. The main reason lithic materials are seldom used for TL dating
is that there is seldom reason to believe they have been sufficiently heated, although there
are also greater sample preparation problems with lithics. I see no reason why ovenstones
should not have been heated as highly if not higher than open-fired calcite-tempered
pottery.
462

Roviana TL Data:
Eight pottery samples from four archaeological sites were submitted to the University of
Washington Anthropology Department TL laboratory. All were from the quartz-calcite
temper group, which held the best prospect for isolating grains suitable for TL dating
(Feathers 2002).
Environmental Dose:
Nine sediment samples and one ovenstone sample were measured to aid in the estimation
of environmental radiation. Radioactivity of sherds was higher than that of sediments
samples. Volcanic manuports from Zangana yielded higher radioactivity measurements
than the calcite reefal detritus samples. Environmental dose rate was calculated by
averaging the dose from the sediment samples for each site, with the manuport
radioactivity omitted.
Equivalent Dose and Fading:
Two of the smaller samples, one from Gharanga and one from Miho, which also lacked
sufficient quartz gains for inclusion dating, were analysed by the fine-grain method, and
yielded equivalent dose measurements of low precision, due to weak signal-to-noise ratio.
The other sherds were initially analyzed using the inclusion method on extracted quartz,
but unreasonably old ages with high error terms were obtained, leading Feathers to
conclude there was a spurious luminescence component, and this approach was aborted.
Inclusions obtained by density separation of Potassium feldspars and Aluminium
Silicates produced a sufficiently strong signal for dating, but four of theses samples showed
evidence for anomalous fading (which can make the age too young). A correction for
fading was applied to three of these samples, which resulted in low precision. Results are
given in Table 47.
Feathers rejected the fading correction for uw476 as it seemed too old based on
the information supplied to him, which was that the occupation being dated ran from
463

about 2500-1900BP, however, the corrected age, as initially supplied by him by email, is
entirely consistent with the expected age of the Honiavasa Lapita site, and is therefore
included in Table 47. Looking at the ages uncorrected for fading, the two dates from
Honiavasa were the oldest (uw475 and uw476), while two of the dates from Paniavile
(UW 472 and uw474) and the date from Gharanga (uw471) formed a young group. The
other date from Paniavile could be grouped with the Miho date and the Zangana south date
( uw473, uw477, uw478). While Feathers does not think the dates can be used as relative
ages as they are statistically indistinguishable when corrected for fading, the argument will
be made below that the uncorrected ordination of ages, ignoring their standard deviations,
agrees with the seriation favoured below, and can be reconciled with the AMS C14
evidence also. This correlation with the seriation additionally supports the inference drawn
from the decorative analysis of the quartz-calcite hybrid temper, which was that this temper
group was imported over a substantial period.
Table 47: Thermoluminescence dating results. Precision shown is at confidence limits
of one standard deviation.
Lab# Site Sherd Grain size Date (ka) Date after
fading
correction
(ka)
uw471 Gharanga GH.183 fine 0.984±0.166 1.608±1.043
uw472 Paniavile C.49 coarse 1.095±0.364 2.079±1.202
uw473 Paniavile R379a4 coarse 1.801±0.296
uw474 Paniavile P.165 coarse 0.790±0.363 0.950±0.495
uw475 Honiavasa HV.5.120 coarse 1.903±0.290
uw476 Honiavasa HV.1.48 coarse 1.903±0.290 2.339±0.750
uw477 Miho MH.61 fine 1.798±0.278
uw478 Zangana B.1 coarse 1.817±0.338
464

UW474 suggests (at two standard deviations) that firing or use of quartz-calcite pottery
continued until after 1940 BP (BP=before 2001 for this TL data) or in calendrical years,
61AD. In combination with the radiocarbon evidence from NZA-12353 (Figure 140)
UW474 suggest a minimum total temporal span for the intertidal-zone sites from 550BC
to 61AD, or 611 years.
Seriation:
The technique chosen for seriation was correspondence analysis (CA), using the
WINBASP program developed by Bonn University. Correspondence analysis will only
produce a temporal seriation if the types/attributes in terms of which the units are being
described are chronologically sensitive. “There must be reasons to believe this in the first
Table 48: Definitions of attribute codes used in seriation tables and plots.
N_A_B_NB Neck, applied decoration, being a band, of nubbins
L_F_BOPC Lip, fingernail impression, being a band, of opposed pinching
N_F_B_PC Neck, fingernail impression, being a band, of opposed pinching
F_MB__PC Fingernail impression, multiple bands, of opposed pinching
L_I___BI Lip, impression, band of, inner edge of lip
L_I___BO Lip, impression, being a band of, outer edge of lip
L_I___BT Lip, impression, being a band of, top of lip
L_I___BB Lip, impression, being a band of, both edges
L_I__WAV Lip, impression, into a wave pattern (staggered inner and outer edges)
L_D__WAV Lip, deformation, into a horizontal wave
L_D__OCI Lip, deformation, discontinuous into individual spout-like impressions
G_P____B anywhere on vessel (G=general), punctation, single horizontal band
CLASS1LM Class 1 linear motif, anywhere on vessel
R_CLS2LM Rim, class 2 linear motif
S_CLS2LM Shoulder, class 2 linear motif
STAMPING Anywhere on vessel, stamped decoration (including dentate)
465

place (Shennan 1997:342).” The decorative attribute incidences (sherd counts) tabulated
in Table 42 were used in the initial seriation reported below. To control for form variation,
a number of these attributes were excluded from a second seriation, producing a different
result to that of the total attribute set.
In the seriations, attributes from Table 42 are coded as in Table 48. Schematic
representation of attributes is given in Figure 143.
Seriation 1: All Attributes, All Vessel Forms:
Sample sizes by site for the various attributes are given in Table 42. Component 1 and
Component 2 represent more than 84% of the total inertia, which reassures that the counts
of the various attributes reduce well into two dimensions of variability (Table 49). The Qlt
column of Table 50 indicates that some attributes are not well characterized by these two
components (impressions on the outer lip and impressions on the top of the lip, have the
lowest quality scores, while wave-impression of the lip, deformation of the lip by
impression, and Class 2 linear motif on the shoulder score fairly low, suggesting these
latter attributes are only moderately well characterized by the two summary components).
The Mass column indicates that lip deformation into a wave, a band of pinching at the
neck, and banded punctation have the greatest weight among the types by virtue of their
greater incidence, while stamping, lip impression into a wave and a band of applied nubbins
at the neck have the least sample-size-related weight of the types/attributes. There are no
outliers among the attribute contributions to inertia (Inr). Pinched band at the lip, Class 1
linear motifs, band of punctation, and band of impressions on the inner lip have the greatest
attribute contributions to inertia.
The first Ctr column (Table 50) shows that applied band of nubbins at the neck
(N_A_B_NB) contributes 8.0% of the inertia in Component 1 (80/1000), band of pinching
at the lip (L_F_BOPC) contributes 22.5% of the Component 1 inertia (225/1000), etc.
Thus we see that Component 1 is a product principally of three attribute incidences, these
being: Class 1 linear motif incidence (CLASS1LM), band of pinching of the lip
(L_F_BOPC), and a band of impressions on the inner lip (L_I___BI).
466

Figure 143: Attributes used in seriations, groupings explained in chapter conclusions.
467

Component 2 inertia is a result principally of the distribution through sites of bands of
punctation (G_P____B). Multiple bands of pinching, a band of pinching at the neck, and
Class 2 linear motif on the shoulder are also important contributors to Component 2
inertia. A plot of type (attribute) correspondence values for the two components is given
in Figure 144.
Table 49: Relative contribution of first and second components of CA using all
attributes and forms.
Component Iterations Norm Eigenvalue % Inertia Cumulative%
1 28 0.051 0.583806 51.3 51.3
2 8 0.082 0.376932 33.1 84.4
Table 50: CA diagnostics by attribute, all attributes included.
Attribute
Qlt Mass Inr Comp1 Cor Ctr Comp2 Cor Ctr
N_A_B_NB 971 20 42 -1534 962 80 -148 9 1
L_F_BOPC 969 50 127 -1616 909 225 415 60 23
N_F_B_PC 921 124 76 508 372 55 616 549 125
F_MB__PC 854 50 63 298 62 8 -1062 791 151
L_I___BI 982 61 94 -1307 977 179 -92 5 1
L_I___BO 63 63 15 74 20 1 -106 43 2
L_I___BT 0 61 20 0 0 0 -9 0 0
L_I___BB 893 29 32 -195 30 2 -1043 863 83
L_I__WAV 421 14 44 327 30 3 -1172 391 52
L_D__WAV 669 156 42 409 542 45 198 127 16
L_D__OCI 369 70 23 339 304 14 157 65 5
G_P____B 951 122 136 299 70 19 -1056 881 362
CLASS1LM 983 31 120 -2071 959 225 327 24 9
R_CLS2LM 895 92 76 593 371 55 704 524 121
S_CLS2LM 563 43 39 427 177 13 632 387 46
STAMPING 840 13 49 -1899 815 78 328 24 4
468

Figure 144: Correspondence plot (attributes) using all attributes and forms.
The same sort of diagnostic statistics are given for the sites (Table 51). The two summary
components give a good representation of the variability at Honiavasa (999/1000).
Similarly, but to a lesser extent, the variability at all other sites except the three with the
least mass (small sample sizes) is well represented. The Honiavasa site is an outlier in terms
of its contribution to total inertia, contributing more than 80% of Component 1 inertia.
Gharanga makes the greatest contribution to Component 2 inertia, with Hoghoi, Miho and
Zangana South making significant contributions to Component 2 also. Honiavasa’s
contribution to Component 2 inertia is negligible (1.8%).
469

Table 51: CA diagnostics by site, all attributes included (*=inertia outlier).
SITES Qlt Mass Inr Comp1 Cor Ctr Comp2 Cor Ctr
PANIAVIL 703 187 70 348 284 39 422 418 88
HOGHOI__ 817 165 87 174 50 9 -676 767 201
MIHO____ 794 180 98 507 416 79 484 378 112
HONIAVAS 999 167 418* -1674 985 803 201 14 18
GHARANGA 866 101 164 158 13 4 -1255 852 421
NUSAROVI 39 20 24 -112 9 0 203 30 2
KOPO____ 154 4 15 417 38 1 -733 116 5
ZANNORTH 487 61 41 327 140 11 -516 348 43
ZANSOUTH 761 115 84 520 327 53 599 434 109
A plot of these two components is given in Figure 145. The correspondence scores in the
Cor columns provide the coordinates for these plots. Honiavasa is an outlier along the
Component 1 axis, the most similar site along Component 1 being Nusa Roviana, while the
other sites separate into two groups at opposing ends of Component 2. Nusa Roviana has
one example of a Class 1 linear motif (sherd NR34), but is closer to Honiavasa more by
default, by absence of characteristics, leading to a location on the plot close to the average,
or 0,0 position. The single occurrence of stamping at Nusa Roviana will also have had a
slight effect on the plot. It must be remembered that Nusa Roviana was not well
characterized by these two components. Stamping and Class 1 linear motif occurred on the
same sherd, so the similarity between Nusa Roviana and Honiavasa hinges on a single
sherd, and thus should not be made too much of.
470

Figure 145: Correspondence plot (sites) using all attributes and forms.
As discussed in the form and decoration chapters (Chapters 8 and 9) there are some
decorative attributes that are highly correlated with aspects of vessel form variation, and
which may be decoration specific to vessels of a particular function. In order to minimise
the risk of incorrectly seriating contemporaneous functional variation, these attributes were
excluded from analysis below. Excluded were: Class 1 linear motifs; bands of applied
nubbins at the neck; and stamping; all highly correlated with carinated vessel forms.
Multiple bands of fingernail pinching were also excluded due to occurrence only on short-
rim heavily excurvate Gharanga-type vessels.
Also, some lip impression similarities between Gharanga form/decoration type and
some Honiavasa tall excurvate rims are likely to be analogous rather than homologous
similarity (see Chapter 9), and banded lip impressions are thus excluded from subsequent
471

seriations, with the exception of lip impressions on both edges of the lip forming a wave
pattern; this decorative attribute occurred mainly in the Gharanga site on tall weakly
excurvate rims within the “general purpose” Form 6c/2c category, and indications from
this pattern of occurrence are a relatively restricted temporal occurrence.
Seriation 2: Form-correlated Attributes Removed:
Cumulative inertia of the first two components is 88%, suggesting that the patterning in
the data summarizes well into these components (Table 52). The three components
together produce an average quality of characterization for the attributes of 81%
(812/1000) (Table 52), with three attributes only moderately characterized (wave-
deformation of the lip, discontinuous finger-deformation of the lip and Class 2 linear motif
on the shoulder). Of these, Class 2 linear motif on the shoulder has a low mass figure,
related to small sample size. Component 1 contributions to inertia are almost wholly
dominated by fingernail pinch band at the lip, while the major contributor to Component
2 inertia is banded punctation (G_P___B).
Class 2 linear motif on the rim and a band of pinching at the neck make significant
contributions to Component 2 inertia also, while the reasonably numerous wave
deformation of the lip and discontinuous finger-deformation of the lip classes make very
little contribution to the inertia of either of the first two components. This is particularly
interesting, as these ubiquitous decorative techniques may indicate either a long production
span, or possibly a temporally central production span within the Roviana intertidal milieu,
where occupation spans of sites tend to incorporate discard of these attributes by virtue
of their temporal centrality. Component three may hold the key in this regard, because
these attributes, and particularly the related technique of lip impression into a wave, have
a lot of inertia there. The correspondence scores of these attributes for Components 2 and
3 are plotted in Figure 146.
The position of pinching of the lip on this plot can be disregarded as it is an outlier
472

in Component 1, which is not shown. Lip impression into a wave pattern is also an outlier,
in the low-inertia Component 3, but small sample size must be born in mind here. Although
most examples occur in a single site, absence from others might be a sample size effect.
The tight grouping of neck pinching, Class 2 linear motif on the rim, and Class 2 linear
motif on the shoulder confirms the “Miho” decorative style as a relatively coherent entity,
as suggested in Chapter 9. The separation of lip deformation into a wave and discontinuous
finger deformation of the lip from the “Miho style” grouping results largely from the spatial
separation of these attributes within the greater Zangana site, although wave-deformation
of the lip occurs in almost all sites to some extent, which is why these attributes lie close
to the intersection of the axes. Overall it is reassuring that the attributes order in the same
way in Component 2 as they did in Figure 144 despite omission of some potential problem
attributes.
Table 52: Eigenvalues and contributions to intertia of CA components, for data
excluding form-correlated attributes.
Component Iterations Norm Eigenvalue % Inertia Cumulative
%
1 28 0.051 0.611850 53 53.0
2 9 0.004 0.407938 35.3 88.3
3 10 0.041 0.080177 6.9 95.2
Table 53: CA diagnostic table for attributes, form-correlated attributes omitted (*=intertia
outlier).
Name Qlt Mass Inr Comp1 Cor Ctr Comp2 Cor Ctr Comp3 Cor Ctr
L_F_BOPC 1000 75 479* 2676 969 876 479 31 42 -21 0 0
N_F_B_PC 959 184 56 -111 35 4 -547 850 135 -161 74 60
L_I__WAV 972 21 81 -591 80 12 1255 359 83 -1530 533 624
L_D__WAV 638 233 26 -165 213 10 -133 138 10 192 288 107
L_D__OCI 485 104 18 -90 39 1 -62 19 1 295 426 113
G_P____B 996 182 240 -551 199 90 1099 792 538 89 5 18
R_CLS2LM 862 136 64 -158 46 6 -663 809 147 -62 7 7
S_CLS2LM 586 64 35 36 2 0 -525 441 43 -298 142 71
473

Turning to the table of scores for sites (Table 54) it is clear that all sites except Kopo and
Nusa Roviana are well characterized by the three components of the analysis. The mass
column indicates that these latter sites had small sample sizes, and were given little weight
in the calculation of inertia values. Honiavasa site is still an outlier in its contribution to
total inertia, by virtue of a contribution of almost 87% of Component 1 inertia, which, as
can be seen in Figure 146, is mostly to do with the incidence of a band of pinching at the
lip. Gharanga site dominates the contributions to Component 2 inertia, but Hoghoi, Miho
and Zangana South are important also, the first two by virtue of the high incidence of
punctation, the last because of neck pinching and Class 2 linear motif on the rim. A plot
of the first two component correspondence scores is given in Figure 147.
Figure 146: Correspondence plot (attributes) excluding form-correlated attributes
474

Table 54: CA diagnostics by site, form-correlated attributes omitted (*=inertia outlier).
SITE Qlt Mas
s
Inr Comp1 Cor Ctr Comp
2
475
Cor Ctr Comp
3
Cor Ctr
PANIAVIL 840 230 40 34 6 0 -410 834 95 14 1 1
HOGHOI__ 948 144 104 -439 231 45 727 634 187 264 84 126
MIHO____ 822 225 71 -192 101 14 -502 691 139 104 30 30
HONIAVAS 1000 75 476* 2661 965 86
7
508 35 47 6 0 0
GHARANGA 989 80 180 -609 143 49 1358 710 362 -595 136 354
NUSAROVI 325 19 9 -248 114 2 -102 19 0 322 192 24
KOPO____ 464 5 13 -410 61 1 811 237 9 678 166 31
ZANNORTH 872 70 44 -409 230 19 528 382 47 435 260 164
ZANSOUTH 957 152 63 -107 24 3 -551 635 113 -377 297 270
In comparison to Figure 145, Figure 147 shows a different ordering of sites in
Component 1.
Figure 147: Correspondence plot (sites) excluding form-correlated
attributes.

While this might seem relatively minor in view of the stability of Component 2 ordering
between these two plots, Component 1 is crucial to the overall relative dating of the three
broad styles that make up the series (the Honiavasa sample, Miho style and Gharanga
style) as it is Component 1 which shows which of the site assemblages is more similar to
the Honiavasa Lapita site. Where previously Nusa Roviana, Gharanga and Hoghoi were
closest to Honiavasa in terms of Component 1, now Paniavile is closest, by virtue of the
presence of five examples of opposed-pinch band on the outer lip there. Zangana South
is next closest due to a single occurrence of the same attribute there. These sample sizes
show how tenuous the homologous links between Honiavasa and the other sites become,
when the (potentially analogous) lip impression similarity is excluded. There are echoes in
this situation of the dilemma faced by Specht in deciding whether the transition from Buka
phase to Sohano phase on Bougainville was gradual, but poorly sampled, or whether
cultural replacement had occurred. It seems clear in the present case that the Honiavasa
site is a site largely alone, with the exception of occasional sherds with similar attributes
occurring elsewhere, and that the current sample is insufficient to address this question
from ceramic stylistic evidence alone.
With the omission of lip impressions Gharanga is now the most distant site in
Component 1, and all the sites in which a single band of punctation is dominant are now
at the opposite end of Component 1 to the Honiavasa site. Ordering in Component 1 of
these other sites roughly mirrors ordering in Component 2, with the group nearest to
Honiavasa in Component 1 (Paniavile, Miho, Zangana South and Nusa Roviana) forming
a separate group in Component 2 also.
Seriation 3: Honiavasa Excluded and Form-correlated Attributes Excluded:
If Honiavasa, the outlier, is excluded from the data, some of the minor components might
be expected to assume greater weight in the CA, which is not the case (Table 57, Table
55, Table 56 and Figure 148).
476

Table 57: CA eigenvalues and component contributions to inertia omitting Honiavasa
data and form-correlated attributes
Component Iterations Norm Eigenvalue % Inertia Cumulative
1 8 0.024 0.426248 72 72.0
2 21 0.048 0.081960 13.9 85.9
3 15 0.048 0.046658 7.9 93.8
Table 55: CA diagnostics for attributes, omitting Honiavasa data and form-correlated
attributes.
ATTRIBUTE Qlt Mas
s
Table 56: CA diagnostics by site, omitting Honiavasa data and form-correlated
attributes.
SITE Qlt Mass Inr Comp1 Cor Ctr Comp2 Cor Ctr Comp3 Cor Ctr
PANIAVIL 978 249 122 441 673 114 27 2 2 296 302 465
HOGHOI__ 947 156 188 -779 852 222 -249 87 118 75 8 19
MIHO____ 986 243 117 411 594 96 -142 71 59 -302 321 475
GHARANG
A
990 87 342 -1402 844 400 575 142 350 -105 5 20
NUSAROVI 233 20 16 32 2 0 -325 230 26 16 1 0
KOPO____ 472 6 25 -855 280 10 -680 177 33 -193 14 5
ZANNORT
H
Inr Comp1 Cor Ctr Comp2 Cor Ctr Comp3 Cor Ctr
L_F_BOPC 908 17 58 691 243 19 284 41 17 1106 623 455
N_F_B_PC 944 197 95 494 855 112 145 74 50 -67 16 19
L_I__WAV 981 23 156 -1300 424 92 1457 533 599 -311 24 48
L_D__WAV 561 246 36 92 97 5 -184 390 101 80 74 34
L_D__OCI 567 107 36 34 6 0 -313 487 128 -123 75 34
G_P____B 996 197 444* -1151 991 610 -76 4 14 39 1 7
R_CLS2LM 936 147 107 578 780 116 30 2 2 -256 153 208
S_CLS2LM 877 66 69 539 473 45 332 179 89 371 224 196
855 75 78 -590 566 61 -416 281 158 73 9 9
ZANSOUTH 932 165 113 500 616 97 355 311 254 -45 5 7
477

Only a single dimension of high inertia remains in the data, which is dominated by the
inertia of bands of punctation, and it produces groupings much like Component 2 in the
previous plots, with Gharanga at one extreme (tailed by Kopo, with a tiny sample, and
Hoghoi, then Zangana North). Zangana South is at the opposite extreme, with Paniavile
and Miho nearby, and Nusa Roviana tending in the same direction, although sample size
for that site is negligible. The second component has only 14% inertia, and mainly involves
the count of impression of the lip into a wave pattern, which occurs five times at Gharanga
and twice at Zangana South.
Discussion of Correspondence Analyses: the Alternatives:
The search for temporal variability centered on the exclusion of form-correlated decorative
attributes. Exclusion of CLASS1LM and STAMPING made no difference to the
Figure 148: Correspondence plot (sites), Honiavasa sample and form-correllated
attributes omitted from data-set.
478

ordination of sites or attributes, with the exception of Nusa Roviana, a small site sample
with a single sherd on which two of the excluded attributes occurred (Class 1 linear motifs
and stamping). The excluded sherd tended to make Nusa Roviana more similar to
Honiavasa, the latter site being an outlier in all analyses.
The effect on the seriation of excluding bands of lip impression is worthy of note.
If these analysis are interpreted in temporal terms, then two temporal ordinations of
groupings in the data (groupings are summarized in Table 58) can be constructed, either
A-B-C or A-C-B (or a mirror image of either of these, as seriation does not specify a
direction for time). These two alternatives, when combined with the notion that the
Honiavasa (A) phase is most Lapita-like, and therefore earliest (for which there is some
supporting evidence from TL), yield two alternative developmental modes. If Honiavasa
Table 58: Summary of variability
Phase
Symbol
A
B
C
Multipurpose vessel
decoration
one-piece construction PINCHED
BAND LIPS, IMPRESSED LIPS
(especially the inner lip), some
coarse wave deformation of the lip
One-piece construction PINCHED
BAND NECKS, CLASS 2
LINEAR MOTIF RIMS, CLASS 1
LINEAR MOTIF SHOULDERS,
tall excurvate rim form with thin
fragile lips, thick robust necks.
Tendency towards wavedeformation
of the lip, when
preserved.
One-piece construction
PUNCTATE BAND in the vicinity
if the neck, IMPRESSED LIPS,
some lip impression into a wave
pattern
479
Temporally associated
decorative attributes and
vessel forms/functions
Robust Slab-Constructed carinated
vessels with class 1 linear motifs,
stamped decoration, and band of
applied nubbins at the neck, some
compound or collar rims, some
evidence for shallow bowls and robust
storage vessels with low breakage
rates.
Various small bowls or flared rims,
low discard rate of large robust vessels
(storage vessels?)
Gharanga short-rim variant, thinwalled
in many cases, with multiple
bands of fingernail pinching on the
shoulder and bands of lip impression
(especially the inner lip), occasional
discard of punctate bowls.

is early, Miho style (B) is intermediate and Gharanga style(C) is late then the period over
which discard occurred is characterized by a change from slab-constructed forms to one-
piece forms, initially using a tall rim similar to that used in slab construction, but eventually
replacing this for some purposes with a more rugged short, strongly outcurved rim form
that survives well in the swash zone in spite of very thin construction in some cases.
Alternatively, if Gharanga/Kopo style is earlier than Miho style, then the transition to thin,
short-rim pots from Honiavasa heavy excurvate-rimmed pots is more rapid, the loss of the
heavily carinated Class 2 decorated pots or jars is also relatively rapid, and by the terminal
stage, Miho-style linear motifs (unbounded this time) reappear, along with tall rims arising
from Kopo-style tall rimmed punctate-decorated vessels.
Conclusions: Integrating 14 C, TL and Seriation:
14 C data suggested we can be reasonably confident that the Roviana intertidal sites have
a total occupation span of at least 160 years, unless there is an old-wood problem with the
smoke-derived date from Hoghoi. It was not possible to associate either of these dates
with a particular style of ceramic production, although the thin-necked pot from Hoghoi
with a tall everted rim seems intermediate in some attributes between the Gharanga style
and either Honiavasa plain pottery or Miho-style pottery.
The addition of eight TL dates expands the minimum occupation span of the
intertidal sites to 611 years. Also, if the TL ages from quartz-calcite hybrid tempered
sherds uncorrected for fading are taken as an ordinal-scale indication of age ( Dickinson’s
conclusion was that these tempers all originate from a single coastline, which might
provide some confidence that, although they fade anomalously, they may all fade in the
same anomalous way), then the Honiavasa site is likely to be oldest, while some of the
pottery from Miho, Paniavile and Zangana South is mostly intermediate in age between
Honiavasa and Gharanga. Other pottery from Gharanga and Paniavile is younger still.
More TL dates are necessary to test this argument, and could profitably use some of the
decorated Q-C tempered sherds held in the collections at present, especially sherd MH033,
480

decorated with a band of punctation at the neck.
styles, either:
or:
Seriation analyses can be summarized as producing two different sequences of
HoniavasaMihoGharanga/Kopo
HoniavasaGharanga/KopoMiho
(the seriation was more fine-grained than this, being attribute-based, but the major
chronological issue can be phrased in terms of styles). TL results suggest that the
Honiavasa site is older than the others, and that the Miho style predates Gharanga/Kopo,
although more evidence is needed to test this. This conclusion rests on an assumption that
the feldspar grains in quartz-calcite hybrid tempers used in TL dating all fade in a similar
manner, allowing a relative chronology.
The seriation analysis tended to show groups of attributes clustering into styles,
which suggested the homologous links between Honiavasa site and the other sites were
few (but present), which is most economically explained as resulting from historical
discontinuity in the sample. Also, the phyletic links that could show the relationship
between Miho and Gharanga/Kopo styles in detail were largely absent (but not completely
absent). This again is interpreted as a historical gap in the sample, rather than indicating
any essential, immutable reality to the Miho and Gharanga types. In the absence of
evidence to the contrary, any conclusion that these data clusters represent cultural
replacement of any sort is unwarranted. As discussed in the next chapter, there is non-
ceramic evidence for heritable continuity across the sample. If the corpus of sites were
much larger but these discontinuities persisted, then explanation for the discrete styles
might be sought in historical events. The current state of sampling of the New Georgia
group does not rule out gradual continuous changes in some form and decoration
attributes over time for the larger region. Whether such changes all happened at Roviana
Lagoon, but outside the current site sample, or whether the action moved offstage, beyond
the sampling frame, at some times, is unknown from the present evidence.
481

482

Introduction:
CHAPTER 13:
SUMMARY AND CONCLUSIONS
The key question laid out in Chapter One was whether there was a gap in the past
distribution of Lapita, consistent with an avoidance or leap-frogging colonization model,
or whether the near-oceanic Solomon Islands, as represented by New Georgia, were
settled early in the Lapita ceramic series. As the early Roviana ceramics are all in the sea
there were special questions concerning formation processes of this particular record, and
also a need for adaptation of archaeological method in some respects to fit this
archaeological landscape.
The broad question concerning Lapita distribution across the region of Near
Oceania was broken down into a number of more specific topics: how can the Roviana
intertidal record be interpreted in behavioural terms (and to what extent is patterning due
to natural rather than cultural formation processes)? How can a high-resolution chronology
be constructed to allow fine chronological control and enable the historicist approaches
that are so often attempted using inappropriately coarse radiocarbon chronologies? How
much of a sample is needed, and what sort of sample? how can sample size be quantified?
What potential is there for bias given particular sets of formation processes, and what are
the likely biases present in the samples?
In tackling these questions, conclusions of significance to resource management
and future research were drawn regarding the fragility of these swash-zone sites, the
poverty of information remaining, and extreme sensitivity in terms of information content
to further removal of material. In spite of the apparent poverty of the record, the high
archaeological visibility of lag deposits of pottery and lithics means there is unusually
483

good potential for fine-grained chrono-stylistic studies.
Summary:
A review of approaches to ceramic classification and seriation method in Chapter
1 identified a need for materialist approaches to artifact classification capable of capturing
stylistic drift, compatible with a descent-with-modification mode of ceramic change, in
order to develop fine-grained stylistic chronology. The review identified a need to control
for functional variability, as a means of isolating temporal variability.
A review of theories of form-function correlation found that functional classes were
best kept broad, and that use-life theory predicts that general purpose vessels that included
a cooking function should dominate vessel samples. Seriation should focus on forms that
fell into this broad use-category, and on decoration that cross-cut form-function classes.
The review identified a disjuncture between descriptive/classificatory units for
Lapita and post-Lapita (Mead’s system or Anson’s system principally for Lapita, and the
Frost-Irwin attribute combination approach for later pottery). In order to span the Lapita-
post-Lapita period these differences needed to be resolved. The relative simplicity of
decoration on utilitarian pottery, which dominates post-Lapita assemblages at least, meant
that descriptive and classificatory schemes needed to be sensitive to slight differences in
decoration. Also, in relation to units of quantification, there is a need for measures of
attribute frequency and sample sizes that are not biased by differences in assemblage
brokenness, or differential preservation of parts of the vessel. Accordingly, it was
determined that units of description and classification should pay close attention to
location on the vessel, and should focus on those parts of the vessel that preserve well and
are easily identified even as small fragments, allowing the structure of decoration across
the vessel as represented by sherd samples to be captured.
A review of temporal constructs for Lapita (Chapter 2) found that temporal
484

esolution of the available chronologies was alarmingly similar to the various definitions
of Lapita, and could be broadly characterized as culture-historical in resolution, with
Lapita continuing to have the status of a secure temporal horizon, but with potential in
current higher-resolution constructions of Lapita temporal variability for mis-assignment
of other dimensions of variability to time. This is partly a result a lack of clear sample
evaluation, an important step, particularly when constructing occurrence seriations and
interpreting motif-sharing. These factors undermine the security of current constructions
of the “Lapita Ceramic Series”.
While temporal constructs are seen as secure by several investigators, for example,
Anson's “Early Far Western/Western” distinction (Anson 1983, 1986, 1987, 1990, 2000),
Summerhayes’ Early/Middle/Late universal Lapita series (Summerhayes 2000a, 2001,
2002), and Sand’s Southern Lapita series for New Caledonia (Sand 1997a, 1999, 2000,
2001), there is a tendency for site samples to be slotted in to these schemes without much
control for functional variation especially, and geographic variation to a lesser extent.
While I would not suggest these constructions are necessarily wrong, these are seen here
as important works in progress rather than a secure foundation for comparative dating of
the Roviana materials.
While Best’s sequence at Lakeba was judged to be a more secure stratigraphic
demonstration of temporal changes from Lapita to other styles, uncertainties remain, and
the Eastern chronology need not apply elsewhere. Constructing a phyletic seriation to
argue that it does, as Spriggs does in “The Changing Face of Lapita” (Spriggs 1990) and
as Best does in “A View from the East” (Best 2002), is putting the cart before the horse.
Secure regional sequences need to be compared and cross-matched to assess whether there
are universal Lapita-wide parallel changes before doing this. It cannot be assumed that
there are such changes, to justify picking a phyletic series from across vast swaths of the
Pacific, because this assumes the answer to a primary question to be asked of the material.
In this final chapter some of the difficulties encountered in trying to reconcile the Roviana
485

early ceramics with other temporal constructs will be elaborated.
An increased emphasis on oceanography (wave processes) and the consequences
of these for archaeological preservation and probability of detection was advocated in
Chapter 3. Previous survey method and results in near Oceania were reviewed, and the
results of two surveys of the Roviana Lagoon region were presented in terms of a sample-
surveying approach. For Lapita, the informant prospection survey (Roviana survey)
obtained similar results to other informant-prospection surveys in the Near-Oceanic
Solomon Islands, i.e. no hits (other than a single sherd), in spite of informants volunteering
the locations of a number of post-Lapita or Lapita-derived intertidal sites. More intensive
intertidal coverage of the complex coastline of the more restricted Kaliquongu area located
a single Lapita site.
Although a sample of one suggests attribution of different results to different
survey methods should be made with caution, I think we have a better idea, as a result of
the Kaliquongu survey, how to locate sites of the Lapita period in the New Georgia area.
Despite the small site-sample size for Lapita, Lapita recorded site density (as opposed to
past site density) was well within the range recorded elsewhere in Near Oceania, offering
some evidence in support of a model of rapid and comprehensive spread of Lapita rather
than leap-frog colonization or avoidance modes. Given that the Kaliquongu survey covered
only a small fraction of the New Georgia intertidal zone and only one major reef passage
into the chain of landlocked lagoons encircling New Georgia, it is highly likely that
temporal-stylistic diversity will increase with further survey of similar locations around
New Georgia, which, if this turns out on further analysis to be the case, would in turn
strengthen support for a model of early, rapid spread. While speculative, this offers a
testable hypothesis for future work.
What is clear from the Kaliquongu results is that we would need a lot more
evidence than currently in hand to rule out early Lapita settlement of the New Georgia
group. Furthermore, given the results of the review of the Lapita ceramic series, it is
486

difficult to say on current evidence just how old the Honiavasa materials are, although they
suggest a sufficient period of divergence from an early-Lapita ancestor to lack pedestalled
bowls and flat-based dishes (although absence from the current sample could be a sampling
error given the small sample of slab-built and carinated Lapita forms currently recorded
from Roviana Lagoon).
Chapter 4 gave details of the ceramic database, and coding of units of ceramic
description in the core files on the appended data CD. Core sherd attributes of size, weight
and fabric were recorded in the “Master.db” table, while thickness, form and decoration
were coded in separated tables organizing multiple records by vessel part for each sherd.
This is a departure from the “diagnostic sherds’ approach where analysis focussed only on
a small subset of sherds which met (usually unstated) criteria. The approach allows close
control for part representation when querying the data, and allows convenient access to
a large set of attributes through a form layout of master record and client tables. Salient
attributes from this relational database were reconstructed into a single flat table for spatial
analysis and seriation.
Fabric classes included several placering variants of volcanic sands, most of which
are thought to be local, and an exotic quartz-calcite hybrid temper, thought on current
evidence to be from a continental location to the west, but other less exotic sources cannot
be ruled out on current evidence. There were few calcite-only tempered sherds. The
quartz-calcite tempers were much more quartzose than any quartz-calcite hybrid tempers
from the Bismarck Archipelago, with ratio of coarse K-feldspar-to-Plagioclase suggestive
of granitic origin. This raises the prospect of low frequency of trans-Solomon Sea or trans-
Coral-Sea pottery transfer, which would accord well with some of the materials from
Reef-Santa-Cruz Lapita sites, and with Irwin’s voyaging models, but is anomalous in that
no other sites in Near Oceania are known to have yielded similar pottery temper sand
evidence (Felgate & Dickinson 2001). The explanation preferred by Felgate and
Dickinson is thus unlikely to find much favour with the wider Pacific archaeological
487

audience unless such a source of pottery is physically located, which is beyond the scope
of the current research, so the implications of this temper remain an open question.
Form of vessel parts was recorded with the aim of providing sufficient information
to enable a reasonably accurate depiction of vessel shape variability. Nominal
morphological classes such as everted rim, restricted neck, carination, etc were thus
supplemented with a series of orientation, curvature and distance measurements. These
together with thickness measurements at a series of defined points enable more traditional
nominal classes of vessel form to be deconstructed for the purpose of an exploratory
analysis of variability, in tune with the materialist, evolutionist approach to identification
of stylistic and functional variation.
Similarly, decoration was coded by vessel part, employing formulaic descriptions
of the patterned layout of elements for simple banded decorations, and reference to
drawings for more complex or larger, or fragmentary designs. There was a conscious effort
in coming up with this scheme to separate decorative technique from decorative element,
and from the layout of elements. Metric variability of formulaic banded designs was
recorded, some of which information turned out to be salient to some of the basic research
questions when analysed in Chapter 8.
Chapter 5 provided a detailed investigation of vessel brokenness and completeness,
with several practical outcomes important to the conclusions of the thesis. While many of
the samples comprised quite large sherds, vessel completeness as evidenced by
construction of lip-sherd vessel families was extremely low, with mostly singleton lip
sherds present. A total breakage population was estimated by constructing a virtual
assemblage of whole rims broken in the manner of the sample, and iteratively sampling it
at various levels to see what range of sampling fractions yielded completeness data similar
to that of the real sample (Felgate & Bickler n.d). A statistical approach using the same
completeness data and a “number of species” algorithm obtained similar results to the
simulation, but the simulation results were preferred as more precise due to the use
488

of brokenness information.
Around 99% of the sherdage of the original breakage population has not made it
into the sample, and the breakage population vessel count would have been at least five
times larger than the vessel count represented in the lip sample. As little of the missing
pottery was buried on site, a taphonomic regime resulting in substantial sherd attrition was
inferred, leading to the conclusion that there was high potential for taphonomic bias in the
recovered sample. Also, accumulations-based inferences that use quantities of recovered
pottery to estimate intensity/duration of occupation are not applicable to data of this
quality. This finding led to a question being raised in regard to Wickler’s inference of low
intensity of occupation for similar sites in the Buka region. Also, comparisons using
relative abundances of styles should be made with caution, particularly when comparing
with well-preserved sites such as the Arawe Islands stilt villages and the Mussau stilt
villages, where preservation appears to be extremely good and taphonomic bias is less
likely. Similarly, comparison with terrestrial sites should be made with extreme caution,
as while biases are likely in heavily gardened sites like those of the Reef/Santa-Cruz area,
we do not yet know whether the biases are the same, or whether they are likely to create
differences in stylistic composition of assemblages.
How fragile is the site type? Chapter 5 informed on the state of preservation of
samples in a particular setting, the Roviana lagoon, where wave exposure is uniformly low
and sea level history is similar all sites in the study. Even in this sheltered setting, the sites
are in a very poor state of preservation, although visibility for those that remain exposed
on the surface is high. This finding is of fundamental importance to the central question
of this thesis, how to interpret the Lapita gap in Near-Oceania Solomon
Islands/Bougainville. If the sites are poorly preserved in this sheltered setting, along most
other coastlines exposure to a height level in relation to wave processes where
archaeological visibility is high would result in total destruction of the ceramic component
in most cases. Along such coastlines, where sea levels have fallen since Lapita times we
489

should be looking for vestiges of sites such as stormwash accumulations rather than in-situ
deposits, or deeply submerged sites if there is a history of relative rise in sea levels.
The low state of completeness of the Roviana intertidal pottery provides some
assurance that sherd counts are mostly independent observations. Sherd counts thus
provide a useful measure of observation sample size, and the relative frequencies of pottery
attributes in site samples. It should be noted that the same cannot be said of well preserved
sites, where many sherds in the sample can derive from a single vessel.
Chapter 6 extended the analysis of site preservation, adding the wave exposure
component to a model of preservation. While information on prevailing winds conflicted,
and environmental information on the extent and depths of reefs, sheltering sediments and
seagrass beds was not of sufficient quality to generate precise comparative results, it is
clear that none of the sites where pottery was found in quantity have more than 6 km of
wave exposure, and most have mitigating factors that prevent any substantial waves from
affecting the sites at low tide. Thus the results of Chapter 5 can be extended to other
regions as follows: where fetch is greater than 6km, and sea-levels have undergone a
similar fall as in the Roviana setting, preserved intertidal ceramic sites are not expected to
have survived except as lithic scatters or storm-redeposited material. This is just a
hypothesis at present given some of the measurement uncertainties, but provides a rule-of-
thumb starting point from which future survey of this sort can be targeted towards
relatively sheltered locations.
Analysis of formation process evidence in Chapter 7 considered a series of models
of cultural formation processes, followed by a consideration of the evidence pertaining to
natural or taphonomic processes as they affected ceramics. The absence of adjacent
evidence on land, and the absence of erosion features other than solution scarring of Plio-
Pleistocene upraised reefs are evidence against erosion of terrestrial sites into the sea. The
fresh state of preservation of Acropora corals on the upraised shore platform to landward
of most sites suggests a recent high stand, the upper limit of coral growth, at which time
490

the site locations would have been soft backreef lagoonal sediments unaffected by swash
processes. Paleoshoreline data (Mann et al. 1998) strongly support this interpretation, and
suggest a high stand of at least 1.5m above present at 3000 BP, probably more.
Subsequent sea-level fall led to to emergence of the backreef ceramic sites into the swash
zone, such as it is. Coral growth accretions on some sherds also support this conclusion,
as no modern coral growth occurs at these levels. Additional evidence against terrestrial
settlement is complete absence of acid-leaching of carbonate grains (disregarding surface
etching caused by cleaning sherds in acetic acid).
Taking all these factors together, including the recent information from geology
which was unavailable at the outset of the study, coral artificial islets seem an unlikely
formation process for these sites, as the water would have been at least chest-deep at low
tide, and probably deeper, requiring phenomenal labour to construct walling for a
settlement. Stilt dwellings over shallow water, with piles set in soft sediments now
removed by swash processes as a result of uplift, are the most likely scenario.
The artefact inventory of all intertidal sites was similar, dominated by pottery and
lithic manuports, with rare adzes, flakes of meta-basalt or metamorphosed mudstone, very
rare chert flakes, and a variety of abrader manuports. Coral abraders were not seen, but
the corraline gravels (dominated by small Acropora fragments)forming the matrix of most
sites would have made these difficult to find. There was no evidence for site abandonment
in the artefact inventory, or none that had survived postdepositional processes. All
complete adzes were heavily resharpened in comparison to one unused preform in a private
collection, suggesting sites were the focus of extended settlement resulting in the discard
of ceramics, ovenstones/net weights and worn out adzes.
Taphonomic analysis and spatial analysis (the latter at Zangana) offered support
for a transportation model of sherd attrition, while presence of ferromagnesian mineral
grains in a sediment sample from the coralline Hoghoi shoreline suggested dissolution
was also a factor in sherd attrition at some stage in the history of the sites. Taphonomic
491

analysis of sherd samples most clearly supported the collector and collection intensity
models, suggesting the potsherd scatters are highly sensitive to collection events. This
finding is important for archaeological resource management and also places an additional
burden on future investigators to thoroughly document the provenance of each item
collected, and attend to desalination procedures. We cannot pick up a sample expecting
to be able to go back later and get more of the same. This also suggests proactive
education on the significance and fragility of the resource is desirable at the local level.
Were artifact collectors to be successful in illegally selling items collected from such sites,
we could expect the accessible surface-site information resource to rapidly degrade to an
irretrievable level.
Taphonomic analysis in Chapter 8 also provided information on the types of bias
likely to be present in the ceramic samples. Strength variation between temper classes as
evidenced by sherd size was low, but sherd thickness was clearly related to sherd size,
suggesting thinner sherds are weaker and break more easily, and that thinner pottery is
likely, or thinner parts of pots are likely to be underrespresented in the lag deposits as a
result. Data was presented showing that some forms were stronger than others: Miho style
pottery tended to be thicker at the neck and thin and fragile at the lip, suggesting this style
is principally represented by neck sherds. Gharanga-style vessels with short, heavily-
excurvate-to-rolled rims preserve well due to form strength, and lips of this form are thus
more commonly attached to rims, leading to a larger count of Lip/Rim/Neck/Shoulder
sherds.
The analysis of pottery form variability and function in Chapter 8 focused on
vertical contour of the upper vessel, and also used negative evidence, the absence of flat
based sherds and stands (although there was one possible stand from Zangana). Six broad
functional vessel form classes were identified on the basis of upper body contour
variation, and a seventh was defined using body sherd thickness and form. These classes
were not all mutually exclusive, as some of the larger and more robust upper vessel sherds
492

perhaps should have been classed together with the Form 7 sherds. The aims of the
functional classification were the identification of a general purpose class that can be
expected to have high use rates, exposure to thermal shock and short use life/ high discard
rate, for use in form-controlled seriation analysis. A sub-class of Form 2 and a sub-class
of Form 6 were judged to best meet these criteria, and were used for seriation analysis in
Chapter 12.
In Chapter 9 decorative variability was explored, and the Zangana site was split
into two areas on the basis of decorative differences. Class 1 linear motifs
(bounded/defined with double-line zone-markers, either dentate or incised) were largely
restricted to the Honiavasa site, where they always occurred in association with carinated
Form 1 vessels, as did a band of circular nubbins at the neck. It was these sherds that had
led to the assignment of the site to the Lapita period from first discovery. Also present in
some quantity were everted-rim necked vessels with either lip notching/impression of
various sorts, or a band of pinching on the outer edge of the lip.
There were only two examples of deformation of the lip into a wave form at
Honiavasa, both of these coarsely executed, whereas this decorative technique was
ubiquitous at Nusa Roviana, Paniavile and Zangana. Deformation of the lip into a
discontinuous series of small spout-like decorative finger impressions was most common
in Miho site, and occurred in all other sites for which sample size was adequate. There
were only two occurrences of this attribute at Honiavasa, compared with 22 occurrences
at Miho. Class 2 linear motifs on the shoulder or rim were common at Miho, Paniavile and
Zangana South, and absent from Honiavasa. A single band of fingernail pinching at the
neck was common in the same sites as Class 2 linear motifs, and often occurred on the
same sherds. This complex of decoration was termed the Miho decorative class or “Miho
style” after the site where it was first recognised.
The decorative attribute of a single punctate band at the neck was common in two
sites, and was associated either with tall weakly everted rims above a slight neck
493

estriction (Kopo style) or with short heavily everted rims (Gharanga style), the latter often
associated also with multiple bands of fingernail pinching on the shoulder. Gharanga/Kopo
style sherds showed a tendency towards thinner necks than Miho style, supporting the
conclusions in Chapter 7 regarding the nature of stylistic-taphonomic bias.
A heterogeneous sample of applied decoration proved difficult to classify and was
omitted from the attributes used in seriations, although there is some discussion of possible
temporal variability given below.
Analysis of lip impression/crenation/notching in relation to lip form found a
correlation between lip orientation angle and the location of decoration which cross-cut
drastic differences in vessel form and decoration, suggesting that lip impression attributes
could cause faulty ordering of a seriation. Location of lip impression bands seemed to be
correlated with lip orientation, but as this measurement was dependent on rim orientation
measurements, which had poor precision, this requires confirmation from an enlarged
sample of sherds of 10% EVE or greater (below which rim orientation is difficult to
measure).
Neck decoration for all styles (Honiavasa, Miho and Gharanga-Kopo styles) was
dominated by a band of circular elements encircling the neck, while technique of
execution varied between these styles. Within an evolutionary definition of style and
function, following Dunnell, this suggests, if these are temporal types, that the pattern is
constrained and remaining constant, while the technique of execution varies, either
through drift or selection. The dominance of this pattern at the neck of the pot and the
temporal constancy suggests the pattern is fundamental to the decorative ideas of the
potters, and is functional in some way. A parallel was seen with Spriggs’ “changing face
of Lapita” in that these circular elements could be seen as the eyes in the Lapita face, and
a protective ideofunction was suggested, where the eyes were an example of the
“technology of enchantment or the enchantment of technology” (Gell 1992), perhaps
conferring protection from illness or poisoning to the contents of the pot and to eaters of
494

the contents.
Analysis of lithics in Chapter 10 focussed on petrographic classification of lithic
manuports, analysis of size distribution of these, morphological and petrographic
description of adzes, and petrographic description of sandstone abrader fragments. The
diversity of sources represented suggests enormous potential of such sites for lithic
sourcing studies, as a means of reconstructing patterns of raw material transport in the
past. This conclusion should add archaeological value to these intertidal/underwater
surface scatters, and encourage more intensive recording and sampling in the course of
future research and resource management.
Some of the unfractured water-rounded lithic manuports seemed too small to have
been procured as ovenstones, although small stones are sometimes collected for this
purpose in Oceania today (Nojima 2002: Pers. Comm.). Another functional possibility is
as net weights for large nets such as used ethnographically for turtle or dugong, both of
which are common today in the Lagoon, dugong especially in extensive seagrass meadows
adjacent to Hoghoi.
While most manuports probably were sourced from Rendova, some may have
originated from further afield (Viru Harbour or Kolombangara, both about 50km distant),
although this suggestion requires testing through examination of likely source streams.
Adzes are from a diversity of sources, showing a preference for metamorphosed
siltstones or metamorphosed fine-grained volcanic rocks. While some of these adze rocks
are conceivably from Tertiary sedimentary formations of the forearc chain of
Ranonnga/Southern Rendova/Tetepare, 15km or more distant from Roviana, the
recrystallization evident petrographically may indicate an older origin further afield,
perhaps among the complex geology of Choiseul, or possibly some of these result from
more local contact metamorphism associated with plutonic intrusions.
Adze forms are consistent with the Buka/Nissan Lapita reef-site evidence, and
suggest heritable continuity with Lapita, supporting Reeve’s contention that the Paniavile
495

material was Lapita-derived. This evidence for heritable continuity was also useful for
seriation, as a demonstration of heritable continuity is important for evolution-based
techniques such as seriation. The Near-Oceania stone adze sample from the Lapita period
is now dominated by the materials obtained from reef/intertidal sites, suggesting surface
archaeology holds the key to acquiring a large sample with which to study variability of
this rare artefact class in detail.
Petrography of abrasives indicated procurement from a variety of sources, and the
diversity of sources (no repeats) suggests the parent population of sources utilized is large.
Our sample of these sorts of materials is far from saturated. Detailed petrographic and
morphological comparisons with materials recovered by Wickler from Buka would be a
rewarding exercise.
Analysis of spatial structure (Chapter 11) in the ceramic data and the lithic Hoghoi
data identified evidence for size and density sorting, supporting the sherd transport natural
formation process model of Chapter 7. The analysis also looked for structured distribution
of tempers and styles. There was a clear separation of Miho decorative class from
Gharanga/Kopo decorative class at Zangana, supporting division into Zangana North and
Zangana South for seriation. By contrast, wave-deformation of the lip occurred ten times
in each of the two halves, suggesting this attribute had a different temporal trajectory to
the Miho decorative class despite association on some sherds with unbounded incised rim
decoration.
There was no strong evidence for a spatial separation of the exotic quartz-calcite
temper class from other tempers, which suggests, in conjunction with TL dating results,
stylistic analysis, and occurrence in all sites, that this class of temper had an extended
period of production and importation, despite the low sherd count.
In Chapter 12 three lines of chronological evidence were integrated: these being
radiocarbon dates; thermoluminescence; and ceramic seriation. C 14 data were interpreted
as allowing reasonable confidence that the Roviana intertidal sites have an occupation
496

span of at least 160 years, unless there is an old-wood problem with the determination on
smoke-derived carbon from the surface of a sherd in the Hoghoi site. Direct dating was
successful in two out of three attempts, with one case of dating of a sub-fossil organic
inclusion clearly documented. The prospects for AMS direct dating of pottery are good,
and throw the spotlight onto developing techniques for locating additional samples within
sherds, both from the current collection and from future samples. Demonstrating that it is
now possible to derive independent chronological data from the physical properties of
sherds and lithics in Near Oceania in this way, using TL and Radiocarbon, places added
archaeological value on “disturbed” archaeological contexts such as intertidal lag scatters.
The addition of eight TL dates expands the minimum occupation span of the
intertidal pattern to 611years at 2 s.d.. If the TL dates from the Quartz-calcite temper are
read as relative ages, disregarding corrections for anomalous fading, these suggest
Honiavasa is the oldest site, while Miho-style pottery from Miho, Zangana and Paniavile
is mostly slightly younger, with the youngest dates coming from Paniavile and Gharanga.
More TL data are needed to test the notion that the uncorrected ages provide a relative
chronology due to origin from a common source, and TL/carbon pairs would be an ideal
scenario. If a systematic model of Luminescence fading can be developed for the
feldspathic component of the quartz-calcite temper, then TL dating could become a key
tool for defining Lapita chronology in the Western Solomon Islands.
Seriation using CA produced different orderings depending on which attributes
were included, and the results of seriation from which lip impressions were excluded were
preferred, but this cannot be independently confirmed on current data. The tendency for
ceramic variability to coalesce into three styles (Honiavasa, Miho and Gharanga/Kopo was
partly an result of choice and sample sizes of attributes, but is likely also to be a
consequence of a historically incomplete sample. Honiavasa site has very little in common
ceramically with the other sites, although enough to suggest heritable continuity. Taken
together with continuity in settlement types, the preferred explanation for the three styles
497

is that we are missing the transition from Honiavasa utilitarian ceramics to Miho and
Gharanga/Kopo utilitarian ceramics, unless some of the ubiquitous attributes span this gap,
for example lip deformation on plain rims. Additional sampling of the regional record is
required, including areas beyond the current sampling frame, as concluded on other
evidence in Chapter 3 also. Should these stylistic discontinuities persist with additional
sampling, this conclusion would have to be reevaluated.
When the three different chronologies are synthesised, a Honiavasa>Miho-
style>Gharanga/Kopo style sequence seems most likely. Wave-deformation of the lip
seems likely to begin during the production span of the Miho style and continue till tall-rim
Kopo-style production commences. Gharanga-style short-rim vessels may be
contemporaneous functional variants of the Gharanga/Kopo period, or may be a
development out of Kopo style, with multi-band pinching of the shoulder, and thinner
construction being innovations following the reduction of rim height.
Under Balfet’s model as applied by Irwin (Balfet 1965, Irwin 1985), this sequence
may document replacement of domestic idiosyncratic low-skill production with specialist
standardised technically-accomplished production. The chunky Form 1 carinated jars in the
Honiavasa site with individualistic decoration, and heavy Form 6 pots, have been replaced,
if this series is correct, by progressively thinner and simpler, more standardised pots of a
more general-purpose nature, with better thermal properties as a result of reduction in
thickness, also potentially allowing more rapid drying prior to firing.
External Comparisons:
Having critiqued the use of presence-absence data for comparing subsamples of past
behaviour in Chapter 1 and 2, I am unwilling to make such comparisons using the
Honiavasa data as I believe these will simply mislead. The Honiavasa sample contains a
498

small corpus of carinated vessels with Lapita motifs, but sample size is insufficient for
relative-frequency based motif comparisons also. Comparisons are thus restricted to some
general statements. The Honiavasa carinated vessels with Class 1 motifs have clear
homologous similarity to the Lapita ceramic horizon. There are no flat bases or cylinder
stands in the sample, but this does not mean these were absent in the past. The complex
curvilinear motifs characteristic of early Lapita according to Best are not present in the
sample. Ishimura’s recent analysis, by contrast, has these techniques of decoration evolving
over a more lengthy period across the temporal dimension of the Lapita horizon.
A number of Anson’s motifs, or similar motifs are present in the small Honiavasa
carinated vessel sample. Mead motifs 16, something similar to motif 11 or 16.3, something
similar to motif 15, and a geometric motif similar to 17.1 or 30 are present among the small
sample of carinated Form 1 vessels. Mead’s 3-dimensional design elements N1.1, VB2,
TB3.1, and TB3.3 occur in the Honiavasa sample. Design elements 1.2, 5 and 6 occur at
Honiavasa. These characteristics also serve to link Honiavasa to the Lapita ceramic
horizon.
Carinated vessel forms at Honiavasa are characteristic of Summerhayes’ early
Lapita in the Bismarck Archipelago (Summerhayes Form V), but the rare occurrences of
dentate-stamping argue for a later date under the Summerhayes scheme. While this site
was initially described as late-Lapita (Felgate 2002) I would now prefer to have a detailed
regional sequence as a basis for cross matching, and am less confident in the temporal
constructs of others than at the outset of analysis.
I initially thought that the Roviana intertidal ceramics must date to around 500BC
or younger, based on the information from Watom and Buka, and the first C14 date from
Paniavile was consistent with this assignment. Feathers, however, was forced to reject his
measurements from Honiavasa as too old, when he compared the results obtained to the
expected age. Since then, a date likely to be around 800BC from Hoghoi has suggested
that the Honiavasa Lapita subtantially predates 500BC, also that the TL measurements
499

were giving a reasonable ordination chronology, and that Lapita is effectively over by
about 800BC at Roviana Lagoon.
As a result of this information, combined with the results of a critical review of the
underpinnings of current constructs of Lapita variability, I am now inclined to put the
absolute age of the Honiavasa carinated pottery at no younger than 850BC, and possibly
older. Miho style pottery may be being produced by about 800BC, and my inclination,
although as discussed in Chapter 12 this is unconfirmed by solid evidence, is to have
Gharanga/Kopo style produced sometime between 800BC and 0.AD.
Several occurrences of compound rims at Honiavasa provide similarity with rim
profiles from Buka (especially Wickler 2001: Figure 4.2k from site DAF). These rims are
listed by Sand as characteristic of the Southern Lapita Province (Sand 2000), and the Buka
and Roviana materials provide a suggestion that they are a more widespread trait. Poulsen
calls these “flange” rims for Tonga, and these are Mead’s 3-dimensional design element
2.8, defined from a Sigatoka sample, so they are clearly widely distributed, although the
best examples are from the large whole vessels recently found in New Caledonia. Given
the widespread distribution of these I hesitate to attribute them to a late date, and regard
them as potentially part of the repertoire of rim/vessel forms of the initial spread into
Remote Oceania, and thus having an origin of similar antiquity to the curvilinear/face
motifs such as Mead M33. Alternatively, the widespread occurrence of these rim forms
may arise through reticulate processes in RemoteOceania and the Near-Oceanic Solomon
Islands/Buka. I am tempted to call these an indication of a “Central Lapita Province”,
rather than Western Lapita, the latter something of a misnomer when Western Lapita is
found as far east as Tonga (eg. Burley & Dickinson 2001).
It is possible that the Roviana early ceramic sequence, in spite of the low number
of sites, few dates and patchy sample, is amplifying the temporal resolution obtainable
from excavated samples from elsewhere in Near Oceania for the Lapita/post-Lapita
transition, as predicted by the extended review in the opening chapters of the nature and
500

potential of seriation of surface sites as an approach to the archaeology of a region. The
Garua pottery from FSZ looks like a mixture of attributes that are spatially separated in the
Roviana surface sites. Wavy stamping, as illustrated by Summerhayes for the Garua sample
(Summerhayes 2000a:146, Figure 9.2, sherds 1620, 1101, 1927) occurs on several
Honiavasa sherds, in similarly abstract arrangements, with hints of use as a latitudinal zone
marker in some cases. Also, Summerhayes' Sherd 1275, for instance, is a typical Roviana
wave-deformed lip, occurring in high frequency at Miho, Paniavile, Zangana and Nusa
Roviana, but only rarely and in an odd, coarse, heavy variant at the Honiavasa Lapita site.
The separation of Honiavasa and Paniavile using linear motifs provides another clue
that in the Roviana surface-collected data we may be seeing an amplification of temporal
resolution in the relatively unrestricted landscape with plenty of places to move a stilt
settlement to. Miho style is dominated by variants of Mead motif 18, while motif 24 also
occurs several times. On the shoulder of Miho-style vessels it is not uncommon to see a
simplified form of motif 16, being just the diagonal lines forming a zigzag, expressed using
the single-line (Mead GZ3) rather than twinned lines. The motif listed as Mead M16.1
(Green 1979) (different to that illustrated by Mead) is present at Paniavile, where Miho
style predominates, and is associated with pinching at the neck, a feature diagnostic of the
Miho style.
This idea of temporal amplification obtains additional support from the
occurrences of Gharanga/Kopo style punctation elsewhere. The characteristic punctation
or stick impression diagnostic of Gharanga/Kopo style, where the still-plastic clay is
deformed inwards slightly as a bump on the interior surface of the neck region occurs on
more than 60 sherds in the sample, and is completely absent from the large Honiavasa
vessel sample. This decoration is illustrated by Anson from Ambitle (Anson 1983: Figure
X), and occurs on two sherds at site DES on Nissan (Wickler 2001:120), and at the EKQ
site on Mussau (Kirch et al. 1991: Figure 4c). Compared to these samples, the Roviana
501

sample, particularly from Gharanga and Hoghoi, represents the parent lode, unless this is
simply analogous similarity, but if the latter is the case, why is it so rare elsewhere?
The single quartz-calcite tempered example of this class of decoration may indicate
an extended zone of production, design emulation, or exchange of potters as well as pots,
as suggested by Summerhayes also from other data. Most of these vessels have placered
volcanic temper, and it is not certain that there is much benefit in petrographic comparisons
with the other examples from Buka and Papua New Guinea, although it never hurts to
look.
Gharanga-style multiple bands of fingernail impression on the shoulder occur in low
frequency in many samples elsewhere, and has generally been investigated at the level of
decorative technique rather than at the level of structured pattern across the vessel (e.g.
Anson 1983:Figure XI) although Bedford is much more specific in this respect (see for
example Bedford 2000: early and late Ifo ware, Figures 7.2-7.4). Until chronologies are
more clearly defined it will be difficult to distinguish between homologies and analogies
in respect of these simple decorative techniques, and Bedford’s reassessment of the incised
and applied relief “tradition” is apposite here too, providing a caution against jumping to
hastily to conclusions and making unwarranted connections.
If there is any connection, and/or if decorative technique and pattern is separating
out by site at Roviana, where it is mixed in with other pottery to the northwest or Vanuatu,
either the Roviana surface scatters are amplifying chrono-stylistic structure in the data and
thus temporal resolution, where this is more temporally mixed in excavated samples to the
northwest, or the trajectories of ceramic change are so different between Roviana and the
Buka/New Britain area that Roviana pottery may have changed beyond recognition as
Lapita by 800BC, while Lapita-style is still supposedly thriving in the Bismark
Archipelago.
Whether Gharanga/Kopo pottery is thus a valuable index fossil (O'Brien & Lyman
2000), and similar sites further north are yet to be discovered, or whether this is just
502

analogous similarity (unlikely in view of the rarity of occurrence further north) cannot be
resolved until temporal variability is better understood at the local level in each of these
areas. The preferred explanation is that we have yet to fully sample and define temporal
variability fully for the Lapita period and what came after, in New Britain, at Buka, and at
Roviana Lagoon, and that the current samples from throughout the region, including
Roviana and Buka and the Bismarck Archipelago, are insufficiently historically complete
and site-structured to allow the sort of high-resolution chronology that can be used to
discriminate between homologous and analogous similarity of simpler designs between
regions.
Late prehistoric adzes from Gharanga and Paniavile tie into Specht and Wickler’s
terrestrial chronology, and suggest intertidal settlement may have continued until circa
1500 AD, but this is speculative as the adzes are reported to have come from the intertidal,
but are not securely provenanced. Even if these were found in the intertidal scatters, this
does not mean they are the same age as the bulk of pottery in those sites, as surface
collections, to a greater extent than some buried sites can easily incorporate materials from
various periods.
The Lapita Gap as an Area of Low Probability of Detection :
The Roviana study shows that Lapita is most likely continuously distributed across near
Oceania, because the gap in the recorded distribution can now be seen to reflect
preservation and visibility rather than past behaviour. Now that we know the poor state
of preservation of the emergent Roviana stilt villages, and the likely depth of water in
which they were constructed, the absence of evidence from Choiseul, Ysabel and Ngela
is explicable by the geological evidence for subsidence in those areas (Bruns et al. 1986).
If the main chain adjacent to the New Georgia group is subsiding while Roviana is
uplifting, sites at the sorts of locations evident in the Roviana data will be either grown
503

over with coral or buried in soft reefal detrital sediments, and thus be both well preserved
and invisible, like the Mulifunua Ferry Berth site in Western Samoa, discovered by
fortuitous dredging. In areas with more substantial uplift than Roviana, such as the forearc
chain of Rendova, Tetepare and Ranonnga, preservation is likely to be worse than in the
Roviana Lagoon, in some areas through wave exposure, but also because uplift has
brought the pottery through the swash zone where it cannot survive wave exposure. This
can be expected to have substantially reduced probability of detection. Evidence suggesting
a Lapita gap in the Near-Oceanic Solomon Islands as avoidance or leap-frog colonization
is thus substantially eliminated.
This thesis can be tested by looking in places where high probability of past
settlement, archaeological visibility, preservation and survey methods combine to promote
a high probability of site detection. The scale of survey required is suggested in Chapter
3 to be of the order of 900km 2 of lagoon, or 40 reef passages, or about 150km of coastline
with high detectability, to detect a sample of ten or more sites with good samples of
Lapita-era and a larger number of derivatives of Lapita. With this sort of information in
hand, the behavioural questions that require us to be able to tell time with high temporal
resolution (Spriggs 2001) including consideration of spatial factors such as site density and
mobility models (Anderson 2002) can be matched to appropriate data.
Intertidal-Zone and Shallow-Water Archaeology:
This thesis has been as much about figuring out how to proceed in researching an intertidal
and shallow water archaeological distribution as about the distribution of Lapita pottery.
The key chapters in this regard are those on survey (Chapter 3) and formation processes
(Chapters 5, 6, 7 and 11). These could all have been vastly improved had the
environmental variables used (wave-exposure; sediment characteristics and vegetation
cover; relative sea level and tidal fluctuations; and collection site micro-topography) and
504

the locations of artefacts and manuports been recorded more systematically and in more
detail. This would have necessitated either a lot more survey time or much better
equipment (probably both). In retrospect, if accurate electronic survey equipment was
being used to capture these environmental details, point-provenancing of all sherds and
lithics would have been a simple matter too.
The wet tropical environment of the Roviana study is hard on such gear, which
must be carefully maintained to function well, may fail unexpectedly, and adds to the
expense and logistics of fieldwork Many of the suggestions for practice might seem trivial
to those working in urban environments or in the rural hinterland of industrialised societies,
where power supplies are usually available either from the main grid or from vehicle
batteries, and where electronic equipment is available either through University
departments or Resource Management Agencies or hire companies, and is often used as
a matter of course. Some of these suggestions might be very difficult to implement,
however, by resource management agencies in non-industrialised nations or regions, like
Solomon Islands, particularly where funding is meagre. The core information, though, that
cannot be reconstructed after the event of picking up a sample, is where the piece came
from, which is brought home to me by analysis of data which reflects my own poor efforts
in this regard.
A low-tech solutions to such recording can include detailed triangulation mapping
of sample locations using measuring tapes from a permanently marked baseline, and is thus
possible with simple equipment, although time consuming. Desalination of porous materials
can be done in rainwater, as long as some method of labelling artefacts to retain their
provenance information can be devised. Even cheaper, if this cannot be done, is to
encourage protection of the site at the local level, where possible, and to leave the items
where they are unless there is an immediate threat to their survival. Non-destructive in-situ
recording of form and decoration for research purposes is always an option if there is
sufficient skilled labour available. In some circumstances marine growth on sherds will
505

obscure the details of decoration, in which case laboratory cleaning might be required.
Broad fabric classifications could also be done in the field in many cases using a 10x hand
lens, if a small corner of each sherd is snapped, with a type-series retained for petrographic
analysis. These suggestions for practice may not all turn out to be entirely practical, but
are the range of possibilities I would experiment with in any future fieldwork of this sort.
The focus on formation processes, which involved treating sherds and lithics as
sedimentary particles, required descriptive systematics adapted to that purpose. Sherd size,
in particular, was recorded in more detail than is the norm in Pacific archaeology, as was
lithic manuport size and form. Point provenancing of artefacts, and recording of orientation
could be useful developments of this approach.
While the search for undisturbed buried villages continues in Lapita archaeology,
there has been little explicit consideration of the problem of what to do with these when
they are found (exceptions to this trend are noted in the introduction to Chapter 11).
Relatively well-preserved stilt-house remains have been found in the Arawes and Mussau,
and terrestrial buried landscapes have been reported from Manus and Vanuatu, but none
of these have been extensively excavated, because to do so would take an enormous
amount of time, money and skilled labour. Settlement pattern studies in the Reef-Santa
Cruz islands and Fiji using both the surface and excavation record have been rare
exceptions rather than the rule, enabled in part by shallow burial. Investigation of deeply-
buried places has tended to become a search for layercake chronology, largely due to the
difficulties of excavating anything other than a small fraction of the areas in a
reconnaissance mode. There is nothing intrinsically wrong with this, provided stringent
attention is paid to identification of formation processes, and age-depth relationships are
carefully constructed rather than assumed to be readable as time. These relatively small
holes will seldom be easily demonstrated to contain representative samples of variability
for any of the layers, due to limited spatial sampling and the difficulty of quantifying
506

samples, an area in which this thesis has sought to make some progress.
Well preserved ceramic sites yield a lot of sherdage, which can foster complacence
that the sample obtained from a small excavation is representative of the settlement as a
whole, both across space and through time. The analysis in Chapter 5 provides an example
showing how distantly related physical quantity of ceramics and sample size can be. It is
this aspect of a tendency to search for layer upon layer of buried villages that is the focus
of much of the critique in Chapter 2, and the argument is put forward in various places in
this thesis that the temporal resolution of a seriation chronology constructed from a large
number of large samples will almost inevitably surpass that obtained from reconnaissance
excavation squares.
In making external comparisons using the Roviana data above, the suggestion is
put forward that some of the difficulty in matching the Roviana series to that from other
areas may be to do with differences in temporal resolution, where horizontal structure is
separating out styles that are mixed in other buried contexts. An argument for large area
excavations and surface collection has been championed by Green for many years, and in
the lithics chapter (Chapter 10) it is clear that the stone adze sample from Lapita-era sites
in Near Oceania is dominated by materials from surface collections (this probably holds
true for remote Oceania also, if large-area excavations are counted as equivalent in the
scope of sampling to surface collections).
I put it to research archaeologists interested in Lapita that this is probably true also
of pottery, it is just that we have been lax about quantifying sample size, as few have been
willing to battle with the problem of establishing vessel brokenness and completeness, and
to develop methods of assessing sample size in various sampling contexts using these
properties of samples. For these reasons we should value the surface record, including lag
deposits in the sea, especially in Near Oceania (where settlement in such locations in the
past is now well documented), as a source of large, easily acquired samples of ceramics
and lithics, that can potentially form the backbone of high-resolution chronologies on
507

which so much behavioural inference depends.
Landscapes which offer surface exposure of samples of Lapita-age materials
sufficiently large to allow exploration of ceramic variability, of which the New Georgia
region is one, should not be too hastily bypassed in the noble quest for Wobst’s mythical
single point where all the region’s culture-historical complexes can be found superposed,
separated by sterile deposits.
The End
508

REFERENCES
Allen, J., 1984. Pots and Poor Princes: A Multidimensional Approach to the Role of
Pottery Trading in Coastal Papua. In S. E. Van Der Leeuw and A. C. Pritchard
(eds) The Many Dimensions of Pottery: Ceramics in Archaeology and
Anthropology. Amsterdam: Universiteit van Amsterdam,
Allen, J., Gosden C., 1991. Report of the Lapita Homeland Project. Canberra:
Department of Prehistory, Research School of Pacific Studies, ANU.
Allen, J., Gosden C., White J. P., 1989. Human Pleistocene Adaptations in the Tropical
Island Pacific: Recent Evidence from New Ireland, a Greater Australian Outlier.
Antiquity 63: 548-61
Allen, J., Specht J., Ambrose W. R., Yen D. E., 1984. Lapita Homeland Project: Report
of the 1984 Field Season Department of Prehistory, Research School of Pacific
Studies, ANU, Canberra
Ambrose, W., 1991. Lapita or Not Lapita: The Case of the Manus Pots. In J. Allen and C.
Gosden (eds) Report of the Lapita Homeland Project. Canberra, Australia:
Department of Prehistory, Research School of Pacific Studies, The Australian
National University, pp.103-12.
Ambrose, W., Gosden C., 1991. Investigations on Boduna Island. In J. Allen and C.
Gosden (eds) Report of the Lapita Homeland Project. Canberra, Australia:
Department of Prehistory, Research School of Pacific Studies, The Australian
National University, pp.182-8.
Anderson, A., 2002. Mobility Models of Lapita Migration. In G. R. Clark, A. J. Anderson
and T. Vunidilo (eds) The Archaeology of the Lapita Dispersal in Oceania:
Papers from the 4th Lapita Conference, June 2000, Canberra, Australia.
Canberra: Pandanus Press, Research School of Pacific and Asian studies, ANU,
pp.15-23.
509

Anson, D., 1983. Lapita Pottery of the Bismarck Archipelago and Its Affinities. PhD
thesis. University of Sydney.
---, 1986. Lapita Pottery of the Bismarck Archipelago and Its Affinities. Archaeology in
Oceania 21: 157-63
---, 1987. Reply to Kirch Et Al. Archaeology in Oceania 22: 127-8
---, 1990. Aspiring to Paradise. In M. Spriggs (ed) Lapita Design, Form and
Composition: Proceedings of the Lapita Design Workshop, Canberra, Australia,
December 1988. Canberra Australia: Department of Prehistory, Research School
of Pacific Studies, Australian National University, pp.53-8.
---, 2000. Reber-Rakival Dentate-Stamped Motifs: Documentation and Comparative
Implications. New Zealand Journal of Archaeology 20(1998): 119-35
Anthes, R. A., 1982. Tropical Cyclones, Their Evolution, Structure and Effects.
Meteorological Monographs No. 41. American Meteorological Society.
Arnold, D. E., 1985. Ceramic Theory and Cultural Process. Cambridge University Press.
Arqueologicas, P. N. D. P., 1970. Brazilian Archaeology in 1968: And Interim Report on
the National Program of Archaeological Research. American Antiquity 35 (1): 1-
23
Aswani, S., 1997. Customary Marine Tenure and Artisanal Fishing in the Roviana and
Vonavona Lagoons, Solomon Islands: The Evolutionary Ecology of Marine
Resource Utilization. PhD thesis. Anthropology, University of Hawaii.
Balfet, H., 1965. Ethnographic Observations in North Africa and Archaeological
Interpretation:The Pottery of the Maghreb. In F. R. Matson (ed) Ceramics and
Man. New York: Viking,
Bath, M., Deguara J., 2003. Southern Hemisphere Tropical Cyclones: Historical Tracks
and Data. Electronic maps prepared using Australian Bureau of Meteorology
data. http://australiasevereweather.com/cyclones/history.htm
Baxter, M. J., 1994. Exploratory Multivariate Analysis in Archaeology. Edinburgh
510

University Press.
Bedford, S., 1999. Lapita and Post-Lapita Ceramic Sequences from Erromango, Southern
Vanuatu. In J. C. Galipaud and I. Lilley (eds) The Pacific, 5000-2000 BP:
Colonizations and Transformations. Paris: IRD Editions, pp.127-37.
---, 2000. Pieces of the Vanuatu Puzzle: Archaeology of the North, South and Centre.
PhD thesis. Department of Archaeology and Natural History, Research School of
Pacific and Asian Studies, Australian National University.
---, 2002. Personal Communication: to M. Felgate. Auckland
Bellwood, P., 1978. Man's Conquest of the Pacific. Sydney: Collins.
Best, S., 1984. Lakeba: The Prehistory of a Fijian Island. PhD thesis. Anthropology
Department, University of Auckland.
---, 2002. Lapita: A View from the East. New Zealand Archaeological Association
Monograph 24. Auckland.
Bickler, S. H., 1998. Eating Stone and Dying: Archaeological Survey on Woodlark
Island, Milne Bay Province, Papua New Guinea. PhD thesis. Department of
Anthropology, University of Virginia.
Binford, L. R., 1962. Archaeology as Anthropology. American Antiquity 44: 455-70
---, 1964. A Consideration of Archaeological Research Design. American Antiquity 29:
425-41
Blankholm, H. P., 1991. Intrasite Spatial Analysis in Theory and Practice. Aarhus:
Aarhus University Press.
Braun, D. P., 1983. Pots as Tools. In J. A. Moore and A. S. Keene (eds) Archaeological
Hammers and Theories. New York, London: Academic Press, pp.107-36.
Brenchley, P. J., Harper D. T., 1998. Paleoecology: Ecosystems, Environments and
Evolution. Chapman and Hall.
Bruns, T. R., Cooper A. K., Mann D., Vedder J. G., 1986. Seismic Stratigraphy and
Structure of Sedimentary Basins in the Solomon Islands Region. In J. G. Vedder
511

(ed) Geology and Offshore Resources of Pacific Island Arcs - Central and
Western Solomon Islands. AAPG Circum-Pacific Council for Energy and Mineral
Resources, 4, pp.177-214.
Bunge, J., Fitzpatrick M., 1993. Estimating the Number of Species: A Review. Journal of
the American Statistical Association 88 (421): 364-73
Burley, D. V., 1994. Settlement Pattern and Tongan Prehistory. Journal of the Polynesian
Society 103 (4): 379-411
Burley, D. V., Dickinson W. R., 2001. Origin and Significance of a Founding Settlement
in Polynesia. Proceedings of the National Academy of Sciences 98 (20): 11829-31
Chao, A., 1984. Nonparametric Estimation of the Number of Classes in a Population.
Scandinavian Journal of Statistics 11: 265-70
Chao, A., Lee S.-M., 1992. Estimating the Number of Classes Via Sample Coverage.
Journal of the American Statistical Association 87 (417): 210-7
Clark, G. R., 1999. Post-Lapita Fiji: Cultural Transformation in the Mid-Sequence. PhD
thesis. Department of Archaeology and Natural History, Research School of
Pacific and Asian Studies, Australian National University.
Clarke, D. L., 1968. Analytical Archaeology. London: Methuen.
Colwell, R. K., 1997. Estimates: Statistical Estimation of Species Richness and Shared
Species from Samples. Connecticut: http://viceroy.eeb.uconn.edu/estimates.
Cowgill, G. L., 1972. Models, Methods and Techniques for Seriation. In D. L. Clarke (ed)
Models in Archaeology. London: Methuen, pp.381-424.
---, 1975. A Selection of Samplers: Comments on Archaeostatistics. In J. W. Mueller (ed)
Sampling in Archaeology. Tucson: The University of Arizons Press, pp.258-74.
Cowgill, G. L., Altshul J. H., Sload R. S., 1984. Spatial Analysis of Teotihuacan: A
Mesoamerican Metropolis. In H. Hietala (ed) Intrasite Spatial Analysis in
Archaeology. Cambridge University Press, pp.154-95.
Davenport, W., 1972. Preliminary Excavations on Santa Ana Island, Eastern Solomon
512

Islands. Archaeology and Physical Anthropology in Oceania 7 (3): 165-83
David, B., Wilson M., 1999. Re-Reading the Landscape: Place and Identity in NE
Australia During the Late Holocene. Cambridge Archaeological Journal 9 (2):
163-88
David, N., 1972. On the Life Span of Pottery, Type Frequencies, and Archaeological
Inference. American Antiquity 37 (1): 141-2
Davis, W. M., 1928. The Coral Reef Problem. American Geographic Society Special
Publication 9.
Day, T. J., 1980. A Study of the Transport of Graded Sediments. Wallingford, England :
Hydraulics Research Station.
de Barros, P., 1982. The Effects of Variable Site Occupation Span on the Results of
Frequency Seriation. American Antiquity 47 (2): 291-315
De Boer, W. R., 1983. The Archaeological Assemblage as Preserved Death Assemblage.
In J. A. Moore and A. S. Keene (eds) Archaeological Hammers and Theories.
New York: Academic Press,
Dickinson, W. R., 2000a. Petrography of Sand Tempers in Prehistoric Sherds from
Roviana Lagoon, New Georgia, Solomon Islands. Tucson: Unpublished Report,
University of Arizona, Geology Department, pp.10.
---, 2000b. Petrography of Sand Tempers in Prehistoric Watom Sherds and Comparison
with Other Temper Suites of the Bismarck Archipelago. New Zealand Journal of
Archaeology 20 (1998): 161-82
Dickinson, W. R., Green R. C., 1998. Geoarchaeological Context of Holocene Subsidence
at the Ferry Berth Lapita Site, Mulifanua, Upolu, Samoa. Geoarchaeology 13 (3):
239-67
Donovan, L. J., 1973. A Study of the Decorative System of the Lapita Potters in Reefs and
Santa Cruz Islands. MA thesis. Department of Anthropology, University of
Auckland.
513

Duff, A. I., 1996. Ceramic Micro-Seriation: Types or Attributes? American Antiquity 61
(1): 89-101
Dunkley, P. N., 1986. The Geology of the New Georgia Group, Western Solomon Islands
British Geological Survey Overseas Directorate, Keyworth, Nottinghamshire
Dunnell, R. C., 1970. Seriation Method and Its Evaluation. American Antiquity 35 (3):
305-19
---, 1971. Sabloff and Smith's "the Importance of Both Analytic and Taxonomic
Classification in the Type-Variety System". American Antiquity 36 (1): 115-8
---, 1978. Style and Function: A Fundamental Dichotomy. American Antiquity 43: 192-
202
---, 1986. Methodological Issues in Americanist Artifact Classification. In M. B. Schiffer
(ed) Advances in Archaeological Method and Theory. Academic Press, pp.149-
207.
---, 2001. Foreword. In T. D. Hurt and G. F. M. Rakita (eds) Style and Function:
Conceptual Issues in Evolutionary Archaeology. Westport, Connecticutt: Bergin
and Harvey, pp.xiii-xxiv.
Egloff, B. J., 1971. Collingwood Bay and the Trobriand Islands in Recent Prehistory.
PhD thesis. Australian National University.
---, 1973. A Method for Counting Ceramic Rim Sherds. American Antiquity 38 (3): 351-3
---, 1979. Recent Prehistory in Southeast Papua. Terra Australis. 4, Canberra:
Department of Prehistory, Research School of Pacific Studies, ANU.
Ericson, J. E., Read D. W., Burke C., 1972. Research Design: The Relationships between
the Primary Functions and the Physical Properties of Ceramic Vessels and Their
Implications for Distributions on an Archaeological Site. Anthropology UCLA 3
(2): 84-95
Feathers, J. K., 1997. The Application of Luminescence Dating in American Archaeology.
Journal of Archaeological Method and Theory 4 (1): 1-67
514

---, 2002. Luminescence Dating of Pottery Sherds from the Solomon Islands. Unpublished
Analysis Report, Seattle: University of Washington.
Felgate, M. W., 2002. A Roviana Ceramic Sequence and the Prehistory of Near Oceania:
Work in Progress. In G. R. Clark, A. J. Anderson and T. Vunidilo (eds) The
Archaeology of the Lapita Dispersal in Oceania: Papers from the 4th Lapita
Conference, June 2000, Canberra, Australia. Pandanus Press, Research School
of Pacific and Asian studies, ANU, pp.39-60.
Felgate, M. W., Bickler S. H., n.d. Pots as Species: Estimating a Parent Population of
Vessels from a Sherd Sample. Unpublished Manuscript in the Posession of the
Authors.
Felgate, M. W., Dickinson W. R., 2001. Late-Lapita and Post-Lapita Pottery Transfers:
Evidence from Intertidal-Zone Sites of Roviana Lagoon, Western Province,
Solomon Islands. In M. Jones and P. J. Sheppard (eds) Proceedings of the 2001
Austalasian Archaeolmentry Conference. Auckland: Research Papers in
Anthropology and Linguistics, pp.105-22.
Ferrari, B., Adams J., 1990. Biogenetic Modifications of Marine Sediments and Their
Influence on Archaeological Material. International Journal of Nautical
Archaeology and Underwater Exploration 19: 139-51
Flemming, N. C., 1983. Survival of Submerged Lithic and Bronze Age Artifact Sites: A
Review. In P. M. Masters and N. C. Flemming (eds) Quaternary Coastlines and
Marine Archaeology. Academic Press, pp.135-73.
Fontana, V., 1998. Procedures to Analyse Intra-Site Pottery Distribution, Applied to the
Neolithic Site of Fimon, Molino Casarotto (Italy), Housesite No.3. Journal of
Archaeological Science 25 (11): 1067-72
Frost, E. L., 1974. Archaeological Excavations of Fortified Sites on Tavenui, Fiji. Asian
and Pacific Archaeology Series. 6, Honolulu: Social Science Research Institute,
University of Hawaii.
515

Gaffney, V. L., Bintcliff J., Slapsak B., 1991. Site Formation Processes and the Hvar
Survey Project, Yugoslavia. In A. J. Schofield (ed) Interpreting Atrefact Scatters:
Contributions to Ploughzone Archaeology. Oxford: Oxbow Books,
Galipaud, J.-C., 1996. New Caledonia: Some Recent Archaeological Perspectives. In J.
M. Davidson, G. Irwin, B. F. Leach, A. Pawley and D. Brown (eds) Oceanic
Culture History: Essays in Honour of Roger Green. New Zealand Journal of
Archaeology Special Publication, pp.297-305.
---, 1998. The Lapita Site of Atanoasao: Malo, Vanuatu ORSTOM, Port Vila
Gardin, J. C., 1967. Methods for the Descriptive Analysis of Archaeological Material.
American Antiquity 32: 13-30
Gell, A., 1992. The Technology of Enchantment and the Enchantment of Technology. In
J. Coote and A. Shelton (eds) Anthropology, Art and Aesthetes. Oxford:
Clarendon Press, pp.40-63.
Gifford, E. W., Shutler R. J., 1956. Archaeological Excavations in New Caledonia.
Anthropological Records. 18(1), Berkeley and Los Angeles: University of
California Press.
Goldmann, K., 1971. Some Archaeological Criteria for Chronological Seriation. In F. R.
Hodson, D. G. Kendall and P. Tautu (eds) Mathematics in the Archaeological and
Information Sciences. Edinburgh: Edinburgh University Press,
Golson, J., 1971. Lapita Ware and Its Transformations. In R. C. Green and M. Kelly (eds)
Studies in Oceanic Culture History. Honolulu: Department of Anthropology,
Bernice P. Bishop Museum, pp.67-76.
---, 1991. Two Sites at Lasigi, New Ireland. In J. Allen and C. Gosden (eds) Report of the
Lapita Homeland Project. Canberra, Australia: Department of Prehistory,
Research School of Pacific Studies, The Australian National University, pp.244-59.
---, 1992. The Pottery from Lasigi, New Ireland. Presented at Poterie Lapita et
516

Peuplement: Actes Du Colloque Lapita, Janvier 1992, Noumea, New Caledonia
Gorecki, P., 1985. Test Excavation at Lemau, New Ireland Province and Report of an
Excursion to Langovai, New Hanover, New Ireland Province. In J. Allen (ed)
Lapita Homeland Project: Report of the 1985 Field Season. Melbourne: La Trobe
University, pp.20-2.
---, 1992. A Lapita Smoke Screen ? In J. C. Galipaud (ed) Poterie Lapita Et Peuplement:
Actes Du Colloque Lapita, Janvier 1992. Noumea, New Caledonia:
O.R.S.T.R.O.M., pp.27-47.
Gosden, C., 1985. An Archaeological Survey of the Arawe Islands, West New Britain.
Lapita Homeland Project: Report of the 1985 Field Season. Melbourne: La Trobe
University, pp.54-5.
---, 1989. Prehistoric Social Landscapes of the Arawe Islands, West New Britain Province,
Papua New Guinea. Archaeology in Oceania 24 (2): 45-58
---, 1991a. Learning About Lapita in the Bismarck Archipelago. In J. Allen and C. Gosden
(eds) Report of the Lapita Homeland Project. Canberra, Australia: Department of
Prehistory, Research School of Pacific Studies, The Australian National University,
pp.260-8.
---, 1991b. Towards an Understanding of the Regional Archaeological Record from the
Arawe Islands, West New Britain, Papua New Guinea. In J. Allen and C. Gosden
(eds) Report of the Lapita Homeland Project. Canberra, Australia: Department of
Prehistory, Research School of Pacific Studies, The Australian National University,
pp.205-16.
Gosden, C., Allen J., Ambrose W. R., Anson D., Golson J., et al, 1989. Lapita Sites of the
Bismarck Archipelago. Antiquity 63: 561-86
Gosden, C., Head L., 1994. Landscape - a Usefully Ambiguous Concept. Archaeology in
Oceania 29 (3): 113-6
Gosden, C., Webb J., 1994. The Creation of a Papua-New Guinean Landscape:
517

Archaeological and Geomorphological Evidence. Journal of Field Archaeology
21: 29-51
Gould, S. J., Vbra E., 1982. Exaptation- a Missing Term in the Science of Form.
Paleobiology 8: 4-15
Graves, M. W., 1998. The History of Method and Theory in the Study of Prehistoric
Puebloan Pottery Style in the American Southwest. Journal of Archaeological
Method and Theory 5 (4): 309-43
Graves, M. W., Cachola-Abad C. K., 1996. Seriation as a Method of Chronologically
Ordering Architectural Design Traits: An Example from Hawai'i. Archaeology in
Oceania 31 (1): 19-32
Green, R. C., 1974a. Excavation of the Prehistoric Occupations of Su-Sa-3. In R. C.
Green and J. Davidson (eds) Archaeology in Western Samoa: Volume II.
Auckland: Auckland Institute and Museum, pp.108-62.
---, 1974b. Review of Portable Artefacts from Western Samoa. In R. C. Green and J.
Davidson (eds) Archaeology in Western Samoa: Volume II. pp.245-74.
---, 1976a. Lapita Sites in the Santa Cruz Group. In R. C. Green and M. M. Cresswell
(eds) Southeast Solomons Culture History. Wellington, New Zealand: The Royal
Society of New Zealand, Bulletin 11, pp.245-65.
---, 1976b. A Late Prehistoric Sequence from Su' Ena Village, Uki. In R. C. Green and M.
M. Cresswell (eds) Southeast Solomons Culture History. Wellington, New
Zealand: The Royal Society of New Zealand, Bulletin 11, pp.181-91.
---, 1976c. A Late-Prehistoric Settlement in Star Harbour. In R. C. Green and M. M.
Cresswell (eds) Southeast Solomons Culture History. Wellington, New Zealand:
The Royal Society of New Zealand, Bulletin 11, pp.133-47.
---, 1978. New Sites with Lapita Pottery and Their Implications for an Understanding of
the Settlement of the Western Pacific. Working Papers in Anthropology,
Archaeology, Linguistics and Maori Studies. 51, Auckland, New Zealand:
518

Department of Anthropology, University of Auckland.
---, 1979. Lapita. In J. D. Jennings (ed) The Prehistory of Polynesia. Cambridge,
Massachusetts: Harvard University Press, pp.27-59.
---, 1990. Lapita Design Analysis: The Mead System and Its Use; a Potted History.
Presented at Lapita Design, Form and Composition: Proceedings of the Lapita
Design Workshop, Canberra, Australia, December 1988, Canberra Australia
---, 1991a. The Lapita Cultural Complex: Current Evidence and Proposed Models.
Bulletin of the Indo-Pacific Prehistory Association 11: 295-305
---, 1991b. Near and Remote Oceania -- Disestablishing "Melanesia" in Culture History.
In A. Pawley (ed) Man and a Half: Essays in Pacific Anthropology and
Ethnobiology in Honour of Ralph Bulmer. Auckland: The Polynesian Society,
pp.491-502.
---, 1991c. A Reappraisal of the Dating for Some Lapita Sites in the Reef/Santa Cruz
Group of the Southeast Solomons. Journal of the Polynesian Society 100: 197-
207
---, 1992. Definitions of the Lapita Cultural Complex and Its Non-Ceramic Component.
in Poterie Lapita et Peuplement: Actes Du Colloque Lapita, Janvier 1992, pp. 7-
21. Noumea, New Caledonia: O.R.S.T.R.O.M.
---, 1994. Prehistoric Transfers of Portable Items During the Lapita Horizon in Remote
Oceania: A Review. Ippa 15th Congress: Chiang Mai -Jan. 5-12.
Green, R. C., Anson D., 1991. The Reber-Rakival Site on Watom. Implications of the
1985 Excavations at the SAC and SDI Localities. In J. Allen and C. Gosden (eds)
Report of the Lapita Homeland Project. Canberra, Australia: Department of
Prehistory, Research School of Pacific Studies, The Australian National University,
pp.170-81.
---, 2000a. Archaeological Investigations on Watom Island: Early Work, Outcomes of
Recent Investigations, and Future Prospects. New Zealand Journal of Archaeology
519

20 (1998): 183-97
---, 2000b. Excavations at Kainapirina (SAC), Watom Island, Papua New Guinea. New
Zealand Journal of Archaeology 20 (1998): 29-94
Green, R. C., Davidson J. M., 1969. Description and Classification of Samoan Adzes. In
R. C. Green and J. M. Davidson (eds) Archaeology in Western Samoa. Auckland:
Auckland Institute and Museum, pp.21-32.
Hagstrum, M. B., Hildebrand J. A., 1990. The Two-Curvature Method for Reconstructing
Ceramic Morphology. American Antiquity 55 (2): 388-403
Hamilton, D., Woodward R., 1984. A Sunken 17th Century City: Port Royal, Jamaica.
Archaeology 37: 38-45
Hedrick, J. D., n.d. Archaeological Investigation of Malo Prehistory:Lapita Settlement
Strategy in the Northern New Hebrides. Unfinished draft PhD thesis. University
of Pennsylvania.
Henderson, A. G., 1998. Middle Fort Ancient Villages and Organizational Complexity
in Central Kentucky. PhD thesis. university of kentucky.
Henderson, H. M., 1992. The Strickler Site: The Analysis of a Nineteenth to Twentieth
Century Urban Farmstead in Markham County, Ontario (Canada). MA thesis.
University of Toronto.
Hendren, G. H., 1976. Recent Settlement Pattern Changes on Ulawa, Southeast Solomon
Islands. In R. C. Green and M. M. Cresswell (eds) Southeast Solomons Culture
History. Wellington, New Zealand: The Royal Society of New Zealand, Bulletin
11, pp.149-59.
Henrickson, E. F., 1990. Investigating Ancient Ceramic Form and Use. In W. D. Kingery
(ed) The Changing Roles of Ceramics in Society: 26,000 B.P. To the Present.
Westerville, Ohio: The American Ceramic Society, Inc., pp.83-117.
Henrickson, E. F., Mcdonald M. A., 1985. Ceramic Form and Function: An Ethnographic
Search and Archaeological Application. American Anthropologist 85: 630-43
520

Hodder, I., Orton C., 1976. Spatial Analysis in Archaeology. New Studies in
Archaeology. Cambridge University Press.
Hydrographic-Office, G. B., 1971. Pacific Islands Pilot Volume I: The Western Groups
Comprising the Territory of Papua Including the Louisiade Archipelago, the
North-East and North Coasts of New Guinea, the Solomon Islands, the Bismarck
Archipelago, and the Caroline and Marianas Islands. Hydrographer of the Navy.
Irwin, G., 1972. An Archaeological Survey in the Shortland Islands, B.S.I.P. MA thesis.
Anthropology Department, University of Auckland.
---, 1985. The Emergence of Mailu: As a Central Place in Coastal Papuan Prehistory.
Terra Australis. 10, Canberra: Department of Prehistory, Research School of
Pacific Studies, ANU.
---, 1992. The Prehistoric Exploration and Colonization of the Pacific. Cambridge
University Press.
Ishimura, T., 2002. In the Wake of Lapita: Transformation of Lapita Designs and Gradual
Dispersal of the Lapita Peoples. People and Culture in Oceania 18: 77-97
Jensen, K., Jensen J., Clegg C., 1999. Inferring Intensity of Site Use from the Breakdown
Rate and Discard Patterns of Fire-Cracked Rock at Playa View Dune. Utah
Archaeology 12: 51-64
Kintigh, K. W., 1984. Measuring Archaeological Diversity by Comparison with Simulated
Assemblages. American Antiquity 49 (1): 44-54
---, 1990. Intrasite Spatial Analysis: A Commentary on Major Methods. In A. Voorrips
(ed) Mathematics and Information Science in Archaeology: A Flexible
Framework. Bonn: Holos, pp.165-200.
Kirch, P. V., 1985. Archaeological Fieldwork on Eloaua and Emananus Islands, St.
Matthias Group, New Ireland Province. In J. Allen (ed) Lapita Homeland Project:
Report of the 1985 Field Season. Melbourne: La Trobe University, pp.14-9.
---, 1987a. Is There an Early Far Western Lapita Province? Sample Size Effects and New
521

Evidence from Eloaua Island. Archaeology in Oceania 22: 123-6
---, 1987b. Lapita and Oceanic Cultural Origins: Excavations in the Mussau Islands,
Bismarck Archipelago, 1985. Journal of Field Archaeology 14: 163-80
---, 1988a. Niuatopupapu: The Prehistory of a Polynesian Chiefdom. Monograph 5,
Seattle: Burke Museum, Seattle.
---, 1988b. Problems and Issues in Lapita Archaeology. In P. V. Kirch and T. L. Hunt
(eds) Archaeology of the Lapita Cultural Complex: A Critical Review. Seattle:
Thomas Burke Memorial Washington State Museum Research Report No. 5,
pp.167-5.
---, 1988c. The Talepakemalai Lapita Site and Oceanic Prehistory. National Geographic
Research 4 (3): 38-342
---, 1997. The Lapita Peoples: Ancestors of the Oceanic World. Blackwell.
---, 2001. Lapita and Its Transformations in near Oceania: Archaeological Investigations
in the Mussau Islands, Papua New Guinea, 1985-88. Berkeley: University of
California.
Kirch, P. V., Green R. C., 2001. Hawaiki, Ancestral Polynesia: An Essay in Historical
Anthropology. Cambridge University Press.
Kirch, P. V., Hunt T. L., 1988. The Spatial and Temporal Boundaries of Lapita. In P. V.
Kirch and T. L. Hunt (eds) Archaeology of the Lapita Cultural Complex: A
Critical Review. Seattle: Thomas Burke Memorial Washington State Museum
Research Report No. 5, pp.9-31.
Kirch, P. V., Hunt T. L., Weisler M., Butler V., Allen M. S., 1991. Mussau Islands
Prehistory: Results of the 1985-1986 Excavations. In J. Allen and C. Gosden (eds)
Report of the Lapita Homeland Project. Canberra, Australia: Department of
Prehistory, Research School of Pacific Studies, The Australian National University,
pp.144-63.
Kirch, P. V., Rosendahl P., 1973. Archaeological Investigation of Anuta. In D. E. Yen and
522

J. Gordon (eds) Anuta: A Polynesian Outlier in the Solomon Islands. Honolulu:
Bishop Museum,
Krantz, G. S., 1968. A New Method for Counting Mammal Bones. American Journal of
Archaeology 72: 286-8
Lampert, R. J., 1966. Archaeological Reconnaissance in Papua and New Guinea.
Canberra: Unpublished Mimeograph, Department of Prehistory, Research School
of Pacific Studies, ANU.
Le Blanc, S. A., 1975. Micro-Seriation: A Method for Chronologic Differentiation.
American Antiquity 40 (1): 22-38
Lepofsky, D., 1988. The Environmental Context of Lapita Settlement Locations. In P. V.
Kirch and T. L. Hunt (eds) Archaeology of the Lapita Cultural Complex: A
Critical Review. Seattle: Burke Museum, Seattle, pp.33-48.
Lilley, I., 1985. Archaeological Fieldwork in Duke of York Islands, East New Britain
Province. In J. Allen (ed) Lapita Homeland Project: Report of the 1985 Field
Season. Melbourne: La Trobe University, pp.41-2.
---, 1986. Prehistoric Exchange in the Vitiaz Strait, Papua New Guinea. PhD thesis.
Australian National University.
---, 1991. Lapita Sites in the Duke of York Islands. In J. Allen and C. Gosden (eds) Report
of the Lapita Homeland Project. Canberra, Australia: Department of Prehistory,
Research School of Pacific Studies, The Australian National University, pp.164-9.
Linton, R. E., 1925. Archaeology of the Marquesas Islands. Bernice P Bishop Museum
Bulletin 23. Honolulu: Bayard Dominick Expedition Bulletin Number 10.
---, 1944. North American Cooking Pots. American Antiquity IX (4): 369-80
Longacre, W. A., 1991. Sources of Ceramic Variability among the Kalinga of Northern
Luzon. In W. A. Longacre (ed) Ceramic Ethnoarchaeology. Tucson: University
of Arizona Press, pp.95-111.
Lyman, R. L., Wolverton S., O'Brien M. J., 1998. Seriation, Superposition, and
523

Interdigitation: A History of Americanist Graphic Depictions of Culture Change.
American Antiquity 63 (2): 239-61
Macmillan, D. H., 1966. Tides. New York: American Elsevier Publishing Company Inc.
Mann, P., Taylor F. W., Lagoe M. B., Quarles A., Burr G., 1998. Accelerating Late
Quaternary Uplift of the New Georgia Island Group (Solomon Island Arc) in
Response to Subduction of the Recently Active Woodlark Spreading Centre and
Coleman Seamount. Tectonophysics 295: 259-306
Marquardt, W. H., 1978. Advances in Archaeological Seriation. Advances in
Archaeological Method and Theory. Journal of Archaeological Method and
Theory, pp.257-314.
Mccoy, P. C., Cleghorn P. L., 1988. Archaeological Investigations on Santa Cruz
(Nendo), Southeast Solomon Islands: Summary Report. Archaeology in Oceania
23: 104-15
McNutt, C. H., 1973. On the Methodological Validity of Frequency Seriation. American
Antiquity 38 (1): 45-60
Mead, S. M., 1975. The Decorative System of the Lapita Potters of Sigatoka, Fiji. In S.
M. Mead, H. Birks, L. Birks and E. Shaw (eds) The Lapita Pottey Style of Fiji and
Its Associations. Wellington: The Polynesian Society, pp.19-43.
Mead, S. M., Birks L., Birks H., Shaw E., 1975. The Lapita Pottery Style of Fiji and Its
Associations. Wellington: The Polynesian Society.
Miller, D., 1985. Artefacts as Categories: A Study of Ceramic Variability in Central
India. New Studies in Archaeology. London: Cambridge University Press.
Miller, D., Roe D., 1982. The Solomon Islands National Sites Survey: The First Phase.
Bulletin of the Indo-Pacific Prehistory Association 3: 47-51
Mills, B. J., 1989. Integrating Functional Analyses of Vessels and Sherds through Models
of Ceramic Assemblage Formation. World Archaeology 21 (1): 133-47
Nojima, Y., 2002. Personal Communication: to M. Felgate.
524

O'Brien, M. J., Lyman R. L., 2000. Seriation, Stratigraphy and Index Fossils: The
Backbone of Archaeological Dating. New York: Kluwer Academic/Plenum.
Open University Oceanography Team, 1989. Waves, Tides and Shallow-Water Processes.
Oxford: Pergamon Press.
Oreihaka, E., 1997. Freshwater and Marine Aquatic Resources in Solomon Islands.
Orton, C., 1982. Computer Simulation Experiments to Assess the Performance of
Measures of Quantity of Pottery. World Archaeology 14 (1): 1-20
---, 1993. How Many Pots Make Five?- an Historical Review of Pottery Quantification.
Archaeometry 35 (2): 169-84
---, 2000. Sampling in Archaeology. Cambridge Manuals in Archaeology. Cambridge
University Press.
Orton, C., Tyers P., Vince A., 1993. Pottery in Archaeology. Cambridge Manuals in
Archaeology. Cambridge University Press.
Orton, C., Tyers P. A., 1991. Counting Broken Objects: The Statistics of Ceramic
Assemblages. Proceedings of the British Academy 77: 163-84
Parker, V. N. M., 1981. Vessel Forms of the Reef Islands Se-Rf-2 Site and Their
Relationships to Vessel Forms in Other Western Lapita Sites of the Reef/Santa
Cruz and Island Melanesian Area. MA thesis. Anthropology Department,
University of Auckland.
Petchey, F. J., 1995. The Archaeology of Kudon: Archaeological Analysis of Lapita
Ceramics from Mulifanua, Samoa, and Sigatoka, Fiji. MA thesis. Anthropology
Department, University of Auckland.
Phelan, M., 1997. Scratching the Surface: The Lapita Pottery of Makegur, Papua New
Guinea. MA thesis. School of archaeology, La Trobe University.
Plog, S., 1985. Estimating Vessel Orifice Diameters: Methods and Measurement Error. In
B. A. Nelson (ed) Decoding Prehistoric Ceramics. Carbondale and Edwardsville:
Southern Illinois University Press, pp.243-53.
525

Poulsen, J. I., Polach H. A., 1972. Outlier Archaeology: Bellona. A Preliminary Report on
Field Work and Radiocarbon Dates. Archaeology and Physical Anthropology in
Oceania 7: 184-214
Prior, C., 2000. Personal Communication: Rafter Radiocarbon Laboratory, Institute of
Geological and Nuclear Sciences, Lower Hutt, New Zealand, to M. Felgate.
Rathje, R., 2000. Lapita Development as the First Holiday Inns. In A. Anderson and T.
Murray (eds) Australian Archaeologist: Collected Papers in Honour of Jim Allen.
Canberra: Coombs Academic Publishing. pp 62-68.
Reeve, R., 1989. Recent Work on the Prehistory of the Western Solomons, Melanesia.
Bulletin of the Indo-Pacific Prehistory Association 9: 44-67
Rice, P., 1982. Pottery Production, Pottery Classification, and the Role of
Physicochemical Analyses. In J. S. Olin and A. D. Franklin (eds) Archaeological
Ceramics. Washington DC: Smithsonian Institute Press, pp.47-56.
---, 1987. Pottery Analysis: A Sourcebook. University of Chicago Press.
Robertson, I. G., 2001. Mapping the Social Landscape of an Early Urban Centre. PhD
thesis. Arizona State University.
Roe, D., 1992. Investigations into the Prehistory of the Central Solomons: Some Old and
Some New Data from Northwest Guadalcanal. In J.-C. Galipaud (ed) Poterie
Lapita Et Peuplement: Actes Du Colloque Lapita, Janvier 1992. Noumea, New
Caledonia: O.R.S.T.R.O.M., pp.91-101.
---, 1993. Prehistory without Pots: Prehistoric Settlement and Economy of Northwest
Guadalcanal, Solomon Islands. PhD thesis. Australian National University.
Rogers, G., 1973. Report on Archaeological Survey in Niuatoputapu Island, Tonga.
Working Papers in Anthropology, Archaeology, Linguistics and Maori Studies No.
28. Department of Anthropology, University of Auckland.
Ross, M. D., 1996. Pottery Terms in Proto Oceanic. In J. M. Davidson, G. Irwin, B. F.
Leach, A. Pawley and D. Brown (eds) Oceanic Culture History: Essays in Honour
526

of Roger Green. New Zealand Journal of Archaeology Special Publication, pp.67-
82.
Rouse, I. B., 1960. The Classification of Artifacts in Archaeology. American Antiquity 25:
313-23
Rowland, M. J., 1989. Population Increase:Intensification or a Result of Preservation?
Explaining Site Distribution Patterns on the Coast of Queensland. Australian
Aboriginal Studies 2: 32-42
Ruig, J., 1999. Collectors as Taphonomic Agents in the Archaeological Record. In M.
Mountain and D. Bowdery (eds) Taphonomy: The Analysis of Processes from
Phytoliths to Megafauna. Canberra: ANH publications: Australian National
University, pp.109-15.
Rye, O. S., 1981. Pottery Technology: Principles and Reconstruction. Manual on
Archaeology 4. Washington: Taraxacum.
Sand, C., 1997a. The Chronology of Lapita Ware in New Caledonia. Antiquity 71: 539-47
---, 1997b. Variété De L'habitat Ancien En Nouvelle-Calédonie: Étude De Cas Sur Des
Vestiges Archéologiques Du Centre-Nord De La Grande Terre. Journal, Société
des Océanistes - Paris 104: 39-66
---, 1998. Archaeological Report on Localities WKO013a and WKO013b at the Site of
Lapita(Kone, New Caledonia). Journal of the Polynesian Society 107: 7-33
---, 1999. Lapita and Non-Lapita Ware During New Caledonia's First Millenium of
Austronesian Settlement. In J.-C. Galipaud and I. Lilley (eds) The Pacific from
5000 to 2000 BP: Colonisation and Transformations. Patis: Editions de IRD,
pp.139-59.
---, 2000. The Specificities of the "Southern Lapita Province", the New Caledonian Case.
Archaeology in Oceania 35: 20-33
---, 2001. Evolutions in the Lapita Cultural Complex: A View from the Southern Lapita
Province. Archaeology in Oceania 36: 65-76
527

Schiffer, M. B., 1987. Formation Processes of the Archaelogical Record. Albuquerque:
University of New Mexico Press.
---, 1995a. Archaeological Context and Systemic Context. Behavioural Archaeology:
First Principles. Salt Lake City: University Of Utah Press, pp.25-34.
---, 1995b. Archaeology as Behavioural Science. Behavioural Archaeology: First
Principles. Salt Lake City: University Of Utah Press, pp.46-54.
---, 1995c. Toward the Identification of Formation Processes. In M. B. Schiffer (ed)
Behavioural Archaeology: First Principles. Salt Lake City: University Of Utah
Press, pp.171-95.
Schiffer, M. B., Sullivan A. P., Klinger T. C., 1978. The Design of Archaeological
Surveys. World Archaeology 10 (1): 1-28
Seber, G. A. F., 1986. A Review of Estimating Animal Abundance. Biometrics 42 (2):
267-92
Sergeantson, S. W., Gao X., 1995. Homo Sapiens Is an Evolving Species: Origins of the
Austronesians. In P. Bellwood, J. J. Fox and D. Tryon (eds) The Austronesians:
Historical and Comparative Perspectives. Canberra: Department of Anthropology,
Research School of Pacific and Asian Studies, Australian National University,
Shackley, M. L., 1974. Stream Abrasion of Flint Implements. Nature 248: 501-2
---, 1978. The Behaviour of Artefacts as Sedimentary Particles in a Fluvatile Environment.
Archaeometry 20: 55-61
Sharp, N. D., 1988. Style and Substance: A Reconsideration of the Lapita Decorative
System. In P. V. Kirch and T. L. Hunt (eds) Archaeology of the Lapita Cultural
Complex: A Critical Review. Seattle: Burke Museum, pp.61-81.
---, 1991. Lapita as Text: The Meaning of Pottery in Melanesian Prehistory. Bulletin of the
Indo-Pacific Prehistory Association 11: 323-32
Shennan, S., 1997. Quantifying Archaeology. Edinburgh: Edinburgh University Press.
Shepard, A. O., 1963. Ceramics for the Archaeologist. Washington DC: Carnegie
528

Institution of Washington.
Sheppard, P. J., 1993. Lapita Lithics: Trade/Exchang and Technology. A View from Thr
Reefs/Santa Cruz. Archaeology in Oceania 28: 121-37
---, 1996. Annual Report of the New Georgia Archaeological Survey 1996, Roviana
Lagoon Survey Year 1 Centre for Archaeological Research, University of Auckland
---, 2003. Personal Communication: to M. Felgate.
Sheppard, P. J., Felgate M., Roga K., Keopo J., Walter R., 1999. A Ceramic Sequence
from Roviana Lagoon (New Georgia, Solomon Islands). In J.-C. Galipaud and I.
Lilley (eds) The Pacific from 5000 to 2000 BP: Colonization and
Transformations. Paris: Editions de IRD, pp.313-22.
Sheppard, P. J., Green R. C., 1991. Spatial Analysis of the Nenumbo (Se-Rf-2) Lapita
Site, Solomon Islands. Archaeology in Oceania 26: 89-101
Shott, M. J., 1989a. Diversity, Organization, and Behaviour in the Material Record.
Current Anthropology 30 (3): 283-315
---, 1989b. On Tool-Class Use Lives and the Formation of Archaeological Assemblages.
American Antiquity 54 (1): 9-30
---, 1996. Mortal Pots: On Use Life and Vessel Size in the Formation of Archaeological
Assemblages. American Antiquity 61 (3): 463-82
---, 1998. Status and Role of Formation Theory in Contemporary Archaeological Practice.
Journal of Archaeological Research 6 (4): 299-329
Shott, M. J., Sillitoe P., 2001. The Mortality of Things: Correlates of Use Life in Wola
Material Culture Using Age-at-Census Data. Journal of Archaeological Method
and Theory 8 (3): 269-302
Shutler, R., Shutler M. E., 1964. Potsherds from Bougainville Island. Asian Perspectives
8: 181-3
Sinopoli, C. M., 1991. Approaches to Archaeological Ceramics. New York: Plenum
Press.
529

Siorat, J. P., 1990. A Technological Analysis of Lapita Pottery Decoration. in Lapita
Design, Form and Composition: Proceedings of the Lapita Design Workshop,
Canberra, Australia, December 1988, pp. 59-82. Canberra Australia: Department
of Prehistory, Research School of Pacific Studies, Australian National University
Skibo, J., Schiffer M. B., 1987. The Effects of Water on Processes of Ceramic Abrasion.
Journal of Archaeological Science 14: 83-96
Skibo, J. N., 1992. Pottery Function: A Use Alteration Perspective. Interdisciplinary
Contributions to Archaeology. New York: Plenum Press.
Smith, M. F., Jr., 1983. The Study of Ceramic Function from Artefact Size and Shape.
PhD thesis. University of Oregon.
---, 1985. Toward an Economic Interpretation of Ceramics: Relating Vessel Size and
Shape to Use. In B. A. Nelson (ed) Decoding Prehistoric Ceramics. Carbondale
and Edwardsville: Southern Illinois University Press, pp.254-309.
Snow, B. E., Shutler R., 1985. The Archaeology of Fuga Moro Island. Cebu City: San
Carlos Publications.
Sokal, R. R., Sneath P. H. A., 1963. Principles of Numerical Taxonomy. San Francisco:
W. H. Freeman and Co.
Solow, 1994. On the Bayesian Estimation of the Number of Species in a Community.
Ecology 75 (7): 2139-42
Spaulding, A. C., 1953. Statistical Techniques for the Discovery of Artefact Types.
American Antiquity 18: 305-14
Specht, J., 1968. Preliminary Report of Excavations on Watom Island. Journal of the
Polynesian Society 77: 117-34
---, 1969. Prehistoric and Modern Pottery Industries of Buka Island, T.P.N.G. PhD thesis.
Research School of Archaeology and Natural History, Australian National
University.
---, 1974. Lapita Pottery at Talasea, West New Britain. Antiquity 48: 302-6
530

---, 1977. Book Review - the Lapita Pottery Style in Fiji and Its Associations. Mankind 11
(1): 75-6
---, 1981. Obsidian Sources at Talasea, West New Britain, Papua New Guinea. Journal
of the Polynesian Society 90 (3): 337-56
---, 1985. Report on Archaeological Research in the Arawe and Kandrian Areas of West
New Britain Province. In J. Allen (eds) Lapita Homeland Project: Report of the
1985 Field Season. Melbourne: La Trobe University, pp.48-53.
---, 1991. Kreslo: A Lapita Pottery Site in Southeast New Britain, Papua New Guinea. In
J. Allen and C. Gosden (eds) Report of the Lapita Homeland Project. Canberra,
Australia: Department of Prehistory, Research School of Pacific Studies, The
Australian National University, pp.189-204.
---, 2002. Personal Communication: to M. Felgate. Noumea
Specht, J., Fields J., 1984. Frank Hurley in Papua: Photographs of the 1920-1923
Expeditions. Bathhurst, NSW, Australia: Robert Brown and Associates in
association with the Australian Museum Trust.
Specht, J., Fullagar R., Torrence R., 1991. What Was the Significance of Lapita Pottery
at Talasea? Bulletin of the Indo-Pacific Prehistory Association 11: 281-94
Spennemann, D. H. R., 1987. The Impact of Cyclonic Surge on Archaeological Sites in
Tonga. Bulletin of the Indo-Pacific Prehistory Association 7: 75-87
Spriggs, M., 1985. Archaeological Fieldwork on Nissan Island, North Solomons Province.
In J. Allen (ed) Lapita Homeland Project: Report of the 1985 Field Season.
Melbourne: La Trobe University, pp.58-63.
---, 1990. The Changing Face of Lapita: Transformation of a Design. In M. Spriggs (ed)
Lapita Design, Form and Composition: Proceedings of the Lapita Design
Workshop, Canberra, Australia, December 1988. Canberra Australia: Department
of Prehistory, Research School of Pacific Studies, Australian National University,
pp.83-122.
531

---, 1991. Nissan, the Island in the Middle. Summary Report on Excavations at the North
End of the Solomons and the South End of the Bismarcks. In J. Allen and C.
Gosden (eds) Report of the Lapita Homeland Project. Canberra, Australia:
Department of Prehistory, Research School of Pacific Studies, The Australian
National University, pp.222-43.
---, 1993. Island Melanesia: The Last 10 000 Years. In M. Spriggs, D. Yen, W. Ambrose,
R. Jones, A. Thorne and A. Andrews (eds) A Community of Culture: The People
and Prehistory of the Pacific. Canberra: Department of Prehistory, Research
School of Pacific Studies, ANU, pp.187-205.
---, 1997. The Island Melanesians. The Peoples of Southeast Asia and the Pacific.
London: Blackwell.
---, 2000. The Solomon Islands as Bridge and Barrier in the Settlement of the Pacific. In
A. Anderson and T. Murray (eds) Australian Archaeologist: Collected Papers in
Honour of Jim Allen. Canberra: Coombs Academic Publishing, ANU,
---, 2001. Who Cares What Time It Is? The Importance of Chronology in Pacific
Archaeology. In A. Anderson, I. Lilley and S. O'Connor (eds) Histories of Old
Ages: Essays in Honour of Rhys Jones. Canberra: Pandanus Books, Research
School of Pacific and Asian studies, ANU,
Stewart, D. J., 1999. Formation Processes Affecting Submerged Archaeological Sites: An
Overview. Geoarchaeology 14 (6): 565-87
Stoddart, D. R., 1969a. Geomorphology of the Marovo Elevated Barrier Reef, New
Georgia. Philosophical Transactions of the Royal Society of London, Series B,
Biological Sciences 255 (800): 383-402
---, 1969b. Geomorphology of the Solomon Islands Coral Reefs. Philosophical
Transactions of the Royal Society of London, Series B, Biological Sciences 255
(800): 355-82
Summerhayes, G., 1996. Interaction in Pacific Prehistory. PhD thesis. School of
532

Archaeology, Faculty of Humanities, La Trobe University.
---, 2000a. Lapita Interaction. Terra Australis 15. Canberra: Department of Archaeology
and Natural History and Centre for Archaeological Research, Australian National
University.
---, 2000b. Recent Archaeological Investigations in the Bismarck Archipelago, Anir - New
Ireland Province, Papua New Guinea. Bulletin of the Indo-Pacific Prehistory
Association 19: 167-74
---, 2000c. What's in a Pot? In A. Anderson and T. Murray (eds) Australian
Archaeologist: Collected Papers in Honour of Jim Allen. Canberra: Coombs
Academic Publishing, ANU, pp.291-307.
---, 2001. Lapita in the Far West: Recent Developments. Archaeology in Oceania 36: 53-
63
---, 2002. Defining the Chronology of Lapita in the Bismarck Archipelago. In G. R. Clark,
A. J. Anderson and T. Vunidilo (eds) The Archaeology of the Lapita Dispersal in
Oceania: Papers from the 4th Lapita Conference, June 2000, Canberra,
Australia. Pandanus Press, Research School of Pacific and Asian studies, ANU,
pp.25-38.
Swadling, P. L., 1976. The Occupation Sequence and Settlement Pattern on Santa Ana.
In R. C. Green and M. M. Cresswell (eds) Southeast Solomons Culture History.
Wellington, New Zealand: The Royal Society of New Zealand, Bulletin 11,
pp.123-32.
Tani, M., Longacre W. A., 1999. On Methods of Estimating Ceramic Uselife: A Revision
of the Uselife Estimates of Cooking Vessels among the Kalinga, Phillipines.
American Antiquity 64 (2): 299-308
Tarlton, F. J., 1996. Effects of Southwest Pacific Tropical Cyclones on Archaeological
Sites. MA thesis. Anthropology Department, University of Auckland.
Terrell, J. E., 1976. Perspectives on the Prehistory of Bougainville Island, Papua New
533

Guinea: A Study in the Human Biogeography of the Southwestern Pacific. PhD
thesis. Anthropology, Harvard University.
Thurman, H. V., 1981. Introductory Oceanograpgy. Sydney: Charles E. Merrill Publishing
Company.
Torrence, R., Stevenson C., 2000. Beyond the Beach: Changing Lapita Landscapes on
Garua Island, Papua New Guinea. In A. Anderson and T. Murray (eds) Australian
Archaeologist: Collected Papers in Honour of Jim Allen. Canberra: Coombs
Academic Publishing, ANU, pp.324-45.
Torrence, R., White J. P., 2002. Tattooed Faces from Boduna Island, Papua New Guinea.
In G. R. Clark, A. J. Anderson and T. Vunidilo (eds) The Archaeology of the
Lapita Dispersal in Oceania: Papers from the 4th Lapita Conference, June 2000,
Canberra, Australia. Pandanus Press, Research School of Pacific and Asian
studies, ANU, pp.135-40.
Trigger, B., 1989. A History of Archaeological Thought. Cambridge University Press.
Turner, M., 2000. The Function, Design and Distribution of New Zealand Adzes. PhD
thesis. Anthropology, University of Auckland.
Vanderwal, R. L., 1973. The Torres Strait: Protohistory and Beyond. Occasional Papers.
Anthropology Museum, University of Queensland, pp.157-94.
Varien, M. D., Mills B. J., 1997. Accumulations Research: Problems and Prospects for
Estimating Site Occupation Span. Journal of Archaeological Method and Theory
4 (4): 141-91
Varien, M. D., Potter J. M., 1997. Unpacking the Discard Equation: Simulating the
Accumulation of Artefacts in the Archaeological Record. American Antiquity 62
(2): 194-213
Wahome, E. W., 1999. Ceramics and Prehistoric Exchange in the Admiralty Islands,
Papua New Guinea. PhD thesis. Australian National University.
Wandsnider, L., 1996. Describing and Comparing Archaeological Spatial Structures.
534

Journal of Archaeological Method and Theory 3 (4): 319-84
Ward, G. K., 1976. The Archaeology of Settlements Associated with the Chert Industry
of Ulawa. In R. C. Green and M. M. Cresswell (eds) Southeast Solomons Culture
History. Wellington, New Zealand: The Royal Society of New Zealand, Bulletin
11, pp.161-80.
Waterhouse, J. H. L., 1926. A Roviana Phrasebook. Sydney: Epworth Press.
Waters, M. R., 1992. Principles of Geoarchaeology: A North American Perspective.
University of Arizona Press.
Weisler, M. I., 2001. Lapita Rockshelters of Elouaua and Mussau Islands (Sites EHM,
EHN, EKO, EKP, and EKQ). In P. V. Kirch (ed) Lapita and Its Transformations
in Near Oceania. Berkeley: Archaeological Research Facility, University of
California, pp.146-74.
White, J. P., Coroneos C., Neall V., Boyd W., Torrence R., 2002. FEA Site, Boduna
Island: Further Investigations. In S. Bedford, C. Sand and D. V. Burley (eds) Fifty
Years in the Field. Essays in Honour and Celebration of Richard Shutler Jr's
Archaeological Career. Auckland: New Zealand Archaeological Association
Monograph 25, pp.101-8.
White, J. P., Downie J. E., 1980. Excavations at Lesu, New Ireland. Asian Perspectives
23 (2): 193-220
White, J. P., Gorecki P., O'Brien R., Anderson P., 1985. Excavations at Balof and
Explorations Elsewhere in the New Ireland Province, 17 August - 23 September
1985. In J. Allen (eds) Lapita Homeland Project: Report of the 1985 Field
Season. Melbourne: La Trobe University, pp.9-.
White, J. P., Harris M.-N., 1997. Changing Sources: Early Lapita Period Obsidian in the
Bismarck Archipelago. Archaeology in Oceania 32: 97-107
Wickler, S., 1995. Twenty-Nine Thousand Years on Buka: Long-Term Cultural Change
in the Northern Solomon Islands. PhD thesis. Anthropology Department,
535

University of Hawaii.
---, 2001. The Prehistory of Buka: A Stepping Stone Island in the North Solomons. Terra
Australis 16. Canberra: Department of Archaeology and Natural History and
Centre for Archaeological Research, Australian National University.
Wickler, S., Spriggs M., 1988. Pleistocene Human Occupation of the Solomon Islands,
Melanesia. Antiquity 62: 703-6
Willey, G. R., Phillips P., 1958. Method and Theory in American Archaeology. Chicago:
University of Chicago Press.
Wobst, H. M., 1983. We Cant See the Forest for the Trees: Sampling and the Shapes of
Archaeological Distributions. In J. A. Moore and A. S. Keene (eds)
Archaeological Hammers and Theories. Academic Press, pp.37-85.
Yen, D. E., 1976. Inland Settlement on Santa Cruz Island (Nendo). In R. C. Green and M.
M. Cresswell (eds) Southeast Solomons Culture History. Wellington, New
Zealand: The Royal Society of New Zealand, Bulletin 11, pp.203-24.
536