Ravin, Yael, and Claudia Leacock (eds.) (2000) Polysemy: Theoretical and
Computational Approaches. Oxford University Press. Hardbound, ISBN:
0-19-823842-8. Pages: 240. Price - $74.00
Niladri Sekhar Dash, Indian Statistical Institute, Kolkata, India
Synopsis
The term 'polysemy' has a long history in linguistics,
stylistics, psychology and literature. The feature was first
noted by the Stoics who applied some principles for
identification of polysemy. Presently, the term is used both
in semantic and lexical analysis with a special connotation
where it implies a word with multiple meanings or
senses. Although such words generate little difficulties in
everyday communication among people, they do pose
impenetrable problems of understanding of actual sense for
linguists and lexicographers. The contributors in this
volume considerably ventilate into these problems for
linguistic theory and how they can be used in computational
linguistics.
In their introduction (pp.1-29), Yael Ravin and Claudia Leacock present an
overview of polysemy with reference to different models employed for
understanding and interpreting the problem at hand. After discussing
briefly what polysemy is, they present a comparative discussion between
polysemy versus homonymy and indeterminacy. In section 3, they try to
capture the relation between polysemy and the context while in section 4,
they initiate an effort for capturing the core of different theories of
meaning (the classical approach, the prototypical approach, and the
relational approach etc.). In their concluding section, they deal with
polysemy from a computational point of view to provide a good exposure to
the people interested in automatic word sense disambiguation both at
lexical and syntactic level.
In Chapter 2 (pp.30-51), D. Alan Cruse discusses the micro-structure of
the word meanings. In the introductory section, he addresses one of the
central problems of lexical semantics: the sensitivity of word meaning to
context which creates difficulties for the description of the content of
the meaning of a word. In the second section, he illustrates the symptoms
of distinctness in sense modules of words with close reference to
antagonism and discreteness. The third section is very brief with
suggestions for unity/integrity of the meaning of a word. In the fourth
section, to illustrate the difference of views of a single element by
different perceptors, he introduces the idea of discontinuity in word
meaning where he argues that there are a few factors which play important
roles in case of discontinuity of word meaning. The first source of
discontinuity is sub-sense of words, the second one is the facets, while
the third one is what he calls 'ways of seeing' (WOS). To substantiate his
first proposition (sub-sense), he furnishes some examples which support
his argument. The idea of superordinate and co-hyponyms further
strengthens his proposition. The properties of facets, illustrated here,
are based on discreteness and unity. The WOS are classified as seeing
something meronymically, taxonomically, and in terms of its interactions
with other things. In conclusion, following Lyons (1963:80) Cruse states
that there is no such thing as 'the meaning of a word' in isolation from
particular contexts: decontexualization of meaning is variable, and in
principle, always incomplete (p.51).
In Chapter 3 (pp.52-67), Christine Fellbaum discusses autotroponymy: the
semantic drift, conceptual shift and cognitive flexibility at the time of
semantic perception of a lexical item. In the introductory section,
Fellbaum provides some general principles of lexicalization and polysemy.
Here she presents some patterns for mapping distinct senses onto one word
proposed by different scholars. In section 2, she observes polysemy in
verb lexicon and proposes, with examples, that the majority of verbs (in
English verb lexicon) refer to specific manners of performing actions
denoted by other verbs (p.54). Here she provides some general principles,
under the head of 'polysemy promotion' and 'polysemy blocking', for
showing contrast between the causative-inchoative and transitive-middle
pairs on the one hand and the verbs related to troponymy ('manner'
relation) on the other. Section 3 is very brief giving only hints and
examples of some unusual cases of verb polysemy. In section 4, she
distinguishes different types of autotroponymy on the basis of syntactic
criteria. Here she shows that conflation is a common phenomenon that
yields new words and word meanings. With reference to conflation she shows
the derivation of denominal verbs, conflation of superordinate noun
arguments, conflation of measure nouns and other cases of noun conflation.
The remaining two sections are spent dealing with adjective and adverb
conflation in English, respectively. In conclusion, she registers the
relevance of autotroponymy in semantics of the conflated nouns, verbs or
adjectives.
In Chapter 4 (pp.68-90), James Pustejovsky addresses the general nature of
argumenthood and what logical distinction is possible between argument
types while examining the syntactic and semantic behaviour of some English
verbs. In introduction, he introduces the concept of lexical shadowing
which, according to him, can be defined as the relation between an
argument and the underlying semantic expression, which blocks its
syntactic projection in the syntax (p.68). After a short analysis, he
identifies three types of lexical shadowing:(i) argument shadowing, (ii)
complementary shadowing, and (iii)co-compositional shadowing. In section
2, he gives an outline how shadowing is performed by the grammar where he
recasts the framework of a generative lexicon (GL) and reviews some of the
basic assumptions of the theory to estimate how they bear on the problems
at hand. His proposed GL framework is characterised by (i) argument
structure, (ii) event structure, (iii) qualia structure, and (iv) lexical
structure - four basic levels of linguistic representation. He
distinguishes four types of argument which are used for refining
classification between argument and adjacent phrases. Moreover, he
distinguishes two types of argument closure:(a) lexical closure which can
be used for describing default arguments and shadow arguments, and (b)
functional closure wherefrom polyvalency phenomena and pragmatically
controlled deletion can be seen arising. He sums up by suggesting that the
grammatical distinctions between lexically and functionally closed
arguments are semantically derived, and accounted for by basic differences
in the type of the lexical items involved (p.77). In section 3, he returns
to the cases of complementary shadowing and coercion with reference to the
peculiar behaviour of shadowing of a particular English verb. In section
4, he discusses the nature of argument expression in a certain class of
verbs, which he refers to as 'complex relations'. Here he defines
'containment relation' which is encoded directly in the semantics of a
container-like concept as the formal qualia value. He argues that a
complex relation is one which decomposes into simpler component parts,
each of which is itself a relation. Thus, Pustejovsky characterises
different types of optionality on the expression of arguments that a
predicate may take, and the methods with which they achieve existential
closure in the semantics.
In Chapter 5 (pp. 91-110), Charles J. Fillmore and B.T.S. Atkins describe
polysemy with an example of a lexical item obtained from different
dictionaries either compiled manually or developed compiling data from
corpora. In section 1, by citing examples from American Heritage
Dictionary of the English Language (AHD, 1992) and Collins English
Dictionary (CED, 1991), they show how dictionaries can recognise multiple
senses of a single word. In section 2, they observe and analyse corpus
attestations of a word comparing with the citations obtained from four
major British learner's dictionaries: the Cambridge International
Dictionary of English (CIDE, 1995), The Collins-Cobuild Dictionary of
English (COBUILD, 1995), the Longman Dictionary of Contemporary English
(LDOCE, 1995) and the Oxford Advanced Learner's Dictionary (OALD, 1995).
Different figures and tables are furnished to support their argument that
the number of sense distinctions that show up in the corpus far exceeds
the number of distinctions that are provided in the dictionaries.
Moreover, the dictionaries fail to capture many varied metaphorical uses
of the words to be found in corpora. In section 3, they argue that lexical
semantics is in a poor position to solve the problem of polysemy because
polysemy is a prototypical concept having a few marked features which
probably can be accessed if an investigation on polysemy includes (i)
corpus-based lexicography, (ii) combinational properties of lexical items,
and (iii) the design of inference systems built on natural language texts.
In section 4, they consider if the translation-equivalents in bilingual
corpora have same kind of matching of senses. After citing a
translation-equivalent of an English word in French, they summarise that
word meaning analysis allows more detailed description of the factors
involved to facilitate a more delicate cross-linguistic matching of sense
(p.108). In appendix, they report a project (Berkeley FrameNet) designed
for devising a comprehensive and internally consistent account of word
meanings and their combinational properties.
In Chapter 6 (pp.111-128), David Dowty is concerned with the problem of
argument alternation (e.g. passivisation, raising, Tough-Movement,
'Middle' Construction, spray/load-Alternation etc.) to show that alternate
forms definitely serve to convey significantly different meanings though
it is usually assumed that the sentences with such alternance express the
same propositions and differ only in their syntactic form (p.111). In next
two sections, he discusses intransitive alternation with 'with' and how
the intransitive 'swarm'-alternation differs from the transitive
'spray-load'-alternation. In section 4, he presents four observation for
the semantic properties of the Location-Subject (LS) form. Here he locates
the semantic classes of verbs appearing in the L-subject form, and argues
that (i) the with-phrase object must be semantically 'unquantified', (ii)
the with-object can be a sound source but not an agent, and (iii) the
L-subject form is more suited to metaphor than the A-subject form. In
section 5, he notes some aspects of an analysis of the LS-form. Here he
makes ten claims about the L-subject form which are supported with
examples and discussions. In concluding section, Dowty investigates the
relation of the LS-form to other lexical semantic extension patterns
(metonymic semantic extension etc.) and considers the history of the
construction in English and related languages citing examples from
Germanic, Dutch and French.
In Chapter 7 (pp.129-151), Cliff Goddard provides outlines of 'natural
semantic metalanguage' (NSM) method of semantic analysis and shows that
this method enables the traditional 'definitional' concept of polysemy to
be applied both to individual lexical items and to lexico-grammatical
constructions (p.129). In section 1, he presents the most basic premises
of the NSM approach and reports about the identification of sixty semantic
primes 'which are impervious to (non-circular) definition and universal in
the sense of having equivalents in all languages' (p.130). Here he argues
that the NSM approach aims to avoid the obscurity, circularity and
indeterminacy which mar most lexicographic work, and to maximize the
explicitness, clarity, and translatability of its explications. In section
2, he deals with lexical polysemy where he argues that multiple meanings
of a lexical item is statable by reductive periphrases terms, and their
validity is testable by substitution. To substantiate his argument he
furnishes examples from Sahaptin, French and English. In section 3,
emphasis is given on the polysemy of grammatical constructions where
Goddard concentrates on a particular form (NP have + AUX a VP-INF) of the
English construction. In section 4, Goddard takes figurative use of
language under his consideration and tries to distinguish between a word's
literal and figurative meaning by analysing some examples compiled from
English. After analysing Michael Yell's (1996) study of metaphorical
language about music he sums up that metaphors from various different
groupings can 'map onto' the same component of literal meaning; and,
conversely, metaphors of a single type can be used to elaborate different
components of literal meaning (p.148). In conclusion, he notifies some
general problems of polysemy and the advantages of the NSM approach that
can efficiently deal with lexical polysemy as well as with polysemy
manifested in certain grammatical constructions.
In Chapter 8 (pp.152-160), George A. Miller and Claudia Leacock consider
what kind of lexical representations are required for sentence processing.
In introducton, they specify what information one can gather while
learning words, to what extant a dictionary can provide information about
words, and why a dictionary cannot be a good theory of the lexical
component of language. In section 2, they deal with contextual
representations of the words with reference to the method of Walter
Charles (1988) to sum up that 'people learn contextual representations (of
words) by observing how word forms are used to express different word
meanings, and that word meanings with similar contextual representations
are judged to be similar' (p.154). In section 3, they emphasise on the
importance of the syntactic category of the words for identifying the
local contexts. With reference to different works on contextual
representations, they argue that it is still dubious how to extract
adequate contextual representation from the local contexts. In section 4,
they consider if topical contexts can provide some more information for
capturing contextual meaning of words. Here they briefly report on
different statistical approaches to the problem of topical context and
refer to their experiment undertaken to capture contextual meaning of
words. In conclusion, they assume that probably some more information
besides contextual information, local context information, information of
part-of-speech, topical information, selectional restrictions, method of
co-reference etc. are required for sense identification of polysemous
words.
In Chapter 9 (pp.160-177), Mark Stevenson and Yorik Wilks deal with large
vocabulary word sense disambiguation (WSD). In introduction, they sum up
three aspects of WSD and report how different NLP modules (part of speech,
semantic preferences, collocating items or classes, thesaural or subject
areas, dictionary definitions, synonym lists, translation equivalents
etc.) are employed for the problem at hand. In section 2, they consider
Machine Readable Dictionaries (MRDs) as reliable sources for WSD as they
provide wide coverage lexicons using a well-understood model of lexical
semantics and also contain large amounts of information useful for WSD.
They use the machine readable version of the Longman Dictionary of
Contemporary English (LDOCE, 1978) to understand homographs and senses,
and to obtain syntactic, semantic and pragmatic information. In section 3,
they study the part-of-speech codes, used in LDOCE, to verify how far
accurate part-of-speech tagging can discriminate senses of words without
any further processing. In next two sections, they report their
implementation of a methodology (designed by using combinational knowledge
resources like MRDs, filters, partial taggers etc.), and describe and
evaluate a sense tagger implemented within the methodology. Their sense
tagger requires pre-processed (tokenized, lemmatised, split into
sentences, grammatically annotated) texts, accesses dictionarial
definitions of words, utilises pragmatic codes, employs information of
selectional restriction, uses broad context, and combines knowledge
resources for sense disambiguation.
In Chapter 10 (pp.178-204), William Dolan, Lucy Vanderwende and Stephen
Richardson deal with polysemy in a Broad-Coverage Natural Language
Processing system, the goal of which is to produce a useful linguistic
analysis of any piece of text passed to it. In introduction, they refer to
a system (MindNet) they use for WSD and postulate if word sense
disambiguation is at all feasible. In section 2, they describe their
system which 'encompasses a set of methodologies for sorting, weighing,
and navigating through linguistic representations produced during the
analysis of a corpus' (p.181). Their system combines in a natural way
paradigmatic, syntagmatic, and statistical information, encoding a
sophisticated analysis of the linguistic context in which each corpus
token appears. In section 3, with illustrations, they show how their
system can efficiently deal with polysemy and WSD. The next two sections
are mainly concerned with different distribution of words in topical
context, and sense discrimination and labelling. In conclusion, they argue
that their system provides the representational capabilities needed to
capture sense modulation to allow the free acquisition of new words, new
meanings, and information about how words are actually used by speakers'
(p.201).
In the concluding chapter (pp.205-219) Hinrich Schuetze is concerned with
disambiguation and connectionism while dealing with a huge corpus for the
purpose of information retrieval and machine translation. In section 1, he
provides a brief survey of some of the connectionist literature relevant
to disambiguation, motivates of vector spaces for representing words,
contexts and senses, and shows how the acquisition of senses can be
modelled as clustering. In section 2, he introduces the context-group
discrimination, a disambiguation algorithm that satisfies the desiderata
of representing activation levels, offering a model of acquisition and
large-scale applicability. In section 3, he describes an application to
information retrieval which demonstrates that the algorithm can be used
for problems of a large size. In section 4, he discusses context-group
discrimination and connectionism in the context of polysemy and word sense
disambiguation.
Critical evaluation
In general the volume provides the readers a broad overview with a
historical and issue-based background for the treatment of polysemy,
examines and contrasts a range of current approaches, and highlights many
unresolved problems in the theoretical understanding of polysemy and the
present computational challenges. In essence, the volume will be highly
useful for those who are working in the area of polysemy and word sense
disambiguation.
The majority of the contributors have rightly appreciated the
importance of a huge database in the form of corpora in the whole process
of sense disambiguation of polysemous words. Let us hope that such
recognition of corpora will enable researchers to be more empirical in
approach towards linguistic studies both in theoretical and applied
domains.
In the context of corpus-based computational analysis of meanings of
words, the last principle of the classical approach (p.7) is probably
defunctional because corpus-based approach is entirely empirical where
multiple finer shades of sense of particular words can be retrieved if its
different contextual uses are appropriately referred.
It is now almost certain that the context of words can provide so much of
information which might not be available from the words if isolated form
their contexts of occurrence. Moreover, contextual information performs an
important role in word sense disambiguation as well as in actual sense
retieval. This notion might have inspired J.R. Firth (1957) to make such
proverbial comments like 'meaning is not a hidden mental process. It is
simply a complex of contextual relations' (p. 19), 'the meaning of a word
can be known by the company it keeps' (p. 21), 'the main concern of
descriptive linguistics is to make statements of meaning' (p. 191), 'there
is no meanings of words, apart from human participants in a social
context' (p. 225) etc. In fact, the same kind of observation was made by
Bhatrihari (AD 450), an ancient Indian grammarian, nearly 1500 hundred
years ago. In his 'Vaakyapadiya' (dealing with the philosophy of grammar)
he commented that the (literal) meaning of a word could be shifted or
extended or changed according to various contexts and that the meaning of
a word is derived from its worldly usage. The divisions into words and
word-meanings are merely useful means in the study of language and have no
reality in themselves. The sentence is the fundamental linguistic fact and
that words are unreal abstractions from the senetence (Verma and
Krishnaswamy 1989: 330).
It is understood that the approaches proposed by Dolan et al. and Schuetze
can be relevant and useful for information retrieval where the task is to
match the query context similar to contexts in the database of documents
(like World Wide Web), but is not clear to us how the application of such
models would be fruitful in machine translation without extensive manual
intervention in the form of encoding (at word, phrase or sentence level).
In the third chapter both the terms 'autotroponymy' and 'conflation'
probably required little more elaboration for establishing arguments of
the author. The description of the factors involved (p.107) is really an
essential aspect for matching of senses of the translation-equivalents in
bilingual corpora. It is equally important, as Biber, Conrad and Reppen
(1998) show, for capturing multiple semantic senses of a single word even
in a monolingual corpus.
A glossary of many new and less known technical and linguistic terms
(MindNet, SemanticNet, WordNet, FrameNet, qualia, coercion etc.) is
required for better understanding.
Finally, some typographical errors can probably be mentioned here in the
hope that these would be corrected in the next edition:
(i) Page 55, line 27-28: the preposition 'in' is
repeated. It should be deleted.
(ii) Page 58, line 8: the article 'an' should be 'a'.
(iii) Page 61, line 2: the citation no (8) does not contain
the word 'smell'.
(iv) Page 70, line 3: the capital form of 'expression'
should be in regular form because the capital form does not
carry any extra value in the context.
(v) Page 79, line 16: 'In order better to understand ...'
should better be 'In order to understand better...'. This
seems more regular and congenial.
(vi) Page 80, line 21: the spelling of 'modeled' is changed
into 'modelled' in page 86 (line 13). Though both the forms
are right, a level of consistency in spelling is expected.
(vii) Page 83, line 1-2: the form 'as well was' should
probably be 'as well as'. In the same page the form 'quale'
should possibly be 'qualia'.
(viii) Page 82 an equation [Type (newspaper) = content ?] of
the citation (51b) is probably missing.
(ix) Page 186, line 20: the name 'Hass' should be 'Haak' as
mentioned in the reference (p. 203).
Bibliography
Biber, D., Conrad, S., and Reppen, R. (1998) Corpus
Linguistics: Investigating Language Structure and
Use. Cambridge: Cambridge University Press.
Firth, J.R. (1957) "Modes of Meaning",in Papers in
Linguistics 1934-1951. London: Oxford University Press.
Leech, G. (1974) Semantics. Middlesex, England: Penguin
Books Ltd.
Lyons, J. (1963) Structural Semantics. Cambridge: Cambridge
University Press.
Verma S.K. and N. Krishnaswamy (1989) Modern Linguistics: An
Introduction. Delhi: Oxford University Press.
A short biography of the reviewer
Niladri Sekhar Dash has passed MA in Linguistics from
Calcutta University in 1991 with First Class. In 1994 he has
completed ANLP (Advanced Natural Language Processing) course
from the Indian Institute of Technology, Kanpur. From 1992
to 1995 he has worked as a Language Analyst in the TDIL
(Technology Development in Indian Languages) project of the
Ministry of Information and Technology, Govt. of India. From
1995 to 1997 he has worked as a Technical Assistant in
Computational Linguistics and NLP (Natural Language
Processing) at Computer Vision and Pattern Recognition Unit
of Indian Statistical Institute, Kolkata. From 1997 he is
working as a Scientific Assistant in the same institute. He
has submitted his Ph.D. thesis on Corpora Design and
Development for Natural Language Processing for doctoral
degree to the Calcutta University. His present areas of
research are: corpus design and development, Word Sense
Disambiguation, word processing, parts-of-speech tagging,
morphological processing, computational and lexical
semantics etc.
|
