Dissertation Proposal: Design and ... - IMS - Universität Stuttgart

Dissertation Proposal:
Design and Implementation of a Data Model for a
Multifunctional Electronic Dictionary
Dennis Spohr
Dennis.Spohr@web.de
May 2006
1 Introduction and Motivation
The structure of electronic dictionaries is a central topic in lexicographic research, and a
variety of approaches have been pursued and implemented over the past decades. There
is, however, an apparent gap between dictionaries for human use and those designed
for applications of Natural Language Processing (NLP). On the one hand, there exist
dictionaries such as the Valency Dictionary of English (Vde; Herbst et al., 2004) which
are targeted solely at human users. The most obvious characteristic such resources
have is that they contain large amounts of unformatted text – in the sense that this
text does not contain any linguistic markup or meta-information – and very few (if
any) links between lexical entries. The consequence is that considerable effort has to
be made in order to make the contained information usable for NLP applications (see
e.g. Spohr, 2004; Asmussen and Ørsnes, 2005). On the other hand, there are resources
such as subcategorisation lexica for formal grammars (e.g. LFG, Lexical Functional
Grammar) or the German morphology dictionary ImsLex (Fitschen, 2004), which have
been designed specifically for one type of NLP application or cover exactly one sort of
linguistic information, i.e. they are very specific.
The main research aim of the proposed thesis is to fill exactly this gap between
dictionaries for human users and those designed for specific NLP applications, and hence
to make a contribution to the notion of a multifunctional electronic dictionary – as
suggested by Heid and Gouws (to appear) in the context of collocations – which is to
serve both human users and different NLP applications.
While the primary prerequisite that an electronic dictionary has to meet with respect
to NLP use is to employ a powerful, expressive and (ideally) standardised formalism, one
of the main requirements for a human use dictionary is that it should go beyond merely
being an electronic version of a printed dictionary. To make this possible, it is necessary
to capture the highly relational nature of lexical data contained in a dictionary, i.e. to
establish connections between the entities that are described. These connections are
however not limited to lexical items, but may also involve links between the descriptions
of such items, e.g. between syntatic and semantic descriptions. For those connections to
be fully explored, it is necessary to provide fast and easy retrieval access to the data in
the model without limiting the ways in which explicit (or even implicit) information can
be queried. In addition to the standard way in which lexical information contained in
an electronic dictionary is accessed, i.e. via the lemma, this issue involves the possibility
1

of non-standard access to lexical information, e.g. via combinations of morphological,
syntactic and/or semantic criteria (“retrieve the nominalisations of those verbs which
are synonyms of . . . ”). With these requirements at the centre of investigation, one could
move away from viewing an electronic dictionary primarily as a collection of lemmata
combined with illustrative material (which is what many dictionaries directed towards
human users are) and towards using it as a network-like instrument of linguistic research
which is capable of not only representing lexical knowledge but also serving as a tool to
identify new connections and relations between lexical items.
To cover the multilingual dimension of lexical knowledge and of many of its applications,
the research questions that have been addressed so far should be investigated with
a view on multilinguality, i.e. considering how the described phenomena behave across
various languages. The proposed dissertation would not, however, aim at providing a
new formalism of semantic description for multilingual NLP, but would rather involve
an analysis of existing formalisms and implementations (see below).
2 Research Questions
The research questions mentioned in the previous section are repeated below, grouped
according to the type of requirement they impose on the dictionary model:
• Linguistic requirements:
– capture the relational nature of lexical data and the interaction between different
levels of description
– generic modelling of linguistic phenomena, i.e. a modelling not bound to a
given single type of application
• User/Usage requirements:
– usability for both NLP applications and human users
– non-standard search and retrieval access
• Formal requirements:
– powerful and expressive standard formalism, e.g. OWL DL 1 (W3C recommendation;
Bechhofer et al., 2004)
– compatibility with current best practice and standards
– querying should be (largely) independent of the particular model structure
The aim of this research is to design and implement a dictionary model which is
to serve both NLP applications and human users, thus contributing to the notion of a
multifunctional electronic dictionary. In addition to this, the model should be populated
with data – ideally extracted from several existing resources – and evaluated according
to its capability of achieving this goal, e.g. by implementing exemplary interfaces to
(probably no more than) two NLP applications (e.g. parsing) and by designing views
on the data with respect to the interests of two different groups of users (e.g. learners
or translation).
1 OWL DL is the Description Logic sublanguage of the Web Ontology Language OWL, which combines
the expressiveness of OWL with the advantageous computational properties of Description Logics,
e.g. non-monotonicity and decidability (see Baader et al. (2003) for a detailed account of the properties
of different Description Logic languages and their use for various kinds of applications).
2

3 Background and Related Work
The Dictionary Writing System iLex (Erlandsen (2004); marketed by Erlandsen Media
Publishing) is a commercial product offering highly efficient query access to very large
XML (Schema) dictionaries. Some aspects of the query mechanism employed in iLex
could be useful for the tasks mentioned in the previous section.
Protégé (Knublauch et al., 2004) and the Sesame framework (Broekstra et al., 2002)
provide attractive solutions for creating, storing and querying Description Logic ontologies
by transforming them into a graph representation using RDF triples. In this
formalisation, stating queries on a model corresponds to specifying paths through this
graph. The use of these tools will be of considerable relevance to the implementation of
the dictionary model.
Mile (Calzolari et al., 2003) and the currently evolving Lexical Markup Framework
(Francopoulo et al., 2006) aim at describing standards for the definition of lexical
resources for NLP, in particular electronic dictionaries. However, they do not account
for the highly relational (i.e. network-like) character of lexical data. As was already
mentioned above, this relational aspect would be a central topic of investigation.
The recently introduced Lexical Systems (Polguère, to appear) represent a kind of
network-like (graph) model which can be generated from existing databases. A current
implementation has been extracted from the DiCo 2 database, mainly describing lexemes
according to lexical functions of Mel’čuk’s Meaning⇔Text-Theory (see e.g. Mel’čuk,
1970). A feature which seems to be absent from this approach, however, is the distinction
between object language and metalanguage vocabulary, which I believe is crucial
e.g. with respect to consistency checking.
Existing knowledge bases, such as WordNet (Fellbaum, 1998), Cyc (Lenat, 1995)
or Sumo (Niles and Pease, 2001), contain taxonomic information about words and the
concepts they denote. The classifications provided by such resources might be relevant
for the description of semantic relations (e.g. synonymy) between lexical items. Attempts
at combining subcategorisation information, WordNet, Cyc and VerbNet (Kipper
et al., 2000) have been described in Crouch and King (2005).
The Papillon multilingual dictionary architecture (Sérasset and Mangeot-Lerebours,
2001) is based on the DiCo formalisation and is thus a model providing various kinds of
relations between lexemes. Its account of multi-word expressions is however very limited
(very often as part of the entry of one of the components), as is query access to the data
contained in the dictionary.
The SIMuLLDA Lexical Database Application (Janssen, 2002) proposes a multilingual
architecture for dictionaries using Formal Concept Analysis (FCA) as basic formalism
for the definition of a Structured Interlingua. However, the focus in this work is
mainly on semantic description and the acquisition of concepts from existing dictionaries
(using FCA) and less emphasis is put on the interfaces between different linguistic
descriptions.
4 Own Work in the Field
In my Diploma Thesis (Spohr, 2005), I have designed a Description Logic data model for
a monolingual collocation dictionary, based on the Web Ontology Language OWL DL.
The main focus was on providing an adequate description of collocations and related
2 http://olst.ling.umontreal.ca/dicouebe/
3

phenomena, such as the modelling of collocation types and of morphosyntactic preferences
of collocations. However, this focus on collocations has often been at the cost
of a detailed account of other linguistic phenomena of lexicographic relevance. These
include, for instance, a detailed description of single-word lexemes and their relations to
other entities in the dictionary, e.g. morphological and lexical semantic relations.
Nonetheless, the work presented in the Diploma Thesis has been able to show that
OWL DL offers a number of attractive solutions for consistent modelling of lexical data,
which has been further emphasised in Spohr and Heid (2006) with respect to multilingual
data. Thus, I would use several aspects of the data model developed there as a starting
point and further elaborate on the representation of the relations between phenomena
from different levels of linguistic description, such as the connection between verbs and
their nominalisations. Further emphasis would be on the integration into the framework
of other NLP applications (e.g. specification of interfaces) and on fast and easy retrieval
access to the information in the model, in order to advance towards a thorough networklike
data model for a multifunctional electronic dictionary which is to serve both NLP
and human users.
In addition to this, I am currently involved in research activities with a group of
researchers in the Salsa project at Saarland University (Erk et al., 2003). One of the
main objectives is the modelling of a lexicon which provides adequate representations of
Salsa’s corpus data and their annotations as well as their relation to the classification
defined in the FrameNet project (Baker et al., 1998). The modelling of this lexicon
involves (frame-)semantic and syntactic descriptions as well as – to some extent – annotations
and classifications of certain kinds of metaphorical uses in corpus sentences.
The knowledge gained during these activities will certainly be extremely relevant to the
proposed task.
5 Structure of the Dissertation
5.1 Specific objectives and approaches
The specific steps towards achieving the goals mentioned in section 2 are listed below,
subdivided into several objectives.
Analysis of current research, especially in the fields of general linguistics, lexicography
and data modelling in order to get a clear picture of the requirements
and possibilities of lexical resource modelling, e.g. with respect to standards (see
above).
Design of a data model for a multifunctional electronic dictionary, primarily based
on the preceding analysis step. Emphasis will be on the modelling of the interrelation
of different levels of linguistic description (e.g. valence information of
morphologically related lexemes), focussing mainly on monolingual descriptions,
however, with a view to the requirements of multilingual lexical data.
Implementation of an application for usage and persistent storage of this data model
(e.g. database), which would most probably be based on the Sesame framework
(cf. Broekstra et al., 2002). Further effort would go into the design of interfaces
for connecting the data model to the frameworks of other NLP applications.
Population of the model in order to be able to evaluate the model to some extent
and to provide a solid basis for the following steps. This objective could probably
4

draw on data from Vde (for valence descriptions; Herbst et al., 2004), FrameNet
(for semantic features and especially valence patterns; Baker et al., 1998), Salsa
(similar to FrameNet, but for German; Erk et al., 2003), GermaNet (Hamp
and Feldweg, 1997) and (Euro)WordNet (for taxonomic relations among concepts;
Fellbaum, 1998), ImsLex (for morphological information; Fitschen, 2004),
and possibly further corpus data available at IMS (cf. Ritz, 2005). The information
extracted from these resources would have to be merged in order to be fully
exploited. One of the ways to do this has been described by Crouch and King
in their approach to building a Unified Lexicon by merging WordNet, Cyc and
further external resources (cf. Crouch and King, 2005). Integrating data from
different resources with focus on (largely) different aspects of linguistic analysis
would be ideal for the evaluation of the model’s ability to capture the interrelation
of different levels of linguistic description.
Enhancement of the querying facility for easier access to the information in the
model. The intention would be to be as generic as possible, i.e. independent of
a particular model structure. Thus, a user or program should be able to query
the underlying model even with minimal prerequisite knowledge of its structure.
Instead of designing a single static query language for this purpose, a possible
approach could be to develop some kind of semi-automatic query language generator
which allows for the definition of path aliases or short-cuts to information
in the model. This objective would require an analysis of graph traversal and
concordance computation algorithms (see e.g. Bohnet (to appear)).
Design of views on the data in the model as well as the presentation thereof, depending
on the needs and interests of a particular application (see above) or group of
users. This would very likely be limited to the specification of no more than two
of each kind, with comments on the general procedure.
5.2 Possible schedule
The table below proposes a schedule for the dissertation. The actual procedure might
deviate from this planning, as some of the objectives might require simultaneous processing.
However, the rough order should correspond to the one specified below.
Year
Objective 1 st 2 nd 3 rd
Design of the dictionary data model × (×)
Implementation and model population (×) ×
Enhancement of the querying facility × (×)
Design of views on the data ×
Writing (×) (×) ×
5

References
Asmussen, Jørg and Bjarne Ørsnes, 2005. Valency information for dictionaries and NLP
lexicons: Adapting valency frames from The Danish Dictionary to an LFG lexicon.
In Ferenc Kiefer and Júlia Pajzs (eds.), Proceedings of the 8th Conference on Computational
Lexicography (COMPLEX 2005) – Papers in Computational Lexicography.
Budapest, Hungary.
Baader, Franz, Diego Calvanese, Deborah L. McGuinness, Daniele Nardi, and Peter F.
Patel-Schneider, 2003. The Description Logic Handbook: Theory, Implementation and
Applications. Cambridge, UK: Cambridge University Press.
Baker, Collin F., Charles Fillmore, and John B. Lowe, 1998. The Berkeley FrameNet
project. In Proceedings of COLING-ACL. Montreal, Canada.
Bechhofer, Sean, Frank van Harmelen, Jim Hendler, Ian Horrocks, Deborah L. McGuinness,
Peter F. Patel-Schneider, and Lynn Andrea Stein, 2004. OWL Web Ontology
Language Reference. Technical report. W3C. http://www.w3.org/TR/owl-ref/. Last
accessed: 15.05.2006.
Bohnet, Bernd, to appear. Textgenerierung durch Transduktion linguistischer Strukturen.
Doctoral Dissertation. Institut für Intelligente Systeme, Universität Stuttgart,
Germany.
Broekstra, Jeen, Arjohn Kampman, and Frank van Hermelen, 2002. Sesame: A Generic
Architecture for Storing and Querying RDF and RDF Schema. In Ian Horrocks and
James Hendler (eds.), Proceedings of the First International Semantic Web Conference
(ISWC 2002). Sardinia, Italy.
Calzolari, Nicoletta, Francesca Bertagna, Alessandro Lenci, and Monica Monachini,
2003. Standards and Best Practice for Multilingual Computational Lexicons & MILE
(the Multilingual ISLE Lexical Entry). ISLE Computational Lexicon Working Group,
Università di Pisa, Italy.
Crouch, Dick and Tracy Holloway King, 2005. Unifying Lexical Resources. In Katrin
Erk, Alissa Melinger, and Sabine Schulte im Walde (eds.), Proceedings of the Interdisciplinary
Workshop on the Identification and Representation of Verb Features and
Verb Classes. Saarbrücken, Germany.
Erk, Katrin, Andrea Kowalski, and Manfred Pinkal, 2003. A corpus resource for lexical
semantics. In Harry Bunt, Ielka van der Sluis, and Roser Morante (eds.), Proceedings
of the fifth International Workshop on Computational Semantics (IWCS-5). Tilburg,
The Netherlands.
Erlandsen, Jens, 2004. iLex – new DWS. In Pavel Smrz, Philippe Climent, Steve Crowdy,
Frantisek Cermak, Laura Elliott, Adam Kilgarriff, and Gilles-Maurice de Schryver
(eds.), Proceedings of the 3rd International Workshop on Dictionary Writing Systems
(DWS-2004). Brno, Czech Republic.
Fellbaum, Christiane, 1998. WordNet: An Electronic Lexical Database. Cambridge,
Massachusetts: MIT Press.
6

Fitschen, Arne, 2004. Ein Computerlinguistisches Lexikon als komplexes System. Doctoral
Dissertation. Arbeitspapiere des Instituts für Maschinelle Sprachverarbeitung
(AIMS), Volume 10(3). Institut für Maschinelle Sprachverarbeitung, Universität
Stuttgart, Germany.
Francopoulo, Gil, Monte George, Nicoletta Calzolari, Monica Monachini, Nuria Bel,
Mandy Pet, and Claudia Soria, 2006. Lexical Markup Framework (LMF). In Nicoletta
Calzolari, Khalid Choukri, Aldo Gangemi, Bente Maegaard, Joseph Mariani, Jan
Odijk, and Daniel Tapias (eds.), Proceedings of the 5th International Conference on
Lexical Resources and Evaluation (LREC-2006). Genoa, Italy.
Hamp, Birgit and Helmut Feldweg, 1997. GermaNet – a Lexical-Semantic Net for
German. In Piek Vossen, Geert Adriaens, Nicoletta Calzolari, Antonio Sanfilippo,
and Yorick Wilks (eds.), Proceedings of the ACL Workshop on Automatic Information
Extraction and Building of Lexical Semantic Resources for NLP Applications. Madrid,
Spain.
Heid, Ulrich and Rufus H. Gouws, to appear. A model for a multifunctional electronic
dictionary of collocations. To appear in Proceedings of the 12th EURALEX International
Congress (EURALEX-2006). Turin, Italy.
Herbst, Thomas, David Heath, Ian Roe, and Dieter Götz, 2004. A Valency Dictionary
of English. Berlin/New York: de Gruyter.
Janssen, Maarten, 2002. SIMuLLDA: a Multilingual Lexical Database Application using
a Structured Interlingua. PhD Thesis. Utrecht Institute of Linguistics OTS, Universiteit
Utrecht, The Netherlands.
Kipper, Karin, Hoa Trang Dang, and Martha Palmer, 2000. Class-based Construction
of a Verb Lexicon. In Harry Bunt, Ielka van der Sluis, and Roser Morante (eds.),
Proceedings of the Seventeenth National Conference on Artificial Intelligence (AAAI-
2000). Austin, TX, USA.
Knublauch, Holger, Ray W. Fergerson, Natalya F. Noy, and Mark A. Musen, 2004.
The Protégé OWL Plugin: An Open Development Environment for Semantic Web
Applications. In Sheila A. McIlraith, Dimitris Plexousakis, and Frank van Harmelen
(eds.), Proceedings of the Third International Semantic Web Conference - ISWC 2004 .
Hiroshima, Japan.
Lenat, Doug, 1995. Cyc: a large-scale investment in knowledge infrastructure. In CACM
38, No. 11 .
Mel’čuk, Igor A., 1970. Towards a functioning model of language. In Manfred Bierwisch
and Karl Erich Heidolph (eds.), Progress in Linguistics. Den Haag, The Netherlands:
Mouton.
Niles, Ian and Adam Pease, 2001. Towards a Standard Upper Ontology. In Chris Welty
and Barry Smith (eds.), Proceedings of the 2nd International Conference on Formal
Ontology in Information Systems (FOIS-2001). Ogunquit, Maine.
Polguère, Alain, to appear. Structural properties of Lexical Systems: Monolingual and
Multilingual Perspectives. Submitted to COLING/ACL Workshop on Multilingual
Language Resources and Interoperability. Sydney, Australia.
7

Ritz, Julia, 2005. Entwicklung eines Systems zur Extraktion von Kollokationen mittels
morphosyntaktischer Features. Diploma Thesis. Institut für Maschinelle Sprachverarbeitung,
Universität Stuttgart, Germany.
Sérasset, Gilles and Mathieu Mangeot-Lerebours, 2001. Papillon lexical database
project: Monolingual dictionaries & interlingual links. In Keh-Yih Su, Yuji Matsumoto,
and Tien-Shun Yao (eds.), Proceedings of the Sixth Natural Language Processing
Pacific Rim Symposium: NLPRS-2001 . Tokyo, Japan.
Spohr, Dennis, 2004. Using A Valency Dictionary of English to enhance the Lexicon of an
English LFG Grammar. Studienarbeit. Institut für Maschinelle Sprachverarbeitung,
Universität Stuttgart, Germany.
Spohr, Dennis, 2005. A Description Logic Approach to Modelling Collocations. Diploma
Thesis. Institut für Maschinelle Sprachverarbeitung, Universität Stuttgart, Germany.
Spohr, Dennis and Ulrich Heid, 2006. Modeling Monolingual and Bilingual Collocation
Dictionaries in Description Logics. In Paul Rayson, Serge Sharoff, and Svenja
Adolphs (eds.), Proceedings of the EACL Workshop on Multi-Word-Expressions in a
Multilingual Context (EACL-2006). Trento, Italy.
8