lexical and semantie relations

6 lexical and semantie relations
Keywords: PARADIGMATIC, Y TAGMATIC, SEMANTIC RELATIO ,
LEXICAL RELATION, SYNONYM, CONTRAST, ANTONYM,
COMPLEMENTARY, CO TRARY, CONVER E, HYPONYM,
HYPERO YM, MERONYM, HOLO YM, DICTIO ARY APPROACH,
THESAURUS APPROACH, EMA TI FIELD, LEXICAL BLOCKI G
6.1 Overview
This chapter examine particular emantic relation among words.
They are called semantic relations because they are relations between senses.
Some ca es of semantic relation can al o be lexical relations in which it is
not ju t the meaning that are related, but al o other a pect of the lexeme ,
like morphological form or collocational pattern . After looking at the detail of
synonymy, hyponymy, antonymy, and other relation, we evaluate two approache
to the representation of semantic relations in the mental lexicon. In the fir t
approach, the lexicon is theorized to be like a dictionary, which record sen e
but not nece arily relations among them. The econd view the lexicon like a
thesauru in which relation, but not meanings, are repre ented.
6.2 Paradigmatic relations:
synonymy, antonymy, hyponymy
6.2.1 Paradigmatic vs. syntagmatic
Relation among word can be divided roughly into two type :
paradigmatic and yntagmatic. Syntagmatic relations are relation between
words that go together in syntactic phra e - like ship 's and captain or dogs
and barko Notice that syntagmatic and syntax are from the same Greek root,
meaning 'touching together' - in other word, word in yntagmatic relations
"touch" each other in phra e . Because they go together in phra e, yntagmatically
related word often belong to different word cla ses - e.g. dog (noun)
+ bark (verb). Syntagmatic relations are studied more and more these day a
108
corpu research highlight the
word rather than other . For i
goes with certain modifier to i
or sound asleep, and that it oec
modifiers that indicate the am
others, like very as/eep. Our fo
tudy for lexical semantic : par
i ues in the later chapters - ineJ
adjective , in chapter ll.
Word in paradigmatic rels
orne characteristics in comme
form a paradigm - that i ,a I
paradigmatic relation i a morp
verb: drink, drank, drunk. otie
in common except their ten e..
involve word senses that hare I
for example, the et of ba ic co
each refer to a different part of ti
word, paradigmatically relatec
For example, red, white, and any
and grammatically occur in the
(l) a redlwhite/greenlblue
(2) a shade of redlwhite/g
Lexical semanticist study parac
relation among entence mean
of what they might tell u abor
ee in §6.3. They are also intere:
Using related words allow u t<
providing varied information an
A few paradigmatic relation
Synonymy i the relation of hal
are ynonyms in many people':
for example, house is a hypor
building. Co-hyponymy ( orne
of en e that make up a et, bt
heart/club/spade/diamond. Tha
type - in this case, playing card
which two words are opposite
the e relations in more depth in
While people of ten talk abou
accurate to talk about senses or II
being synonyms or antonym, i

ans
C RELATION,
M,
l,
PROACH ,
ILOCKJNG
n among words.
i between en es.
n in which it i
i of the lexemes,
g at the detai I of
:etwo approache
icon. In the fir t
:h records sense
he lexicon like a
l.
into two type :
ilations between
captain or dogs
me Greek roots,
gmatic relations
ses, syntagmate.g.
dog (noun)
re these day as
Lexical and semantie relations 109
corpu re earch highlight the way in which word tend to occur with certain
words rather than other. For in tance, we can notice that the adjective asleep
goe with certain modifier to indicate 'ab olute tate of sleep,' as in fast asleep
or sound asleep, and that it occasionally goe with orne other general-purpose
modifier that indicate the arne meaning, like completelyasleep, but le with
others, like very asleep. Our focus in thi chapter i the more traditional area of
tudy for lexical emantic : paradigmatic relation. We'll see orne yntagmatic
is ue in the later chapter - including the issue of which modi fiers go with which
adjectives, in chapter Il.
Words in paradigmatíc relations belong to the ame word cla s and hare
orne characteri tie in common. The word in uch relations can be aid to
form a paradigm - that i , a et of example that how a pattern. One kind of
paradigmatic relation i amorphological paradigrn, uch as the tense forms of a
verb: drink, drank; drunk. Notice that the ve rb in this paradigm have eve ry thing
in common except their ten e. We are interested in emantic paradigms, which
involve word sen e that hare many emantic propertie , but differ in orne. So,
for example, the et of ba ic color adjective form a paradigm who e members
each refer to a different part of the color spectrurn. Unlike syntagmatically related
words, paradigmatically related word are u ually ub titutable for each other.
For exarnple, red, white, and any other member of the color paradigm can en ibly
and grammatically occur in the ame phrases, as in (l) and (2).
(l) a red/white/green/blue house
(2) a hade of red/white/green/blue
Lexical emantici t tudy paradigmatic relation becau e of their roles in logical
relation arno ng entence meanings, uch a entailment (see §2.2), and because
of what they might tell us about how the mental lexicon is organized, as we'lI
ee in §6.3. They are alointeresting for their u e in creating coherent discourse.
U ing related word allow u to describe the ame things in different way, thu
providing varied information and avoiding repetition in di cour e.
A few paradigmatic relations receive the mo t attention in lexical emantics.
Synonymy i the relation of having (nearly) the ame meaning. Couch and sofa
are ynonym in many people' dialect. Hyponymy i the 'type-of' relation;
for example, house i a hyponym of building becau e a house i a type of
building. Co-hyponymy (ometime called lexical contrast) involves a group
of ense that make up a et, but which contra t with one another, for example
heart/club/spade/diamond. That i , they are different varietie within a single
type - in thi ca e, playing card uit. Antonymy i a peci al ca e of contra t in
which two word are oppo ite, for example black/white. We examine each of
these relation in more depth in the following sub ection .
While people of ten talk about word being ynonyms or antonyrn , it i more
accurate to talk about senses or lexical units (form + en e-in-u e - see § 1.2.4) as
being ynonym or antonym, ince a ingle word may have different ynonym

110 LEXICAL MEA ING
or antonym for different en es. For instance, when using the 'ternperature'
en se of hot, it oppo ite i cold. but when u ing the' picy' en e of hot, its
opposite is mild. For thi rea on, ernantic relation are ometimes called sense
relations.
6.2.2 Synonymy
The term synonym come from Greek roots syn 'alike' and onym
'name.' It refer to word that mean the ame as each other, o the equal sign =
is used to signal synonymy. Becau e it i rare for two word to have exactly the
same rneaning/u e, discussions of ynonymy frequently concern word that are
not perfect synonym , but that differ only slightly. The substitutability test is
u ed to determine whether two words are synonym . Word are ub titutable if
there is no change in the mean ing of a sentence when one word is ubstituted
for the other. So, for example, if the truth of (3) entails the truth of (4), and vice
ver a, then we have evidence that person and human are synonyrn
(3) A per on is standing beside me.
(4) A human is tanding be ide me.
If we want to te t whether man is a ynonyrn for per on, then we can compare
(5) and (3).
(5) A man is tanding be ide me.
In thi case, ince we can conclude that ometime it would be true that a person
is standing beside me at the ame time when it is fal e that a man is tanding
be ide me - since there things that can be referred to as person that can not be
called man - namely, women, girls, and boy . Hence man and person are not
ynonyms.
Absolute vs. sense synonyms
Words are said to be absolute syn ony ms if they are substitutable in any po sible
context with no changes in denotation or other a pect of meaning (including
connotation - see §2.2.2). U ing that criterion, it i ea y to see that very few
word are ab olute ynonym . Take for example funny = peculiar and funny =
comical. Where peculiar is substitutable for funny, as in (6), comical probably i
not ub titutable and vice ver a, a in (7).
(6) My turnrny feels a bit funny (= peculiar, 1comical) whenever l eat fish.
(7) Anna told a hilariously funny (l peculiar, = comical) joke.
Funny has different ynonym in different context beca u e it is polysemou ,and
its various en e match up semantically with different set of words. Since they
do not share all their en se .funny and pecu/iar are not ab olute ynonym , and
neither are funny and comical. In tead, they are sense synonyms, in that they
each have one sen e that mean the ame as one of the other word' en e .
For each of the followingpa
ynonyms by giving a enter
Describe why they are not 1
a. safe, ecure
b. fake. false
C. big, large
d. (a tough one!) Ol
The exi tence of large the:
are very close in meaning. :
to find pair that are perfectl
may feel that funny and co,
funny joke is a slightly diff
(1973:12-13) has gone of.
where two expres ion canr
be deceiving our elve ." Tt
technical names for thing
furze = gorse = whin, whic
American example i groutu
leave it to you to decide wh,
myself in claiming that the (
Near-synonyms
Far more common is for wo
[ake ~ false in Puzzle 6-1.
ynonyms (in a particular
contexts, but not every conte)
ub titution te t in (8), but fa
have a sense that roughly me:
are not shared by both word
(8) Ian obtainedlacquir
(9) a. Jan obtained permi
b. Ian acquired a Briti!
So far, we have seen that
being poly emou in differer
ynonym ) or by not having
onyms that are denotationally
way - for exarnple, by belo
different connotations. So, w

ng the 'ternperature'
icy' en e of hot, its
metime called sense
yn 'alike' and onym
o the equal ign =
I to have exactly the
mcern words that are
bstitutability test i
Is are ubstitutable if
_ word i ubstituted
truth of (4), and vice
nonym .
hen we can compare
be true that a person
at a man istanding
erson that can not be
I and person are not
table in any pos ible
meaning (including
lo ee that very few
eculiar andftll1ny =
comical probably is
whenever l eat fish.
)joke.
ti poly emou ,and
of word. Since they
rlnre ynonyms, and
onyms in that they
.worďs sense .
Puzzle 6-1
For each of the following pairs, demonsrrate that the two words are not absolute
Lexical and semantie relations 111
synonyms by giving a sentence in which they are not sub titutable for each other.
Describe why they are not substitutable (i.e. how their ense differ).
a. safe. secure
b. fake, false
c. big. large
d. (a tough one!) somebody, someone
The exi tence of large thesauru eprove that Engli h ha plenty of words that
are very clo e in meaning. But if we look closely at those words, it is very rare
to find pair that are perfectly ynonymou • even for just one of their senses. You
may feel that funny and comical never mean exactly the ame thing, o that a
[unny joke i a slightly different kind of joke than a comica/ joke. Roy Harris
(1973:12-13) ha gone o far as to claim that "If we believe there are instances
where two expre ion cannot be differentiated in re pect of meaning, we must
be deceiving our elve ." The rare candidate for perfect ynonymy tend to be
technical names for things like plants, animals, and chemicais. An example i
[ur;« = gorse = whin, which all name the ame European evergreen plant. An
American example i groundhog = woodchuck, which name the same animal. l
leave it to you to decide whether you think, as Harri would, that I'm deceiving
my elf in claiming that these are perfectly ynonymous.
Near-synonyms and variants
Far more common i for word' en e to overi ap and be near-synonyms, like
fake ~ ja/se in Puzzle 6-1. (We can use ~ to ignal near- ynonymy.) earsynonyms
(in a particular en e) can of ten substitute for each other in some
coruexts, but not every context. An example is obtain and acquire, which pas the
ubstitution te t in (8), but fail it in (9). This is because obtain and acquire both
have a sense that roughly means 'get,' but there are ubtleties to those senses that
are not shared by both words.
(8) Ian obtainedlacquired three diplomas.
(9) a.
b.
Ian obtained permi. sion to land. [?acquired]
[an acquired a British accent. I#oblained]
So far. we have seen that o-called ynonym may differ from one another by
being poly emou in different ways (i.e. en se ynonym that are not absolute
ynonyms) or by not having the exact arne denotation (near- ynonyms). Synonyms
that are denotationally identical may still be different in non-denotational
way - for exarnple, by belonging to different diaJect , regi ters. or by having
different con notation . So, we could ay that the synonym for toilet facilities

112 LEXICAL MEA ING
are not strictly substitutable because the sentences in (lO) would not be equally
appropriate in all ocia1 contexts.
(10) a.
b.
c.
Where is the john?
Where i the lavatory?
Where is the powder room?
John, la vato ry, and powder room are sense synonyms because they denote the
same things, but because they differ in register and connotation, they are variants
of one another. Synonyms can be variants with respect to any number of nondenotational
properties, including connotation, register, dialect, and affect. The
American sense of bathroom (which can refer to a room with a toilet but no ba th)
and British 100 (which refer to a room with a toilet) are ense near-synonyms,
but dialectal variants.
Why so many synonyms?
The moral of the synonym story is that it is very rare for two words to mean
exactly the ame thing. As Cruse (1986:270) has noted, "languages abhor ab olute
synonyms just as nature abhors a vacuum." English has many near-synonyms
because it has readily borrowed words from other languages and because English
productively makes new open-class word through the morphological processes
of derivation and compounding (§1.3.4) and conversion (§5.3.2).
In some case , those new or borrowed word tart out a perfect ynonym
for existing words in the language, but over time one of two things generally
happens. In some cases one of the ynonym "wins" and the other drop out of
use. For example, the Greek-derived Eucharist began to replace the Old English
husi toward the end of the Middle Age, and now husl is obsolete. In other ca e ,
the words' senses diverge. A famous example is what happened when Norman
French words for certain live tock animals were borrowed into Engli h in the
Middle Ages. Before that point, Engli h speakers had words for these animals
and used them also to describe meat from those animals; so they ate ox or sheep or
pig fiesh. After the French words for livestock became available, Engli h peakers
came to use the French livestock word (which became anglicized a beef, mutton,
pork) to refer to meat, while the native English words were retained for denoting
the animal. Other animal words were not affected by French borrowings, either
because the French word was not popularized in English (in the case of lamb)
or because the word was added to the language later (in the ca e of turkey).
Synonyms are also found in ca e where one word i reserved for technical
registers while the other sounds more folksy or poetic. For example, the Latinate
uterus is a more "medical" sounding word than womb, which is more likely to
be used in poetry.
Languages react to ynonyms in this way for a number of reason . First, it i
not economical or helpful to have many word for the ame thing - it takes more
effort to learn and remember two word form when one i sufficient. When we go
through the effort of having two words, we usually work under the a sumption
of a principIe of contrast - ti
different meanings. This mean
the same thing as another WOl
di fference - in denotation, conn
Clark and Clark (1979) give ex
noun into verbs results in a nl
for fo cook and to che! in (ll):
(Il) a.
b.
Phil cooked up a fea I
Phil chef'd up a feast.
To cook is an established verb,
denotes the ame activity a lo
a little different than cook. For
more professionalism or panacJ
In um mary, while language
words that are very close in n
can be seen as enriching, rathei
is aid to be particularly syna
more expressive than language
in meaning can be conveyed a
paralinguistic (e.g. intonational
6.2.3 Hyponymy and h~
Another type of pan
or are contained in, other mee
extension (§2.3. J) of one word ii
the extension of cheddar ia'
i cheddar is al o cheese, but ev
( ince it could be gouda or mo
cheddar is a type of cheese, a:
meaning of cheddar.
Properties of inch
The inclusion relation is asynu
is not a type of cheddar, o we I
relation: cheddar is a hyponyn
dar. The e terms come from the
over,' and this under/over inu
a "family trees," as illustrated
ification, or taxonomy, of foc
peaking, we use the terms hy!
relations between words like c
cheddar itself, or the nation of (

not be equally
hey denote the
~yare variants
umber of nonmd
affect. The
let but no bath)
iear- ynonyms,
word to mean
abhor absolute
near- ynonyms
ecau e English
gical processes
fect ynonyms
ung generally
tel'drops out of
:heOld English
. In other ca es,
when Norman
Engli h in the
r these animal
e ox or sheep or
ngli h speakers
as beef, mutton,
led for denoting
rrowings, either
e case of lamb)
:a e of turkey).
:d for technical
ile, the Latinate
; more likely to
ans. First, it is
~- it takes more
nt, When we go
the as umption
Lexical and semantie relations l 13
of a principie of contrast - that different lingui tic form are associated with
different meanings. This means that if we come across a word that seem to mean
the same thing a another word we already know, we expect it to have orne
difference - in denotarion, connotation, or social meaning - from the other word.
Clark and Clark (1979) give examples of when the common proce of making
nouns into verbs result in a new lexerne that contra ts with the extant verb, a
for to cook and to chef in (l l):
(Il) a. Phil cooked up a feast.
b. Phil cheťd up a feast.
Tocook is an e tabli hed verb, but if we read a new ve rb to ch~fthat ostensibly
denotes the same activity as to cook, then we presurne it mu t mean something
a little different than cook. For instance, one might interpret (l lb) as connoting
more professionalism or panache on Phil's part than (lIa) does.
In summary, while language resist synonyrny, they can nevertheless tolerate
words that are very close in meaning. Having words that overi ap in meaning
can be seen as enriching, rather than duplicating, our vocabulary. While English
is aid to be particularly synonyrn-rich, this should not be taken to say it is
more expressive than languages with fewer synonyms, since subtie differences
in mean ing can be conveyed a well through other yntactic, rnorphological, and
paralinguistic (e.g. intonational, gestural) means.
6.2.3 Hyponymy and hyperonymy: inclusion relations
Another type of paradigmatic relation involves meanings that contain,
or are contained in, other meanings - or, to put it a different way, when the
extension (§2.3. l) of one word is a subset of the extension of another. For example,
the extension of cheddar i a ubset of the extension of chee e; everything that
ischeddar is also cheese, but everything that is chee e is not necessarily cheddar
(since it could be gouda or rnozzarella or feta instead). We could say then that
cheddar is a type of cheese, and that the meaning 'cheese' is included in the
meaning of cheddar.
Properties of inclusion relations
The inclusion relation i asymmetrical; cheddar i a type of cheese, but cheese
i not a type of cheddar, o we need different names for the two directions in the
relation: cheddar i a hyponym of cheese, and cheese is a hyperonym of cheddar.
These terms come from the Greek-derived prefixes hypo- 'under' and hyper-
'over,' and this under/over imagery is useful in picturing hyponyrn paradigms
a "family trees" as illustrated in figure 6.1 which shows the y ternatic classification,
or taxonomy, of foods (with particular attention to chee e). Strictly
speaking we u e the terms hyponytn and hyperonym. when speaking about the
relations between word like cheese and cheddar. lf we talk about cheese or
cheddar itself, or the notion of CHE E S E or CHE O o AR (rather than the words for

114 LEXICAL MEA l G
food
I =--=
meat dairy product fruit vegetable
~
chee e bul/el' yogurt
Cheddar feta Edam Gouda
Figure 6.1 A partial taxonomy of food, particularlv chee e
grain
them), we refer to superordinate and subordinate categorie . So, we can ay
that the word cheese i a hyperonym of the word cheddar and that the category
CHEESE i uperordinatetothecategorYCHEDDAR. Wecanabbreviatethi with
< and >: cheddar < cheese and cheese> cheddar. The ymbol alway points
toward the smaller category. Figure 6.1 show that while one word may have
several hyponym , each hyponym has only one immediate (i.e. on the next level
up) hyperonym.
A well as being asymmetrical the inclusion relation i. often aid to be transitive,
which is to say that if X < Y and Y < Z, then X < Z. For example, cheddar
is a type of cheese and chee e i a type of food, therefore cheddar i a type of
food too. But this is where defining hyponymy get tricky, ince not all 'type-of'
relations are tran itive, as (12) shows:
(12) a.
b.
c.
a peci men cup (as used for urine samples) is a type of cup
a cup i a type of drinking ves el
#!! a peci men cup i a type of drinking ves el
The intransitivity of specimen cup < cup < drinking vessei tem from the fact
that the 'type-of' relations in (12a) and (l2b) are not the ame. The relation
in (l2a) involves proper inclusion; that is, the meaning cup is included in the
meaning specimen cup - a cup i a container with a wide, round opening and
a low profile that can hold liquid , and a peci men cup i a particular kind of
low container with a wide/round opening. On the other hand, (12b) i not proper
inclusion - the meaning of cup doe not include the meaning of drinking vesse/instead
a cup is an example of omething that can be u ed as a drinking ve sel.
Cru e (1986) calls the proper-inclusion type of hyponymy taxonymy ( ince
these are the relations found in cla ic taxonomies), while the ca e in (12b) can
be called functionaJ hyponymy, ince we can say 'is used a ' rather than 'i a
type of' in de cribing the relation. Beside functional hyponymy, there are other
types of hyponymy that many would not con ider to be 'true' hyponymy. A
Cruse (2000b) note , all queen are women, but iť a bit funny to aya queen
is a type of woman (but le funny to say a queen is a type of monarch). In
other words, we'd probably not li t queen in the same taxonomy in which we
have the term woman, so queen < woman i not a taxonymic relation, while
queen < monarch i . We can think of taxonyms as prototypical hyponyms,
ince they are reliably tran i
is a type of Y'
The inclusion/taxonym relation
ynonymy in light of the e tem
Tran itive or intransitive? l yo
near- ynonyms?
Folk vs. scientift
An important thing to notice
valuable to u on an everyda
hyperonym for cheese. In eve
i a type of food bul in lhe '
i a much-needed lexeme th.
chee e belongs to and the Ol
taxonorny that goe traight
while in more specialist con
with more fine-grained level
ation , we are happy to say
need to specify dog < cani
tel' how many level a taxo:
apply.
Even among the smaller rn
alient - i.e. more readily u I
cheese example, notice that if
to caJl it a cheese sandwich r;
of cheddar, you would proba'
have more specific informati
caJl it cheese rather than CI/(
general names in the e ituai
a dairy-product sandwich. T
that we generally attend to c
Rosch (1978) and other h
gui tie and cognitive term . li
we notice and can easily dií
tie level, the names for basi
one morpheme rather than
of the recognized propertie
table 6.1.

e/able grain
. So, we can say
that the category
breviate this with
101always points
word may have
on the next level
aid to be transi-
:xample, cheddar
ddar is a type of
:not all 'type-of"
p
m from the fact
ne. The relation
included in the
Ind opening and
uticular kind of
2b) i not proper
trinking vesseI -
drinking ve el.
rxonymy ( ince
a e in (12b) can
rather than 'is a
" there are other
hyponymy. A
r to aya queen
of monarch). In
ny in which we
relation, while
ica! hyponym ,
lexical and semantie relations I 15
ince they are reliably transitive and are easiest lo de eribe in the terms 'X
is a type of Y. '
Puzzle 6-2
The inclu ion/taxonym relation is asymmetrical and transitive. Reconsider
ynonymy in light of the e terms. Is ynonymy symmetrical or a ymmetrical?
Tran itive or intransitive? Is your an wef the same for perfect ynonym and
near- ynonym ?
Folk vs. scientific taxonomies and the basie level
An important thing to notice about taxonomies is that the level are not equally
valuable to u on an everyday basis. Dairy product i somewhat artificial as a
hyperonym for cheese. In everyday language we would probably ay that cheese
is a type of food, but in the language of supermarket manager, dairy product
i a much-need ed lexeme that allow. for a contra t between the category that
chee e belong to and the one that potatoes belong to. The everyday kind of
taxonomy that goe traight from cheese to food is called a folk taxonomy,
while in more speeialist context, we might need more elaborate taxonornies
with more fine-grained level of cIas ification. For instance, in everyday ituations,
we are happy to ay a dog is a type of animal, but a zoologi t might
need to specify dog < canine < mammal < vertebrate < animal. o matter
how many levels a laxonomy has, though, the same principie generally
apply.
Even among the smaller number of level in folk taxonomy one level i more
alient - i.e. more readily used and noticed - than the others. Sticking with the
cheese example, notice that if you make a sandwich out of cheddar, you are likely
to cali it a cheese andwicli rather than a cheddar sandwich. lf you saw a picture
of cheddar, you would probably say it wa a picture of cheese. Even though you
have more pecific information about the nature of the ehee e. you are likely to
cali it cheese rather than cheddar in many situations. We would not u e more
general name in the e ituations either; you would not ay CI food sandwich or
Q dairy-product sandwich. There i a level between too peeifie and too general
that we generally attend to, called the basie level.
Ro ch (1978) and other have noted that the basie level is pecial, both in linguistic
and cognitive terms. In cognitive term ,the ba ie level i the level at whieh
we notice and can ea ily differentiate between type of thing. At the lingui -
tic level, the name for basie level items tend to be more ba ic them elves -
one morpheme rather than more - and are the term we learn fir t. Some
of the recognized properties of basie level item /term are ummarized in
table 6.1.

116 LEXICAL MEANING
Table 6.1 Properties of the basic-level category CHEESE and the basic-Ievel
word cheese
TypicaI properties of basíc-level categories
Similarity of food (superordinate): different shapes
shape cheeses (ba ic level): similar hapes
cheddar ( ubordinate): shape that are indistinguishable from
one another
Similarity of
motor
movements
food: meat, cheese, ice cream, nut (erc.) are prepared and eaten
with different motions
cheeses: prepared and eaten with similar motions
cheddar : prepared and eaten with identical motions
Typical properties of basic-level words
Most frequently food: used when referring to mixed collections of food (e.g. the
used contents of a grocery bag), but rarely used to refer to cheese
speei fieally
cheese: oeeurs about 5 times per million words in English
cheddar: oecurs less than 1 time per mill ion words in English
(Dahl 1979)
Most readily
used
Morphologieally
simple
Earliest learned
If shown a pieture of some cheddar, most people say it's cheese,
rather than iťs food or it's cheddar.
superordinate: dairy product = 2 morphemes
basie-level: cheese = I
subordinate: we of ten use eompounds like cheddar cheese
(2 morpheme ) or goat's chee e (3 morphemes)
Children learn to name cheese before hyperonyrns jbozŕ or dairy
product or hyponym cheddar.feta, etc.
Thus we can conclude that while we describe things at many different taxonornic
levels, alllevels are not created equal. The basie levej is perceptually more
salient, and therefore the most unmarked (i.e. morphologically implest and mo t
broad ly used) level, lingui tically speaking.
Hyponymy and word classes
So far, all of our examples of hyponymy have been nou ns, and this is typical
in discussions of hyponymy, raising the question of whether inclusion relations
exist in other word cias es. Certainly, it i more natural to u e noun in X is a
type of Y statements, a shown in (13).
(13) a.
b.
c.
Cheddar is a type of cheese. [nouns]
? To march is a type of to walk. [verbs]
? Nauseated is a type of iiI. [adjectives)
Cru e (1986) propo e that
Xing is a way of Ying insn
copular ("linkjng") verb be
(14) a.
b.
Marching i a wa
Being nau eated
But while we can identif
taxonornies do not tend to
example, it is hard to thin
inc1usion chain marching <
a hyperonym for ili could b
state of health, but notice I
paradigmatic relation to ili
cross word-class boundarie
type of incJusion relation
Is hyponymy éI
While incJusion relations ar
tics, it is not clear that ucl
say that the relation betwee
of the relation between com
word s are in a hyponym re
related by inclusion relatio
meaning is only part of the
ub titutable) synonyms ut
and dialect as well a deno
to the e non-denotational ii
even though kitty and anii
cat is a type of animal JUI
pronouncement of X being
something called kitty is jl
thi rea on, it can be said
both a semantic (i.e. denot
since it involves similarity I
hyponymy is just a semanti
6.2.4 Incompatibilit
The relation di
lapping categories. Incom
can not denote the same thi
patible, a there are no fíov
link flower and equatiom
that constitute semantic par

le basic-level
guishable from
.epared and eaten
of food (e.g. the
) refer to cheese
.in English
ord in English
le say it 's cheese,
idar cheese
les)
rmsfood or dairy
y different taxoereeptually
more
imple t and most
Id thi i typical
eiu ion relations
: noun in X is a
Lexical and semantie relations 117
Cruse (1986) proposes that we test for hyponymy in verbs using the test sentence
Xing is a way of Ying instead. We could extend thi to adjectives by using the
copular ("linking") verb being, as in (14b).
(14) a. Marching is a way of walking.
b. Being nauseated is a way of being ill.
But while we can identify hyponyrn relations for verbs and adjective , their
taxonornie do not tend to have as many level as noun taxonomies have. For
example, it is hard to think of more verbs that could go on either end of the
inelusion chain marching < walking < moving and even harder to think of what
a hyperonyrn for iII could be. You might sugge t that it would be something like
state of health, but notice that state of health is a noun phrase, so not really in
paradigrnatic relation to iII. Lyon (1977) refers to relations like these, which
cross word-class boundaries, as quasi-hyponymy. In chapter 10 we look at other
types of inclusion relations for verbs.
Is hyponymy a lexical relation?
While inelu ion relations are definitely relevant to the discussion of lexical semantics,
it i not clear that such relations are specifieally lexical in nature. This is to
say that the relation between the word s cheese and cheddar i a direct reflection
of the relation between concepts (and the objeets) CHE ESE and CH EDDA R. The
words are in a hyponym relation imply because the things that they denote are
related by inclu ion relations. Cornpare thi to ynonyrny, for which denotative
meaning i only part of the story - we noted that word are not "good' (i.e. fully
ubstitutable) synonyrn unless they rnatch on is ue like con notation, register,
and dialect as well as denotative meaning. Hyponyrn relation are le s sensitive
to these non-denotational is ue . It i true to ay that a kitty is a type of animal,
even though kitty and animal differ in register. We might prefer to ay that a
cat is a type of animal, ju t because it is odd to u e a non-standard word in a
pronouncernent of X being a type of Y, but that doe not change the fact that
something cali ed ki/ty is just as much an animal as sornething cali ed cat. For
this rea on, it can be said that synonymy (and, as we shall see, antonyrny) is
both a semantic (i.e, denotational en se) relation and a lexical (word) relation,
sinee it involve imilarity on both denotational and non-denotational levels, but
hyponymy is just a ernantic relation.
6.2.4 Incompatibility, antonymy, and contrast
The relations di eu sed so far hold between words that denote overlapping
categorie . Incompatibility is the logical relation between words that
cannot denote the same thing. So, for example, fiower and equation are incompatible,
as there are no flowers that are al o equation - in fact, there is little that
links flowers and equations. More interesting to us are cases of incornpatibility
that con titute sernantic paradigrns. Of these, the most di cussed iantonymy, the

118
person
adult ~ child
mÔoman bo/"iirl
Figure 6.2 Age/sex taxonomy for person categories
relation of oppo itenes which hold between pair of word , uch a black/white,
old/young, and down/up. (We use a la h "I " to ignal incompatibility.) When
the paradigm involve more than two words, it i called a contrast set or a set
of co-hyponyms - that is, words that have the same hyperonym. We pay more
attention here to antonymy, since it ha particularly interesting properties.
Binarity
Antonymy i the only relation that is particularly binary- i.e. it holds between
pairs of words. Any word sense can have more than one synonym (e.g. fur;e =
gorse = whirú or hyponym (cheese> t cheddarígruyeretfeta, etc.l), but we tend
to think of word a having only one antonym in any particular en e. In orne
cases, this is becau e there are only two co-hyponym in a et, and o by default
the contrast set is a pair. This is the ca e in the taxonomy in figure 6.2. In this
case, adult/child, man/woman, and boy/girl can be con idered to be antonym pair
becau e they are contrast ets of two. Pair of antonym can also ari e through
morphological derivation. Engli h for example, ha several negating prefixes
that can be used to create antonym, uch a asymmetrical, infrequent, unhappy,
non-partisan, and dissatisfied.
In addition to these kind of "natural' binarity, antonyrn pairs ari e in ca e in
which a larger contra t et exi t . For instance black/white belong to the larger et
of achrornatic colors black/grey/white, and happy/sad al o contrast with ang ry,
afraid, and surprised. Nevertheles , we oppo e black/white and happy/sad a
if there were no other possibilitie . It seern that when people want to contra t
things, they like to have ju t one other item to contra t it with. Thi raise the
que tion of why we consider black the oppo ite of white instead of gre)', and why
sad is the oppo ite of happy rather than ang ry. You rnight an wer that black is
the opposite of white because it i more different from white than grey i , but that
an wer doe not work o well for the happy/sad/ang ry problem since sad and
angry are both very different from happy.
Minimal difference
The solution, on the contrary, is not how different the word are but how similar
they are. For instance, hot/cold make a better antonym pair than hot/cool becau e
the senses of hot and cold are more irnilar to each other than the en e of hot
and coal are. Antonymy i a relation between two lexerne that hare all relevant
properties except for one that causes them to be incompatible. Thi is to ay that
the rneanings of antonym
thi definition better than h
are both perception of ter
scale. They are different i
cale. HOl and coal are bo
temperature scale they are
componential analy i in C
than hot and cold are.
(15) hol
+ ternperature
+ high
[
+ extreme
Sirnilarly, white and black a
extrernes in the range of a<
op po ite extremitie of the
white and black and not a
contrasts with sad and ang/
in cornmon with sad, ince
with reversed facial expres
the e facial expression an
each other, they are imila
which are much more diffe
minimal difference allow
when two member within I
than to other member of th
Types of antor
There are several subtype I
Contra ry antonym are
a ertion of one entail the I
entail the assertion of the Ol
(16) a.
b.
Glady i tall. ~
Gordon i not tall
Not tall doe not entail shor,
that a thing or per on could t
in that they describe the t\
CA more sophi ticated ven
adjectives in chapter II.) E
be measured (e.g. l centim
of the cale include shon a
mea urements that are neith

s black/white,
.bility.) When
st set or a set
We pay more
perties.
iold between
I (e.g./Llrze =
), but we tend
ense. In some
loby default
re 6.2. In this
antonym pairs
arise through
ating prefi xes
ent, unhappy,
ise in cases in
) the larger set
t with angry,
happy/sad as
mt to contrast
:hi raises the
gre)', and why
r that black is
re)' i , but that
since sad and
ut how similar
t/cool because
; en es of hot
rre all relevant
s i to say that
Lexical and semantic relations 119
the meanings of antonyms are minimally different (Clark 1970). Hotlcold fits
this definition better than hot/cool, since hot and cold have in cornrnon that they
are both perception of temperature, and both are extremes on the temperature
scale. They are different in that they are on oppo ite ends of the temperature
scale. Hot and cool are both temperatures, but differ in both which side of the
temperarure cale they are on and whether or not they are extreme. A the basie
componential analysi in (IS) how, hot and cool are less similar to each other
than hot and cold are.
(15) hot cold cool
+ temperature]
+ high
[
+ extreme
+ temperature]
- high
[
+ extreme
+ temperature]
- high
[
- extreme
Similarly, white and black are antonyms because they are alike in that they denote
extremes in the range of achromatic colors, but different in that they are at the
opposite extremities of the color range, while grey is both a different color from
white and black and not at an extreme point on the scale. And while happy
contrasts with sad and angry and surprised and afraid, it seems to have the most
in cornmon with sad, since (arnong other reasons) happy and sad reftect states
with reversed facial expressions (smile/frown) and postu res (up/down). While
these facial exp re ion and postures are different, since they are reversals of
each other, they are similar (as opposed to expressions of anger and surprise,
which are much more different expressions). In all these cases the principIe of
rninimal difference allows for binary antonym pairs within a larger contrast set
when two members within the et are perceived to be more similar to one another
than to other members of the set.
Types of antonyms
There are several subtypes of antonyrn relation:
Contrary antonyms are th ose, like short/tall and old/young, in which the
assertion of one entails the negation of the other, but the negation of one does not
entail the as ertion of the other. as illustrated in (16):
(16) a. Gladys is tall. ~ Gladys is not short.
b. Gordon is not tall. -+> Gordon is short.
Not talf does not entaiI short because tall and short do not describe all the heights
that a thing or person could be. We can say that these contraryantonym are seaIar
in that they describe the two extremes of a cale, as illustrated in figure 6.3.
(A more ophisticated ver ion of thi cale is presented in the di cussion of
adjectives in chapter ll.) Each point on the line designates a height, which can
be measured (e.g, l centimeter, 2 cm, 3 cm, ... ). The left and right extremes
of the scale include short and tall height , but the middle area includes height
mea urements that are neither tall nor short. So, Gordon is not tall does not entaii

120 LEXICAL MEA I G
not tall
----------~--------~
~ _ ~
short tall
Figure 6.3 Height scale (preliminary version)
Gordon is short becau e he could have an unremarkable height in lead. Thu tall
and short con titute a contrary pair. Some author (e.g. Lyon 1977, Cru e 1986)
re erve the name antonym for preci ely thi type of scalar (or gradable) contrary
relation.
Cornplementary (alo cali ed contradictory) antonym are tho e for which
the assertion of one entails the negation of the other and vice versa, a in (17).
(17) a.
b.
That integer i even. ~ That integer i not odd.
That integer is not odd. ~ That integer i even.
In other word, the en e of complementary antonym completely bi ect some
domain: every integer is either odd or even; there is no midd1e ground.
ot all adjectival antonyms fit o nicely into the e categories. So-called gradable
complementaries, like dishonest/honest, lie between complementarity and
contrariety. They seem to contradict each other (X is not honest entaii X is
dishonest, and vice versa), but a middle ground seem to exi t, ince we can
a sert that ome person is neither honest nor dishonest. Even cla ic examples
of complementarity, like dead/alive, sometimes take on gradable qualitie (e.g.
he 's more dead than alive).
Cont.rariety and complementarity are of ten discussed with reference to adjectives
( ee also chapter ll), although we can find contrary and complementary
antonym in other part of speech. For example, Love/hale (noun or verb) are
contrarie ,wherea the verb stay/go are in a complementary relation.
Converse antonym describe the same relation or activity from different perspectives,
and follow pattern like: if X i p to Y, then Y i q to X. Sometime
one need to make adju tments to thi te t pattern in order to form a grarnrnatical
sentence, a the examples in (18) how. For example ince Bill (X) i a parent
(P) to Madeline (Y), Madeline (Y) i a child (q) of Bill' (X). In other word,
Bill and Madeline have. a relationship, and Bilľs position in the relationship is
parent, and Madeline' i child.
(18) a.
b.
c.
Bill i Madeline's parent. ~ Madeline is Bilľs child.
John gives to Oxfam. ~ Oxfam receives from John.
Scotland i above England. ~ England is below Scotland.
Reversíve oppo ite involve the undoing of some action: tie/untie, con truction/demolition.
Conver e and rever ive antonym can be collected, along with
other mi cellaneou example (e.g. come/go), in a general category of directional
antonyms.
Some word pairs that an
categorie . For example, le
but they are not different p
that typically co-occur. (1
semantics to the tudents
Other pairs of word that
minirnally different (they a
in being alcoholic/non-alo
the e to be proper antonyn
entailment relation like c
Some of these types of al
ness are nece sarily binary
or a relation with two per p
more generally in larger cr
ha to be adapted to allow
(19):
(19) a.
b.
That ace i a clul
That ace i not a
Determine whether the follo
converse, or no ne of these. E
a. fast/slow
b. student/teacher
c. clean/dirty
d. female/male
e. [eminine/masculu
Is antonymy a
Finally, there i the que ti
emantic relation or al o a I
of the word are relevant la
to be particularly lexical, Il
hould not make much diff
But a k anyone what the or
intuition is backed up by th
but large and Little do not (
for in tance, dead/alive ee
seems "better" than rise/des.
ju t meaning may come intr

stead. Thus tall
17,Cruse 1986)
dable) contrary
ho e for which
W, a in (17).
ely bisect some
ound.
So-calied gradlementarity
and
est entails X is
t, since we can
las ic examples
e qualities (e.g.
erence to adjeccomplementary
un or verb) are
ation.
m different per-
~X. Sometimes
] a granunatical
(X) is a parent
In other words,
! relationship is
untie, construc-
.ted, along wi th
y of directional
Lexical and semantie relations 121
Some word pair that are commonly considered opposites do not fit the above
categories. For example, learn/teach seem to go together in a conver e-like way,
but they are not different perspectives on the same action, but rather two actions
that typically co-occur. (They certainly do not alway co-occur. Lynne taught
semantics to the students does not entaii that the students leamed anything!)
Other pairs of words that go together, such as gin/tonic can be argued to be
minimally different (they are imilar in being part of the same drink, but different
in being alcoholic/non-alcoholic). However, many semanticists do not consider
these to be pro per antonyms SLnce they can not be described in terms of logical
entailment relations like complementary, contrary, and converse relations can.
Some of these types of antonyrny, including gradable contrariety and converseness,
are nece arily binary relation , a they require either a cale with two ends
or a relation with two perspectives. The notion of complementarity can be applied
more generally in larger co-hyponym sets, though the test for complementarity
has to be adapted to allow for more than one complementary possibility, a in
(19):
(19) a. That aee is a club. ---7 That aee is not a heart or a diamond or a spade.
b. That aee is not a club. ---7 That aee is a heart or a diamond or a spade.
Puzzle 6-3
Determine whether the following antonym pairs are eomplementary, contrary,
conver e, or none of these. Briefíy explain your answers.
a. fast/slow
b. student/teacher
c. cleanldirty
d. female!male
e. feminine/masculine
Is antonymy a lexical relation?
Finally, there is the question of whether antonymy, like hyponymy is just a
seman tie relation or al o a lexical relation, in which non-denotational properties
of the words are relevant to the relation as well. Some opposite pairings do seem
to be particularly lexical. If it were ju t the meaning that were relevant, then it
hould not make much difference whether big or large i the antonym of little.
But a k anyone what the opposite of little is, and they will tell you iť big. Thi
intuition i backed up by the fact that big and little occur together of ten in texts,
but large and little do not (Muehleisen ] 997). There are many similar examples;
for instance, dead/alive seems "better" than deceased/alive and ascend/descend
seems "better" than rise/descend. Since antonyms come in pairs, factors other than
just meaning may come into play in determining the "be t" antonym for a word,

122 LEXICAL MEA ING
for instance morphological or phonological sirnilarity (as for ascend/descenďy
and register (deceased/alive).
Because antonyms frequently co-occur in peech and wri ting, it is likely that we
learn antonym pairs like big/little and large/small and these get reinforced to the
extent that we prefer them to other possible oppo ites like large/little. In this case,
we are not just relying on our knowledge of the words' meanings to recognize
their oppositene s; we are also relying on our knowledge of which word go
together in language use. Some lexicologi ts use the term opposite to refer to the
semantic relation and an tonym pecifically for oppo ite that are al o lexically
related. We are les likely to learn co-hyponyrn relations as lexical relations. Since
whole large contrast sets do not co-occur in di course as often a opposite pairs
do, there is less opportunity to learn them as et of related word. For instance,
if l try to name the whole range of emotions, I get as far as happy/sad/ang ry, but
then am less sure about which other items go on that list. (Which hould lincJude:
afraid or [righteneďl Is surprised in thi contra t set?) However, some cases exi t,
like the playing card suit and scientific contrast sets, uch as solid/liquid/gas and
the prismatic colors, red/orange/yellow/green/blue/indigo/violet.
Further evidence that some opposites are also in a lexical antonym relationship
cornes from psycholinguistic experirnents. Pairs like high/low, happy/sad, and
black/white prime each other (§ 1.4.4), so that if you had just read happy, you
would recognize the word sad much quicker than if you had not read happy.
The same kind of test can show that a semantically opposite, but not lexically
antonymous, word like depressed would not be recognized as fast as sad (Becker
1980). People are also quicker and more con i tent in noticing the opposition
between pairs like happy/sad than happy/depressed (Charles et al. 1994). Thi
argues for a lexicological theory that repre ents the opposition of happy/sad in a
more accessible way than happy/depressed, who e opposition we have to figure
out on emantic grounds (see §6.3 below).
6.2.5 Other paradigmatic relations
Synonymy, hyponymy, and antonymy/contrast are the most important
relations for semantic theories, but they are not the only paradigmatic sernantic
relations among words. We could name many more, like the relation between
agents and the fields they work in (teacher-education, actor-theatrei, organizations
and their heads (team-captain; club-president, committee-chairi, or the
relations between animals and the sounds they make (cow-moo, cat-meow, pigoink).
Since these relations apply to relatively narrow ranges of vocabulary and
have fewer logical implications, linguists have not bothered to give them -onym
names.
An additional relation that has been named i the part-whole or 'has-a' relation,
meronymy. Like inclusion, this is an asymmetrical relation, so we say that
finger is a meronym of hand and hand is the holonym of finger. Also like
hyponyrny, meronymy does not rely on the lexical forrns of the words - it is a
direct reflex of the mea
be identified, such as ,
component (car> eng
substanee (pipe> copp
lexical/sernantic organi:
meronyms and holonyn
and their hyperonym . ~
the same part-name of
i always a part-name, b
doors, jugs, suitca e .ai
meronym relation can
or as consistent a the o
6.2.6 Summary
Table 6.2 u
Table 6.2 Properties of
Syn
semantic relation ml!
binary X
syrnrnetrical J
transitive J
lexical relation J
6.3 Two appr
Different cl
emantic relations in th
whether the theory viev
like a thesaurus. Dictio
componentially represei
introduced in chapters 3
Generative Lexicon the
represented in the lexic
an tonym of hoť) becau
ponential semantic repn
hold that semantic relatí
senses) are linked to one
antonym , and hyponyn
tradition of Structurali I

d/descend)
ely that we
erced to the
ln thi case,
) recognize
l word go
refer to the
;0 lexicaJly
ion. Since
po ite pair
)r in tance,
Vangry but
dl include:
ca es exist,
uid/gas and
elation hip
ry/sad, and
happy, you
ead happy.
ot lexically
ad (Becker
oppo ition
1994). This
'Py/sad in a
ve to figure
t important
c emantic
)fi between
'e), organiair),
or the
neow.pigibulary
and
bem -onym
.a' relation,
le say that
,Alo like
'ds - it is a
Lexical and semantie relations 123
direct reflex of the meaning of the words. Different ubtype of meronymy can
be identified, such as whole> segment (month> day), whole> functional
component (car> engine), collection> member (pride> lions, and whole>
substance (pipe> copper). Meronymy is generally not thought to be a central to
lexicall emantic organization as the other -onym relations. The relation between
meronyms and holonym i not a nece ary a the relation between hyponyms
and their hyperonym . Many part are optional (a wingle bird is till a bird) and
the ame part-name of ten apply to many different whole - for instance, handle
is alway a part-name, but i not the part of any one particular kind of thing ince
door ,jug, uitca es, and hammer all have (very different) handles. Thu ,while
meronym relation can be helpful in defining word, they are not a widespread
or as con istent a the other 'onym relations.
6.2.6 Summary
Table 6.2 summarize the facts about the relation discussed above.
Table 6.2 Properties of paradigmatic relations
Synonym Hyponym Antonym Co-hyponym Meronym
semantic relation imilarity inclu ion oppo ition contra t part/whole
binary X X J X X
syrnmetrical J X J v' X
transitive v' v' (taxonyrn) not applicable v' sometimes
lexical relation v' X often sometime X
6.3 Two approaches to relations and the lexicon
Different chool of thought exi t regarding the role of paradigmatic
semantic relations in the mental lexicon. These can be classified according to
whether the theory views the men tal lexicon a more like a dictionary or more
like a the auru . Dictionary approaches hold that the meaning of words are
componentially repre ented in the mind - o the e include the main approache
introduced in chapter 3 and 4 (e.g. Katz and Fodor 1963, Conceptual Semantic ,
Generative Lexicon theory). In thi ca e, emantic relation do not need to be
repre ented in the lexicon (i.e. nothing in the lexicon need to ay cold i the
antonym of hot') becau e tho e relation are derivable from the word ' componential
emantic repre entations. Thesaurus approaches, on the other hand,
hold that semantic relations are represented in the lexicon, In thi ca e word (or
enses) are linked to one another in order to indicate which words are synonyms,
antonyms, and hyponyms of which other word . These mostly derive from the
tradition of Structurali m that begin with Ferdinand de Sau ure (1959/1915).

124 LEXICAL MEANI G
The mo t extreme the aurus- tyle approache hold that there are no definition for
word in the lexicon, as the mean ing of a word can be derived from "the company
it keeps" - i.e. which other word it is linked to. The following sub ection exemplify
the e po itions in turn, though it mu t be pointed out that many lexicologi t
take a middle ground, expecting that both definitional (dictionary) and relational
(the aurus) information i needed for a complete emantic repre entation of a
lexeme.
6.3.1 Lexicon as dictionary
In chapter3 we di eu ed the lexicon-a -dictionary metaphor in term
of how componential theories, like dictionarie , break down the meaning of
lexeme into smaller parts. This mean that the information needed for predicting
synonym, hyponym or antonym relation among word i available in their
componential definition . Thus mo t componential theo ri ts take the view that
the relations themselves do not need to be explicitly mentioned in the words'
lexical entrie. In other word, these theori t think that arepre entation like that
in (20) is ufficient, so that the additional information in (21) is not nece sary.
(20) man [HUMA, ADULT, MALE]
(21) man HUMA, ADULT, MALE
SEX ANTONYM = woman
AGE A TONYM = boy
E R- Y O YM = gentleman, guy, chap.fetlow
HYPONYM (MARITAL TAT ) = bachelor
HYI'ONYM (JOB) = {fireman, postman, handyman ... }
Approache that do not include the relational information in (21) require
another mean of determining which lexeme are ynonyms, antonyms, and
hyponyms. This i done by pecifying rule that determine these relation, uch
a tho e in (22):
(22) a.
b.
c.
X and Y are synonym iff [i.e, 'ir and only ir'] thcy hare all the same components
X and Y are an tonym (and/or co-hyponyms) iff only one of their componenl di ffer ,
X is the hyponym of Y iff it has all of the arne component a Y plu at lea t one more.
Using the e rules, we can tell that lady (in the sense that denotes 'female
adult ' generally) is a ynonym of woman, man i the oppo ite of woman, and
fireman i a hyponym of man, a hown in figure 6.4.
Evaluating the dictionary approach
The dictionary approach i attractive becau e it explain why particular word
are related to one another - that is, because they have semantic components
in common. On the other hand, becau e it only concem the word' emantic
components, non-denotational propertie (like the ound of a word or it ocial
register) cannot contribute to these relation . So, according to the information
[
sa me componen
-7
flwo
synonyms
lady
HUMA HU
ADULT AD
FEMALE FE
Figure 6.4 eOin
in figure 6.4, lady i a go
classical componential th
and not for relation that
There have been attem
non-denotational issue I
matic approach, in whicl
hare all relevant propen
except one. In the case c
i their form, and in the (
meaning. The tricky part
relevant or not within a CI
6.3.2 Lexicon as ti
Theorie that
entries (or en se ubentr
to other lexical entrie (Ol
high would somehow rep
roots in Structurali t view
thinkers - e.g. Co eriu a
orne computationally in
1996' WordNet, Fellbaun:
forgo componential sen e
words that constrain one I
are in the antonym relatir
end of whatever cale tt
different, but related, to te
Diagrammim
One the aurus-type appn
enne Lehrer (1974). Thi:
in figure 6.5, in which ear
fieJd - an organization of

definitions for
"the company
ections exernr
lexicologiats
and relational
sentation of a
rphor in terms
meanings of
for predicting
lable in their
the view that
in the words'
ation like that
t necessary.
(21) require
itonyms, and
elation , such
the ame component
their components differs
I Y, plus at least one more.
iote 'female
: woman, and
ticular words
, components
rd ' semantic
1 Oľ it social
~ information
same components
-7 synonyms
ladyftwoman
HUMAN HUMA
[
ADULT ADULT
FEMALE FEMALE
one different
-7 antonyrn
man
HUMAN
ADULT
MALE
Figure 6.4 Componential approach /0 seman/ic relations
Lexical and semantie relations 125
extra component
-7 hyponym
Ht:~~lQnJ
ADULT
MALE
FIGHTS FJRES
in figure 6.4, lady is as good an antonym for man as woman i . In other words, the
cias ical cornponential theory only accounts for denotational semantie relations,
and not for relations that involve other properties that words may have.
There have been attempts to extend the dictionary approach o that it considers
non-denotational issues as well. For exarnple, Murphy (2003) presents a pragmatic
approach in wh.ich words are related if, within a particular context, they
share all relevant propertie - including relevant non-denotational properties -
except one. In the case of ynonymy, the property that the word s do not share
is their form, and in the case of antonymy, it is typically an a pect of denotative
meaning. The tricky part about this approach is deciding which properties are
relevant or not within a context. (See Murphy 2003 for discus ion.)
6.3.2 Lexicon as thesaurus ...------------------,
Theories that treat the lexicon as thesaurus-Iike hold that lexical
entries (or sense subentries within them) are cross-referenced with or linked
to other lexical entries (or sense subentrie ). Such models hold that the entry for
high would somehow represent that it antonym is low. The e model have their
roots in Structurali t view of emantic (following on from Saus ure and irnilar
thinker - e.g. Co eriu and Geckeler 1981), but have also been carried on in
some computationally in pired approache (e.g. Meaning-Text Theory, Mel'čuk
1996; WordNet, Fellbaum 1998b). Some proponents ofthesauru models aim to
forgo componential sense representations altogether in favor of large network of
words that constrain one another's senses. In such models, because high and low
are in the antonym relation, they are constrained to always denote the opposite
ends of whatever scale they describe, and high will always refer to something
different, but related, to tall, long, and big, since they are all types OfSIZE.
Diagramming relations among words
One thesaurus-type approach is Semantie Field Theory, promoted by Adrienne
Lehrer (1974). This approach can be illustrated by a box diagram, as
in figure 6.5 in which each word cuts out some "sernantic space" in a semantic
field - an organization of related concepts.

126
l'
z
o
vi
:J
..J
U
~ 11
(l)
X -c
..J
<
u
~
'">
~ HORIZONTAL AXIS = CONTRAST
move
crawl walk run jurnp,
211:::~
...c:
~ 0..0 •.... u
::l •....•...• 0::1
~ t; t: r-
(l) 0::1 ~
jog print hop jump-
Figure 6.5 Field representation for selected verbs of human locomotion
Figure 6.5 give a partial emantic field for English verb ofhuman locomotion,
which hows all the relation that a box diagram can illustrate:
The ub cripted numeral on the two in tances of jump indicate its
poly emy. Jumpi has a general sen e 'to pri ng off the ground,' while
jump-. indicates springing and landing with both feel, a opposed to
one foot, as for hop. Jump, then, i an autohyponym, a poly emou
word that ha a en e that i a hyponym of another of it en e .
The horizontal axis represents contra t: if two lexeme on the ame
level are separated by a line, th y contra t with each other and, transitively,
with other word on that line. So, crawl, lValk,jog, run, and
jump i contrast as co-hyponym of rnove. Note that orne lexemes'
boxes inhabit more than one leve!. Jog, for in tance, i a member
of two co-hyponym et: crawl/walk/jog/run/jumpi and jog/sprint
(hyponym of run). In other word ,jog i ometime u ed to contrast
with run (a in 1don 'l run, l just jog) and ometime u ed to de cribe a
kind of running (a in When running ; alternate jogging and sprinting
for an extra good workouťy.
Viewing the vertical axi from the bottorn up, we ee hyponymy
relations. So, march is a hyponym of walk and (transitively) of /110\ e.
We've already seen thatjog i a pecial ca e, a it ha either run or
move as its immediate hyperonym.
Synonymy is indicated by two lexemes in the same box - as for stroll
and amble. (They may not be perfect synonyrn , but we can pretend
that they are for the sake of thi illu tration.)
Overlap in meaning (partia1 ynonymy) is indicated where there i
shading in tead of a horizontal line between two terrn , like saunter
and stroll-amble. Thi mean that orne activitie are definitely saunle
ring or ambling (but not both), but other (in the haded area) could
be called by either name.
Blank areas in box diagram how lexical gaps, which are potential
sense that have not been lexicalized in a language. In thi diagram,
there i a bOJ
We can de (
bipedal locc
cover thi r
Box diagram are al o l
order to ee difference
Devi e a box diagram fc
red, purple yellow, blue
To le t your diagram:
From the br
of' the abox
A k your e'
contra l wil
Alternative to box
network of en e witl
1998b) or by creating
entries have feature li!
for tho e features are o
Mel 'čuk 1996). These!
"tran lated" into a netv
of the "sernantic pace'
u e of walk (e.g, to del
other u e .
Evaluating
Proponents of the aun
word ha to other at II
no rneaning exist in a
Box diagram how the
out" for individual lexe
at their boundarie .
One type of eviden
i the phenomenon kno
meanings of finger and
can mean 'any digit of
meaning i blocked whe
my finger, you would r

union
locomotion,
, indicate it
ound,' while
; oppo ed to
polyemou
en e .
on the ame
er and, tranog,
run, and
ne lexernes'
a member
d jog/sprint
d to contra t
to de cribe a
IId sprint ing
: hyponymy
:ly)of move.
.ither run or
. a for stroll
can pretend
iere there i
like aunter
initely saun-
Iarea) could
are potential
hi diagram,
lexical and semantie relations 127
there i a box above walk.jog, run, andjump that contra t with crawl.
We can describe what that box symbolizes - the category of upright,
bipedal locomotion - but there is no particular English lexeme that
cover this meaning.
Box diagram are also useful for comparing emantic field across language, in
order to ee differences in lexicalization.
Puzzle 6-4
Devise a box diagram for the f llowing COLOR words.
red,purple, yellow. bille, cala 1', lavender, scarlet, orange, green, crimson
To test your diagram:
From the bottom up, ask your elf whether the lower word name 'type
of the above color .
A k your elf whether all the color name on any horizontal level
contrast with one another.
Alternatives to box diagram include repre enting emantic relations a a
network of sense with line linking related enses (e.g. Word et, Fellbaum
1998b) or by creating attribute-value rnatrice (see §3.2.l) in which lexical
entries have features like A TO N Y M and s Y o Y M and the values speci fied
for those features are other lexeme - a done in Meaning- Text Theory (MIT;
Meľčuk 1996). The e are all fairly equivalent (e.g. the AVM in MIT could be
"translated" into a network diagram). The box diagram, however, give a sen e
of the" emantic pace' that the lexemes take up, o one can visualize certain
use of walk (e.g. to denote 'power walking') a clo er to thejog boundary than
other u e .
Evaluating the thesaurus approach
Proponents of thesauru -style lexical model hold that the relation that one
word ha to others at lea t part1y determine the word' sen e. In other word,
no meaning exi t in a vacuum, in tead meanings are created through relation .
Box diagram how the e relation within the ernantic field. Space are "car ved
out" for individual lexeme , and one can imagine the "ten ion" between word
at their boundarie .
One type of evidence for the idea that lexical meaning depend on relation
i the phenomenon known as lexical blocking, which can be exemplified by the
meaning of finger and thumb, hown in the box diagram in figure 6.6. Finger
can mean 'any digit of the hand', as in 1 have ten fingers. But that 'any digiť
meaning i blocked when one refer to a ingle finger. So, if omeone say 1 broke
my finger, you would probably feel mi led if you di covered that the finger in

128 LEXICAL MEANING
finger
(l have ten fingers)
(l broke my finger) thumb
BLO( KING
index pinkie
finger middle ril/g
finger finger little
pointer finger
Figure 6.6 Lexical blocking
question was their thumb. In that case, you would have expected them to say l
broke my thumb. The existence of thumb as a po sible contrast term for finger
bJocks the interpretation of ingular finger to mean 'thurnb. ' Thus we see that the
semantic relation betweenfinger and thumb affect how they are interpreted.
While the aurus-style theorists have sometimes taken the extreme view that
relations are all that is needed for meaning, no successful model of the lexicon
has been able to rely on relations only and completely leave out any kind of
definitional rneaning. Figure 6.5 show us that walking and jumping are different
kinds of moving, but it does not tell us how to walk or jurnp. Similarly, if you
looked at a box diagram of locomotion term for a language you didn't know, you
would be able to say which were the more general terms and which contrasted,
but you probably would not be able to translate the word into Engli h. Thi lack
of definitional information has pro ved difficult for computational model that use
thesaurus-style links among words, Mo t uch model (like WordNet and MIT)
include definitions as well as relations.
This raise the question: if thesaurus-style models require definition , then
why not just use a dictionary-style model, which has to have definitions anyway
and which can account for relations by rules (as in (22) above)? The main
argument in fa vor of representing semantie relations in the lexicon is that not all
sense relations are completely predictable. In this ca e, their relation need to be
represented in the lexicon, the repository of unpredictable facts about language.
Recall the examples of antonymy like big/little, but not large/little, which seem
to indicate that words' seman tie properties are not sufficient to determine which
of a range of synonyms will be its antonym. If thaťs the case and the e relations
are arbitrary to orne degree, then they are fact about the language that must
be recorded in the lexicon. Proponents of the dictionary view argue in reply that
if you look closely enough, you can find emantic difference that predict, for
example, that big/little is a better antonym pair than large/little because big and
little are more minimally different than large and little (recalI your solution to
Puzzle 6-1c). But theori t with a thesaurus point-of-view might till reply that
these words have developed slight variations in meaning preci ely in order to
accommodate the existing relations between them within their semantie field.
6.4 Summary
This chapter i
hyponymy, and antonym:
defining them. It al o de:
of tbe e relations. Dictior
of the words' emantic (
accounts explicitly repres
of how the men tal lexicc
these extremes, since die
non-seman tie propertie
relation , and thesaurus-:
on their own. We return t
relations for verbs in § 1O
antonym type in chapter '
6.5 Further rea
Textbooks by
lots of detail and example
des discussion of the type
sion of history of the dicti:
tie relations in other tiel
cien ce. Sources on the d:
but you can also explore 1
6.6 Answers to
6-1
Your answers will vary,bu
a. The car is safe~
tend to mean di
will come to yo
won't be hit by
locked up and p
b. She was wearin
substitute for re
costurne.
c. A big/large 1110\
large star i phy

;)
lem to say l
m for finger
: ee that the
erpreted.
le view that
: the lexicon
any kind of
are different
Jarly, if you
't know, you
l contra ted,
ih. This lack
dels that use
.t and MIT)
iitions, then
ion anyway
? The main
that not all
IS need to be
ut language.
which seem
nnine which
ese relations
ge that must
in reply that
. predict, for
ause big and
r solution to
.iIIreply that
I in order to
mtic field.
6.4 Summary and conclusion
Lexical and semantie relations 129
This chapter introduced three major semantie relations - synonymy,
hyponymy, and antonymy/contra t - their propertie , ubtypes, and problem in
defining them. It al o described and evaluated two approache to the treatment
of these relations. Dictionary-sty le accounts airn to predict relations on the basis
of the words' semantic propertie in the mental lexicon, while the aurus-style
accounts expJicitly represent those relations in the lexicon. An accurate account
of how the menta! lexicon works is probably to be found somewhere between
these extremes, since dictionary-style approaches cannot explain the effects of
non-sernantic properties (e.g. morphological form) on the "goodness" of these
relations, and the auru s-style approache are not sufficient models of meaning
on their own. We return to semantie relation issues when we examine inclusion
relations for verbs in § 10.6 and the connections between adjective meaning and
antonym type in ehapter J l.
6.5 Further reading
Textbook by Lyon (1977, 1995) and Cruse (1986, 2000b) provide
lots of detail and examples of the relation discussed here. Murphy (2003) includes
discussion of the type and ubtype of relations, as well as extended di cussion
of history of the dictionary/thesaurus debate and of the importance of semantie
relations in other fields, including philo ophy, psychology, and computer
science. Sources on the different the auru -style approaches are given in §6.3.2,
but you can a! o explore WordNet on the web at http://wordnet.princeton.edu/.
6.6 Answers to puzzles
6-1
Your answers will vary, but here are some examples:
a. The car is safe/secure. While both of these are acceptable sentences, they
tend to mean different things. The car is saje mean either that no harm
will come to you if you drive it - or that the car i in a afe place (e.g. it
won't be hit by a falling tree). The car is secure probably mean that it is
locked up and protected against thieves.
b. She was wearing fake/false teeth. False teeth are dentures, used as a
ubstitute for real teeth, but you might u e fake teeth for a vampire
costume.
c. A big/large movie star signed autographs, A big tar is very famou, but a
large tar is phy ically impo ing.

130
d. I want lo be a somebody/(#someone). While the e two word are u ually
6-2
substitutable, somebody (and not someone) ha a noun u 'e that means
'a ignificant per on.'
Perfect sen e synonymy is ymmetricaJ and transitive. It is ymmetrical since if
sofa = couch, then coucli = sofa, and tran itive in that if s/eep = slumber and
slumber = snooze, then it hould follow that sleep = snooze. However, becau e there
are few perfect synonym , there i of ten ome" emantic lippage" among the
near- ynonym that one can find in a thesauru , which re ult in non-tran itivityand
sometime even non- ymmetry. It ha been hown that if you look up a word in a
thesauru , then look up the ynonyms of its ynonym (and so on), the ynonym
path between any word and its antonym i typically ix teps or fewer, a is the ca e
for authentic/unauthentic in The New Collins Thesaurus (reported in Church et al.
1994):
authentic -7 believable -7 probable -7 ostensible -7 pretended -7 spurious=«
unauthentic
6-3
a. contrary:
slow.
thing can have a 'medium' peed a well as beingfasl or
b. converse: if Max i Alice's teacher, then Alice i Max's student.
c.
d.
e.
6-4
contrary or complementary: ometimes dean is used to mean ab olutely
clean - in which ca e one peck of dirt render omething dirty, But when
we are talking about, ay, the rate of omeone hou ekeeping, we often
treat it as contrary - o that one could ay that orneone' hou e i neither
clean nor dirty - it could be better or worse.
usually a sumed to be c mplementary: if omeone told you that the
univer ity has 10,000 tudent, half of them fernale, you would a urne
that the other half i male. While tate between malene and femalene s
are medicaUy po ible, we tend lo ignore that pos ibility in our everyday
u e of the word .
contrary: while male andfemale indicate a per on' phy iological ex,
feminine and masculine de cribe behavior or appearance - and ome
people are neither feminine nor masculine in tho e re pect .
A common mi take in thi exerci c is lo put orange under or above red and yellow,
ince orange i 'made of red and yellow. Thi 'made-of relation may be relevant to
paint that are orange and red, but not to the meaning of the word orange and red,
which are in an overlapping, but contrasting, relation. The diagram hould look
something like figu re 6.7.
All of the contra ting terms overiap, since color boundaries are very fuzzy -
something turquoi e, for in tance, could be called either green or blue. Ideally, the
diagram hould be three-dimen ionalothat purple and red join up and overlap as
well.
crimson
red
scarlet
Figure 6.7 wil
6.7 Exercises
Ad O pt-a-word
A. Explore the iss
synonyms fou
synonymous?
synonymy is d
why the thesa
B. Design a box (
of its senses)
word class (e.
within the fiek
more appropr
l. Show whethe
whether or no
a. quick, fa t
b. ick. iII
c. near, clo. e
2. Devi. e box dl
ju tify your d
Take into aee
sen e that re
a. tool, bar
b. bun, roll,
3. The followins
whether thi: .
sense of tart

ire usually
tmean
ince if
. and
cau e there
: the
iitiviry and
ord in a
ionym
i the ca e
rch et al.
purious ->
S/ or
h olutely
. But when
we often
: i neither
tthe
assume
'emalene s
everyday
:al ex
orne
lyellow,
elevant to
and red,
look
~ylily,
the
erlap a
crim on carlet
B.
Lexicaland semantie relations 131
color
red I orange yellow I green blue purple
Figure 6.7 Lexica/ fie/d for cotor terminology
6.7 Exercises
Adept-a-word
A. Explore the issue of synonymy by comparing your word to up to three
synonyms found in a thesaurus. To what extent are the words really
synonymous? Demonstrate using the substitutability test. Consider how
synonymy is defined and, if the words are not absolutely synonymous,
why the thesaurus proposed them as synonyms.
I
lavender
Design a box diagram for a semantic field that your word (i.e. at least one
of its senses) is in. Limit yourself to about twelve words, all in the same
word class (e.g. noun, verb), if possible. Discuss the semantic relations
within the field and critique the box diagram method. Are such diagrams
more appropriate to some kinds of words than others?
l. Show whether the following word are sense ynonym by dem on trati ng
whether or not they are ub titutable in all context.
a. quick. fa t
b. ick, iII
c. near clo e
2. Devise box diagram that contain the word. in the following lists. Briefly
ju tify your deci ion ba ed on your own under tandings of these words.
Take int a count whether any word are polysemous. but only con ider
enses that relate to the r u a ITUREorBAKED oo o n fields.
a. tool, bar tool, chair, armchair, chaise longue, seat, ofa
b. bun. roll. bread. cake, muffin, . cone. bage1
3. The following words have more than one antonyrn. FOľ each, di cus
whether thi is becau e the word i polysernou . For in tance, doe the
en e of s/ar/ that is an antonym to finish differ from the ense of starr that