Using Explanatory and Combinatorial Lexicologyto Describe Terms

Marie-Claude L’Homme

Université de Montréal

AbstractThis paper is a proposal to apply models developed within the Explanatory and Com-binatorial Lexicology (ECL) Framework to the analysis and description of specializedvocabularies. I will first review previous attempts in the area, and show that they haveled to interesting findings. I will then proceed to demonstrate, using terminologicalunits extracted from different types of specialized corpora (namely computing, retail-ing, and law), that various components of ECL can be used to identify and describespecialized senses. First, criteria help make subtle semantic discriminations betweenrelated senses. Secondly, the actant structure of predicative units can be identifiedaccurately and actant instantiated using semantic labels. Thirdly, lexical functionscapture various semantic relations (paradigmatic and syntagmatic) between terms.Finally, I will demonstrate that minor adaptations can be made to the ECL models toaccommodate special features of specialized terms (e.g. the definition of classes inthe description of actant structures; meronymic lexical functions).

Very often, no clear distinction is made between concep-tual and linguistic analyses. We chose to focus on linguis-tics (Igor Mel’cuk, 2002, in a seminar at the University ofMontreal).

�

1 Introduction

It is always a little surprising to note that there is no true dialogue between mostterminologists and linguists (or, if anything, between terminologists and lexicol-ogists or lexicographers). We can explain this partly by the fact that terminolo-gists designed a specific theoretical framework and created a new vocabulary totalk about problems related to specialized lexica. However, most of these prob-lems are well-known in linguistics and interesting findings on specialized texts

12 L’HOMME

are often made by lexicologists, computer scientists, or specialists who do notsituate their work within the field of terminology, but rather within linguistics.I decided that, if I was going to conduct research in this field, I would addressterminological issues using linguistic concepts and vocabulary.

This vocabulary and, especially, the theoretical framework it relies onwas supplied mostly by the Explanatory and Combinatorial Lexicology, ECL(Mel’cuk et al., 1995) and the application of its principles in the Explanatoryand Combinatorial Dictionary, ECD (Mel’cuk et al., 1984 1988 1992 1999).Due to the importance the lexicon and semantics are given in this framework,it has proved efficient for each aspect of terms I studied. More importantly, onseveral occasions, it made me consider terms from a wider range of viewpoints.Many distinctions made within ECL were very helpful when dealing with a givenproblem or a set of data. And above all—although this is not a reason that willreceive scientific approval—I simply enjoyed applying these principles to theanalysis of terminological data.

ECL has also been used by other researchers to describe specialized vocabu-lary. This paper will show how it can help understand the functioning of termsand their linguistic environment. Of course, readers will only be convinced ifthey admit (following Kocourek, 1991) that a terminological sense is only an-other possible sense a word can acquire and that some senses can be consideredexclusively from the point of view of a specialized domain, such as computing,medicine, or law.

�

First, Section 2 will present some of the basic concepts of terminology. Thiswill help understand how terminology was able to distance itself from lexicol-ogy and, sometimes, even from lexicography. It will also provide an explanationon why some aspects of terms have been studied thoroughly while others havebeen largely ignored. Section 3 will review work carried out by a (small) numberof researchers and lexicographers on specialized vocabulary who have relied onExplanatory and Combinatorial Lexicology or another framework or methodol-ogy compatible with ECL. A selection was made to retain the lexicographicalendeavors that are, in my opinion, closest to ECL-principles. Finally, Section 4will show how Explanatory and Combinatorial Lexicology can be used for ana-lyzing and describing terms in specialized dictionaries. The cases and examplesdiscussed are based on work I have been carrying out at the Laboratoire de lin-guistique informatique (LLI) of the University of Montreal.

�

In this paper, the term word will be used in a non-technical sense and willrefer to any unit that can be considered by lexicographers and can appear as aheadword in general or specialized dictionaries. In addition, I will use sense torefer the meaning a word can convey. Words can have multiples senses, andsome senses can be related to a field of knowledge. Term will be used to referto those senses considered by terminologists and described in specialized dictio-

ECL IN TERMINOLOGY 13

naries or term banks. This vocabulary will be necessary in Sections 2. and 3. Iwill also resort to a more technical expression borrowed from ECL, that of lexicalunit (= LU) (Fr. ‘lexie’). The lexical unit is the basic unit of the Explanatory andCombinatorial Dictionary (ECD) and corresponds to one sense of a word. Lexicalunits associated to the same signifiers with definitions that possess a significantintersection (called semantic bridge in ECL) are organized within vocables (Fr.‘vocable’). This distinction will be used in Section 4. where I will discuss theapplication of ECL-principles in the work carried out at the LLI. Other importantconcepts will be introduced in due course.

2 Basic aspects of terms and how terminology views them

Terminology is concerned with terms, that is senses considered from the pointof view of a body of knowledge (e.g. biology, law, medicine, computer sci-ence). Even though terminology is mostly a set of practices “concerned with thecollection, description, processing and presentation of terms” (Sager, 1990:2),textbooks usually refer to a “theory of terminology”, the “general theory of ter-minology”, or “schools of terminology”.

The most prominent school of terminology is the “school of Vienna”, afterprinciples attributed to Eugen W’uster. Nowadays, it has become fashionableto criticize its principles and an increasing number of scholars list a numberof arguments that aim at showing how wrong these principles were. However,no new model has yet truly replaced that of Vienna and my review of some ofits principles should help understand why specialized dictionaries—and, moreprecisely, terminological dictionaries and term banks—are compiled the waythey are.

First, conventional terminology has been mainly concerned with standardiz-ing lexical usage in special fields of knowledge. Hence, terminologists, with thehelp of specialists in the domain under examination, make decisions aimed at re-ducing polysemy and synonymy and lead to a number of linguistic “standards”.Although these principles are sensible, considering that they are designed to im-prove communication in areas in which precision is a primary objective, theyhave been unattainable in practice.

Secondly, terminology considers that terms reflect the knowledge structure(or “conceptual structure”) of a domain. Terms are seen as the linguistic real-izations of concepts and are analyzed as such. Concepts are studied via theirrelationships with or oppositions to other concepts. For example, PRINTER andPERIPHERAL or HEART and ORGAN will be analyzed in terms of a ‘specific’ �‘generic’ relationship; CARTRIDGE and PRINTER or CLAVICLE and SKELETON

will be described via a ‘part’ � ‘whole’ relationship, and so on and so forth.

14 L’HOMME

Concepts are also grouped into larger classes and sometimes organizedwithin hierarchies or other types of knowledge representations reflecting a tra-dition that prevails in certain disciplines (zoology, for instance, classifies livingorganisms according to the way they reproduce; chemistry refers to a periodictable as a means to organize the elements).

Normally, specialized dictionaries compiled according to conventional ter-minology principles should reflect somehow the conceptual structure of a do-main. This standpoint has a number of consequences on the way terms are listedand described in dictionaries:

� terms are analyzed according to the concept they represent and not accord-ing to the linguistic contexts in which they appear; it has even been saidthat terms are “context-independent”;

� meanings are captured by means of analytical definitions, i.e., Aristoteliandefinitions listing a genus and differentiae;

� terms, or more precisely the concepts they represent, are linked or op-posed to each other according to certain characteristics (often referred toas “intrinsic characteristics”);

� all the lexical units referring to a given concept will appear in the sameentry and considered as true synonyms (e.g., EMAIL, ELECTRONIC MAIL;ELECTRONIC MESSAGING will be listed as variant headwords or at leastplaced in the same article).

Considering what has just been said, it seems natural that terminological dic-tionaries have given prominence to terms referring to entity concepts, i.e., ob-jects, instruments, people, places, etc. Terminology provides an elaborate appa-ratus to represent these conceptual categories. However, it is much less helpful toaccount for terms referring to activities, properties and relations. Consequently,most entries in these dictionaries are terms in noun form.

Another striking characteristic displayed by current specialized dictionariesis the importance they give to complex terms. Most headwords are complexnouns: e.g., host name, home page (Internet), laser printer, analog data (com-puting); bearer cheque, external audit (accounting). General dictionaries onlylist a limited number of multiple word expressions and these are included if theyare complete phrasemes, i.e., multiple word expressions that convey a mean-ing different from those of their respective parts. In specialized dictionaries, avery large portion of complex nouns has a compositional meaning (e.g., a laserprinter is a “printer” that functions with “laser”; an external audit is an “audit”conducted by someone that does not work for the firm concerned).

ECL IN TERMINOLOGY 15

From what has been said up to now, it would appear that terminology—atleast the kind of terminological work that adheres to the Vienna principles—andlexicography do not view words from the same angle. Terminology does notnormally consider units that are not nouns, even if this approach leads to incon-sistencies (e.g. PROGRAM and PROCESSING will be listed in a dictionary oncomputing, but not always [to] PROGRAM, PROGRAMMABLE or [to] PROCESS).Since priority is given to the description of concepts and their characteristics andnot to linguistic usage, predicative lexical units are not taken into account or, ifthey are, their actant structure is not represented clearly. Long lists of complexterms will appear in specialized dictionaries even though they are completelycompositional. Other differences will be discussed later on in this paper.

In the next section, I will show that interesting aspects of terms can be re-vealed if they are studied within a different framework.

3 Previous work resorting to Explanatory and Combinatorial Lexicology(ECL)

A growing number of terminologists and other researchers concerned with spe-cialized texts turn to alternative models to describe terms in dictionaries or forother types of applications, e.g. information retrieval or natural language pro-cessing. This is mainly due to the fact that most work nowadays is carried out onlarge specialized corpora with the help of computer applications, and W’usterianprinciples are not well suited to account for a variety of phenomena displayed bycorpora. However, few researchers resort directly to Explanatory and Combina-torial Lexicology. We will review some of this work in the following subsections.

It should also be pointed out that the Explanatory and Combinatorial Dictio-nary, like most lexicographical endeavors, includes descriptions of what I wouldconsider as specialized lexical units. This will be discussed in Subsection 4.1.

3.1 A concrete proposal to resort to ECL

. . . specialized lexicography is returned to the province of the lex-icographer, who, armed with the tools of a formal dictionary, canuse the specialists as specialists, i.e. informants. This facilitatesthe REAL purpose of specialized lexicography, which is the precisecharacterization of the lexicon, not terminological argument (Fraw-ley, 1988:198).

A few authors (Frawley, 1988; Gentilhomme, 1994, 1995; Marcel, 2000) re-fer to Explanatory and Combinatorial Lexicology (ECL) as a means of describing

16 L’HOMME

specialized terms in science, mathematics and biology respectively. I chose topresent the work by Frawley since it is the most eloquent proposal for using ECL

in the preparation of specialized dictionaries. In addition, it is work on Englishwhereas most work compatible with ECL and conducted on specialized vocabu-laries bears on French.

Frawley (1988) argues that the use of a formal framework such as the ECL

addresses the problem of vagueness common to most definitions found in spe-cialized dictionaries. Vagueness is defined by the author as “non-specificity ofcontextualization” (Frawley, 1988:191) and it results from the fact that dictio-naries seldom provide information on the valence of entries (and, even whenthey do, the information provided is not sufficient to give users what they needto produce the entry in a linguistic context).

Frawley claims that the use of ECL-principles in the compilation of special-ized dictionaries, solves this problem, chiefly because it defines entries “throughactual associations” (Frawley, 1988:199). In existing dictionaries, even thoughsome definitions supply partial information on these associations (e.g., theymight contain a hyperonym, or a typical verb with which a noun combines),the associations are not indicated explicitly. Moreover, many definitions containsupporting information that is not lexical, i.e., that does not concern the worddefined considered as a lexical unit and is encyclopedic or pragmatic.

According to the author, the use of lexical functions and the clear identi-fication of semantic actants should contribute to focus on lexical associations(paradigmatic and syntagmatic), eliminate vagueness and reduce the necessityfor supplying encyclopedic information.

First, I will explain the notions of “predicative lexical units”, “semantic ac-tants” and “lexical functions” and then proceed to show how Frawley appliesthem to specialized lexical units.

Explanatory and Combinatorial Lexicology distinguishes between pred-icative lexical units and non-predicative lexical units. The former refers tosemantic predicates, such as events, properties, relations, and states. Se-mantic predicates necessarily involve participants and these participantsmust appear in the definition. Verbs, adjectives, and adverbs can referto predicative lexical units, but also certain categories of nouns, such asnouns that refer to events or actions (e.g., CONFIGURATION, INSTAL-LATION), properties (e.g., COMPATIBILITY, PORTABILITY), etc. ECL-definitions clearly indicate semantic actants and are composed of two sep-arate parts. First, the defined word is inserted in a propositional form,which is an expression with variables (X, Y, Z, W). These variables re-fer to the semantic actants (which are the expressions of the participantsinvolved in the LU being described).

ECL IN TERMINOLOGY 17

A lexical function (LF) is written f( � ) = � : f represents the function, � ,the argument, and � , the value expressed by the function when appliedto a given argument. The meaning associated to an LF is abstract andgeneral and can produce a relatively high number of values. For example,Magn is a function that expresses intensification. It can be applied todifferent LUs and produce a high set of values (e.g. Magn(smoker) =heavy; Magn(bachelor) = confirmed, etc.) (Mel’cuk et al., 1995:126-127).There is a set of approximately 60 standard lexical functions divided intoparadigmatic and syntagmatic functions (Mel’cuk, 1998; Mel’cuk et al.,1995; Wanner, 1996). A list of the lexical functions mentioned in thispaper is provided in Appendix A.

In accordance with the ECL-approach, Frawley proposes to conduct inter-views with specialists when collecting information on a given term. For ex-ample, geologists could be asked the following questions on the term WINCE

(in geology, an opening or excavation): “Is there a name for a group of winces(Mult)? Is there a way of talking about the closing of a WINCE(Liqu)? Can therebe more or less of a WINCE(Minus, Plus)?”, etc. (Frawley, 1988:201).

Figure 1 is a reproduction of the entry for sc mass in physics that Frawleyuses as an illustration of the application of ECL-principles to specialized lexi-cography.

Frawley’s proposal, although it has not been implemented on a large-scalespecialized vocabulary, shows that specialized senses can be captured using anapparatus such as the propositional form (and according to the author, it shouldbe). It also points into an interesting direction, i.e., the description of varioustypes of semantic relationships between terms by means of lexical functions:paradigmatic relationships such as derivatives, synonyms, and actants; but alsosyntagmatic relationships such as cooccurrence.

3.2 Specialized dictionaries compatible with ECL

Very few specialized dictionaries resort to models that can be related to Explana-tory and Combinatorial Lexicology for reasons I tried to highlight in Section 2.I found two relevant examples for French. The first one is a set of contextualdictionaries (Subsection 3.2.1) and the second one, a dictionary for learners of aspecialized language (Subsection 3.2.2).

I will also describe some work on specialized combinatorics (Subsec-tion 3.2.3) that has attracted interest in the terminological community for thepast couple of decades. Some of this work is based on lexical functions.

18 L’HOMME

MASS: noun

X = 1 Y = 2 Z = 3N N N

A body’s, X’s, ratio of force, Y, acceleration, Z.

Lexical Functions:

Anti A � (mass) = massless Magn(mass) = densityAnti(mass) = zero mass MinusFunc � (mass) = decreaseCentr(mass) = center of mass Mult(mass) = crystalContOper � (mass) = conserve Oper � (mass) = have, be characterized,

possessCulm(mass) = point mass PlusFunc � (mass) = increaseExcess(mass) mass excess [only in sub-atomic physics]

Pred(mass) = be

Fact � S ����� (mass) = pair creation Propt(mass) = because of, out of, fromFinS � (mass) = annihilation, conversion[into energy]

S � (mass) = body, particle

Func � (mass) = circulate S � (mass) = forceFunc � (mass) = react to, determine S � (mass) = accelerationLabor � � (mass) = determine Sres(mass) = weightLabor � � (mass) = change [motion of], ac-celerate

Syn(mass) = energy

Labor � � (mass) = produce Ver(mass) = rest mass, proper mass

Figure 1: Entry MASS in Frawley (1988:203-204)

3.2.1 Contextual Dictionaries

Descamps (1973, 1976) and Descamps et al. (1992) propose a methodologybased on specialized corpora to compile what they refer to as contextual dictio-naries. In line with previous work on the lexicon designed for learners of French,a first specialized dictionary was compiled using a corpus of French texts on ge-ology. Another dictionary covers the vocabulary of politics and was compiledafter the analysis and comparison of corpora on politics.

Lexicographers extract the words to be analyzed from corpora and organize,for each of them (nouns, verbs, adjectives and adverbs), a set of concordances inorder to identify “contextual patterns” (‘schéma d’environnement’), i.e., a set ofsyntactic and dependency relationships. In this project, no clear effort is madeto distinguish terms from other units that are contained in specialized texts. Allwords found in the corpus appear to be equally interesting.

The result of this analysis takes the form of an article which is illustrated inFigure 2 (part of the entry FAVORABLE (Engl. ‘favorable’, ‘in favor’) given inDescamps et al., 1992).

ECL IN TERMINOLOGY 19

favorable, a. 4/3 ; 3/2 (cont. éclat.)

1. PRÉDICAT HUMAIN. CONSÉCUTION

*Professant, donc déclarant d’une façon positive, un certain choix politique.A(EST)«FAVORABLE À» G/M (M=TECHNIQUE DE GOUVERNEMENT)1. D’un premier sondage sur des intentions de vote, il ressort que 36 % des

personnes interrogées se déclaraient favorables à l’UNR (. . . )2. . . .

2. CONSÉCUTION

*Qui est à l’avantage de quelqu’un ou d’un groupe, ou qui fait advenir une situa-tion.

A) M (EST) «FAVORABLE À» A/G/N (N = ACTION DE B)3. Si ceux-ci [les résultats du prochain scrutin] ne lui sont pas favorables . . .4. . . .

B) «FAVORABLE» + ADV (VALEUR DE RENFORCEMENT)(4) largement favorable (5) très favorable(4) toujours largement favorable

Figure 2: Entry FAVORABLE extracted from (Descamps et al., 1992: 53)

The headword appears in bold characters followed by numbers referring toits frequency and dispersion in each corpus studied. Then, the different sensesare clearly separated and identified with numbers (1. and 2. identify the twosenses of FAVORABLE).

Each sense is accompanied by a semantic label that provides a first gen-eral identification of the conceptual category to which the lexical unit belongs.Labels are taken from a finite list the lexicographers have built for a specific vo-cabulary (in this case, the vocabulary of politics). PRÉDICAT HUMAIN is usedfor all qualities that describe a human reality. CONSÉCUTION is used to capturerelations such as cause � effect, factor � product, and so on and so forth. Otherlabels are ACTIVITÉ HUMAINE (Engl. ‘human activity’, AGENT HUMAIN (Engl.‘human agent’), LIEU (Engl. ‘location’). Each sense is then explained by meansof a definition (preceded by an asterisk in the example provided).

Then, one or several contextual patterns are provided under each definition.They are identified with capital letters in Figure 2. Patterns show syntactic anddependency structures in which the term is involved. Each pattern is exempli-fied with sample sentences taken from the corpora and used during the analysisstages.

Another important feature of the entries contained in contextual dictionariesis that cross-references are listed for morphologically-related words. In the en-try FAVORABLE, cross-references are provided for FAVEUR (Engl. ‘favor’), FA-

20 L’HOMME

VORISER (Engl. ‘[to] favor’), FAVORISÉ (Engl. ‘favored’), and DÉFAVORABLE

(Engl. ‘unfavourable’).

Although this work is not explicitly based on ECL,�

many similarities canbe observed. First of all, there is a clear distinction of senses based mainly oncontextual patterns. The description of each contextual pattern is different fromwhat can be found in Explanatory and Combinatorial Lexicology, but there isa clear effort to describe terms according to their functioning as linguistic unitsand their cooccurrence patterns.

The lexicographers also list morphologically-related terms. Although no ex-planation of the relationship of these units and the headword word is provided,an attempt is made to show that some words found in a corpus are associatedsemantically.

Finally, semantic labels, used throughout the contextual dictionaries, can befound in the definitions of more recent versions of the Explanatory and Combi-natorial Dictionary (Mel’cuk et al., 1984b)).

�

3.2.2 A Specialized Learners’ Dictionary

The Dictionnaire d’apprentissage du français des affaires (DAFA) (Binon et al.,2000) is a specialized dictionary for learners of the language of business, themicrostructure of which is in many respects reminiscent of the Explanatory andCombinatorial Dictionary. In fact, is it surprising that no mention is made ofECL in this dictionary since so many similarities can be observed.

The team of lexicographers used a large corpus of French texts on busi-ness and economics. The corpus was explored to make semantic distinctions(is the case of polysemic units), and find collocations, expressions and contextsin which the terms are used. According to the designers of the dictionary, it isintended both for decoding and encoding purposes.

In this dictionary, contrary to the contextual dictionaries described in Sub-section 3.2.1, a selection was made among the words found in the corpus inorder to retain those with a meaning that can be related to business. However,all senses encountered in the corpus are described, even those that do not appearto be associated specifically with the field of business. For example, ARGENT

(Engl. ‘money’) is defined as a sum, and as a measuring device, but also as aprecious metal (Engl. ‘silver’).

Most French terms are listed in alphabetic order with equivalents in German,English, Spanish, Italian, and Dutch. The dictionary also contains 135 thematicentries that deal with “families of terms”. According to the lexicographers, thesefamilies represent the basic vocabulary in the field of business. We will take aclose look at a typical entry, namely ÉPARGNE (Engl. 1. ‘saving’; 2. ‘savings’)

ECL IN TERMINOLOGY 21

(Binon et al., 2000:240-242) and comment on its various components.

First, users are referred to related entries (Figure 3). These cross-referencescontain all sorts of things, but usually they concern near-synonyms, hyperonyms,or hyponyms. In the ECL-framework, they would be captured with a special classof lexical functions, i.e., those designed to represent synonymic relationships(Syn � , Syn � , Syn � ).

ÉPARGNE � placement, investissementEngl. ‘placement’, ‘investment’

Figure 3: Related entries in the DAFA (Binon et al., 2000:240).

Then, a list of formally-related terms is provided (Figure 4). The first columncontains derived and complex nouns; the second column, nouns that refer topersons; the third and fourth columns, the adjectives and verbs respectively. InECL, part of these relationships would be captured using syntactic and semanticderivative lexical functions (S � , S � , V � ).

Another section lists a number of definitions corresponding to the differ-ent senses that were encountered in the corpus analyzed by the lexicographers(Figure 5). Pronunciation, grammatical information, synonyms, antonyms andexamples of usage are also provided in this section. Synonyms and antonymsare explained further in a separate section. Although the definitions may not beordered in an ECL-fashion, the senses are distinguished precisely and numbered.The numbering is used throughout the article in order to indicate the sense thatis being referred to.

Expressions in which the headword appears are also listed (Figure 6). Subse-quently, a very long section is devoted to combinatorics (Figure 7). Adjectives,nouns and verbs that combine with the noun being described are listed and clas-sified according to their meanings. Sometimes, the classification is similar to an

1 l’épargne 2 un épargnant � . . . � 3 épargner4 l’épargne-retraite une épargnante4 l’épargne-pension4 l’épargne-logement4 la désépargne

une assurance(-)épargne

Figure 4: Morphologically-related terms in the DAFA (Binon et al., 2000: 240).‘ � . . . � ’ stands for space left for related adjectives; no one is provided forÉPARGNE

22 L’HOMME

1 l’ÉPARGNE – [epar (@)] – (n.f.)

1.1. (emploi le plus fréquent avec un article défini) Opération par laquelle unagent économique (un particulier, une entreprise, un organisme, un état, unÉtat – X) réserve . . .Syn. : (K 242) Pour en savoir plus, Épargne (sens 1.1.) . . .Les personnes âgées s’étonnent du faible effort d’épargne de la jeunegénération.

1.2. (emploi au sing. et au plur.) Somme d’argent réservée à l’épargne (sens1.1.). . .

2.1 Avantage qu’une personne tire d’une action.Syn. : une économie, un gainLa modernisation de la facturation permis de réaliser une épargne de tempsconsidérable.

Figure 5: Distinctions of senses in the DAFA (Binon et al., 2000:240)

abstract meaning captured by a standard lexical function (especially for verbs).Explanations and examples are also provided in this section.

expressions(sens 1.1.)L’épargne tue l’épargne : phénomène selon lequel l’épargne entraîne une diminutiondes ventes, puis de la production et, enfin, des revenus, suscitant une diminution del’épargne.

Figure 6: Expressions in the DAFA (Binon et al., 2000:241).

Even though only parts of a typical entry were reproduced above, we canbase our comparison between the DAFA and the ECD on them. Senses are clearlydistinguished with numbered definitions. The numbering is then used throughoutthe articles to relate a piece of information to the relevant sense (each collocationand expression is clearly related to a given sense). A large part of the article isdevoted to the listing of typical expressions and collocations, and collocationsare even further classified semantically. Finally, semantically-related terms areprovided in different parts of the article: related concepts next to the headword;synonyms and antonyms are listed next to the relevant definition and explainedat the end of the article; formally-related terms are listed under the headword.

3.2.3 Specialized combinatorics

Recently, terminologists have become increasingly interested in combinatorics.According to many authors, it is no longer sufficient to list terms, provide equiv-

ECL IN TERMINOLOGY 23

+ adjectif(sens 1.2.)L’épargne constituée : somme d’argentépargnée. Je peux bénéficier de l’épargneconstituée dans les mêmes conditions quesi mon époux était resté en vie.

TYPE D’ÉPARGNE (sens 1.1.)L’épargne populaire. (Syn. (plus fréq.)l’épargne des ménages). � L’épargneprivée :

épargne des ménages, des entreprises etdes institutions financières.. . .L’épargne libre, volontaire : suite àune décision prise librement par l’agentéconomique . . .L’épargne contractuelle : . . .L’épargne individuelle : . . .

+ nom(sens 1.1.)

� Un instrument d’épargne : Lecompte d’épargne reste un instru-ment d’épargne très populaire.

� Un compte d’épargne, un livretd’épargne, un carnet d’épargne. . .

. . .CARACTÉRISATION DE L’ÉPARGNE

(sens 1.1.)Une épargne de précaution : épargne

réservée pour faire face aux risques de lavie : maladie, perte d’emploi, . . .

NIVEAU DE L’ÉPARGNE

(sens 1.1.)Le taux d’épargne . . .L’épargne à court terme . . .MESURE DE L’ÉPARGNE

(sens 1.2.)L’épargne à long termeLe volume de l’épargne . . .

+ verbe : qui fait quoi ?(sens 1.1.)

une entreprise faire (un appel) (public) à l’ � �

une mesure stimuler l’ � la stimulation de l’ �

� � décourager l’ � le découragement de l’ �

Cette société a confirmé son intention de faire un appel public à l’épargne sous la formed’une augmentation de capital.. . .

Figure 7: Combinatorics in the DAFA (Binon et al.,2000: 241-242)

24 L’HOMME

alents in other languages and define them with analytical definitions. Users,especially translators, also need to know how to produce them in specializedcontexts. Some of this work can be related to ECL principles.

In a proposal to enrich terminological databases with specialized colloca-tions, Heid and Freibott (1991) describe these combinations using distinctionsproposed in (Mel’cuk et al., 1984a) and (Hausmann, 1979). According to Heidand Freibott, a combination such as créer un fichier (Engl. ‘create a file’) is acollocation, since CRÉER cannot be replaced by another lexical unit that conveysthe same meaning (e.g. *établir un fichier, Engl. ‘establish a file’; *concevoirun fichier, Engl. ‘design a file’). In accordance with ECL, the components ofa specialized collocation are a keyword (FICHIER), whose meaning remains un-changed, and a collocator (CREATE), whose meaning is either specified or ac-quired through its use within the combination.

Other specialized lexicographers have classified collocations using princi-ples similar to those of ECL. We have already pointed out that the DAFA (Sec-tion 3.2.2) uses a finite list of categories to classify groups of collocators ac-cording to their meaning. Cohen (1986) and Laporte and L’Homme (1997) alsoused similar methods for the terms of the stock market and those of pharma-cology respectively. Cohen, for example, in a dictionary entitled Lexique decooccurrents: Bourse et conjoncture économique, lists verbs, nouns and adjec-tives typically found in the vicinity of terms and provides a general classificationfor them. Figure 9 reproduces the article for CAPITAL (Engl. ‘capital’) found inthe dictionary.

Even though the semantic categories chosen by Cohen are very general (andencompass distinctions between parts of speech), similarities between these cat-egories and standard lexical functions can easily be established (Figure 8).

Magn(capital) = appréciable, considérable, élevé, grosIncepPredPlus(capital) = s’accroître, augmenter, croîtreAntiMagn(capital) = petitIncepPredMinus(capital) = baisser, diminuerOper1(capital) = avoir, posséder [ART � ]

Figure 8: Cooccurrents listed in Cohen (1986) and their representation in termsof lexical functions

The work conducted on specialized lexical combinations shows that lexicalfunctions (or at least a classification method based on the same idea) can be usedto represent syntagmatic relationships between terms and the words with whichthey usually combine. We saw that Frawley (1988) uses lexical functions to cap-ture a wider range of relationships, i.e., not only syntagmatic relationships, but

ECL IN TERMINOLOGY 25

CAPITAL2 : Somme d’argent assez considérable, possédée par un individu, uneentreprise ou un État et disponible pour des dépenses ou des investissements.

Noms Verbes Verbes Adjectifs(sujet) (objet)

DÉBUT formation accumulermobilisation amasser

constituerformermobiliser

CROISSANCE s’accroître accroître appréciableaugmenter améliorer considérablecroître augmenter élevé

grosINDÉTERMINÉS circulation circulerDÉCLIN baisser grever petit

diminuer prélever (sur)réduire

FIN absorberépuiser

AUTRES investissement avoirCOOCCURRENTS placement posséder

engagerinvestirfaire valoirplacer

Figure 9: Article CAPITAL2 extracted from (Cohen 1986:20)

also paradigmatic relationships, such as synonymy, antonymy, semantic deriva-tion, etc.

3.3 What These Endeavors Teach Us about Terms

The work described in Sections 3.1 and 3.2 reveals several things about termsthat have not been taken into account by terminology for the reasons we listed inSection 2. I will try to summarize its major repercussions in relation to conven-tional terminology principles.

First, it shows that all types of parts of speech are profitably taken into ac-count in specialized dictionaries, and not only nouns. Descamps (1973, 1976),Descamps et al. (1992), and Binon et al. (2000), for instance, make lists ofmorphologically-related lexical units, and some of these are verbs, adjectives,and adverbs. Research on combinatorics show that words typically used with

26 L’HOMME

terms need to be listed in dictionaries.

Even though polysemy is not always addressed explicitly, it is certainly dealtwith by different methods. Binon et al. (2000) make distinctions based on def-initions, combinatorics, and synonymy relationships. Descamps (1973, 1976)bases his distinctions mostly on contextual patterns. Frawley (1988) speaks of“vagueness” that could be resolved with the help of a model that accounts forthe propositional form and lists lexical functions in a formal dictionary.

Descamps (1973,1976) and Frawley (1988) clearly insist on the importanceof a clear identification of the actant structure (although, they both use otherdesignations to refer to the actant structure) in the description of specializedunits. This work raises another interesting issue: that of the distinction betweenpredicative and non-predicative units in terminology to which I will come backin a later section of this paper.

Finally, combinatorics is a central matter in most of the work described andterms related semantically (or, at least, morphologically) are often considered askeys to understanding the sense of the term under examination.

4 Using ECL at all levels of the description of terms

In the following section, I want to show how the application of certain princi-ples of the Explanatory and Combinatorial Lexicology can help terminologistsmake important distinctions when analyzing terms in specialized corpora. Moreprecisely, I will demonstrate that some criteria designed within ECL provide asound foundation for making fine-grained semantic distinctions (confirm anddistinguish specialized senses). It is also very useful to represent terms withpredicative meaning, since actant structure is expressed formally. Lexical func-tions capture a wide variety of semantic relationships between terms. And, fi-nally, various components in ECL-descriptions, even though they have not beendesigned for that purpose, can be used for extracting classes of semantically-related terms.

Parts of these studies can be related to the work described in Section 3. I willmake cross- references whenever relevant.

4.1 Confirming Specialized Senses

As I pointed out in the introduction, ECL-principles are useful in terminologyonly if one admits that, among the variety of senses a word can acquire, somecan be defined as “terminological”, i.e., senses that should be listed and de-scribed in a specialized dictionary, such as a dictionary of biology, a dictionary

ECL IN TERMINOLOGY 27

of shipbuilding, or any other dictionary the purpose of which is to describe thevocabulary of a specialized field of knowledge.

This means that, when they browse through specialized texts, terminologistsmake a selection among the words that appear in them. They consider only thosewords that convey a sense that can be associated with a specialized domain,and furthermore, only those senses—if the word is polysemic—related to thedomain.

It is appropriate to underline at this point that the Explanatory and Combina-torial Dictionary contains descriptions of lexical units (= LUs) I would consideras “terminological”. For example, ACCUSER2. (Engl. ‘accuse’) covers the sensesconveyed by the unit in texts on law (Figure 10).

2. plutôt au passif. Y est accusé [par X] de Z devant W = Il est déclaré devantl’autorité W par X, en conformité avec la fonction légale que X exerce dansune société et avec la procédure prévue par la loi, que X tient Y officiellementresponsable du fait Z, qui va à l’encontre des lois de cette société ou des règlesd’éthique d’une corporation professionnelle, dans le but qu’un jugement sur laculpabilité de Y et sur la punition correspondante de Y soit rendue par W.

Figure 10: Definition of the lexical unit ACCUSER2. in Mel’cuk et al. (1999:93)

However, from the ECL-perspective, this lexical unit is considered in its re-lation to other LUs within the vocable ACCUSER. In addition, some senses arecaptured in a “generic way” (for example, “head” will be described as the toppart of an implement, which covers several technical senses); terminologists willneed to consider these senses from the point of view of a domain (hence, “head”will be considered differently in separate fields of knowledge). In my view, ter-minological work will focus only on the lexical unit (or lexical units) associatedwith the field being examined and the potential semantic relationships that can begathered from this lexical unit. I will call these specialized lexical units (SLUs).

Thus, the selection of terms is based first and foremost on extralinguisticcriteria, i.e., the possible relation an LU has with the domain under examina-tion. Terminologists can establish this relation once they have acquired a soundknowledge of the field; for difficult cases, they usually refer to specialists.

In specialized texts, some words are exclusively used (or, in a majority ofoccurrences) in their terminological sense. For example, in a corpus composedof texts on computing, memory will always be used to refer to “the storage com-ponent of a computer”, and not to “a human faculty”.

�

Similarly, in a text onconstitutional law, the unit constitution will systematically refer to “A set ofprinciples and guidelines some States define in order regulate their functioning”as shown in (1).

28 L’HOMME

(1) The judiciary is the guardian of the constitution and must, in interpretingits provisions, bear these considerations in mind.. . . but was, nevertheless, legally bound by the new constitution by virtueof its membership in the Canadian unionIn Canada, we were granted a constitution by the British sovereign (Par-liament and the monarch).The point is that some elements of the constitution call for vertical ap-plication . . .

Moreover, a specialized lexical unit that refers to an important concept ina given field can produce several derivatives that are semantically related. Forexample, constitution is related to a set of nouns, a verb, an adjective, and anadverb (contexts have been reproduced in (2)).

(2) Negotiations would avoid the problems inherent in seeking aconstitutional amendment or taking claims to the courts.. . . in challenging the orthodox approach of constitutionalism, JudgeMadala points out that both the South African Constitution and SouthAfrican society are pluralistic.. . . If a government was concerned about the constitutionality of a statute,it could refer the statute to the Court of Appeal (or in the case of Parlia-ment . . . )In Martineau, there is a sense that the Court is rushing the process ofconstitutionalization of fault.The upshot of the three judgments in Lavigne is consistent with thestance taken in the Labour Trilogy not to constitutionalize collective bar-gaining issues.The full range of values and goals of such a society is broader than thelatter, so that, constitutionally and legally, these other values and goalsmay well come into play

However, in many cases, SLUs interfere with other LUs, i.e., senses that can-not be associated to the domain under examination. In addition, the same wordcan be associated with the different LUs that can all be associated with the do-main.

When making semantic distinctions, terminologists can resort to criteriafound in ECL to confirm intuitions they have about potential specialized lexicalunits or make distinctions for polysemic words. We will look at three differentcriteria, namely (a) compatible cooccurrence; (b) differential coccurrence; and(c) differential derivation (more criteria are formulated in Mel’cuk et al. 1995).I will illustrate their application using contexts extracted from specialized texts(i.e., texts on computing and law).

ECL IN TERMINOLOGY 29

a. Compatible cooccurrence (Mel’cuk et al., 1995:64): If, for a given po-tential lexical unit (= LU) ‘. . . , �

�,. . . , �

� �,. . . ’, a normal sentence with com-

patible cooccurrence can be built, then�

must not be separated; we obtainone lexical unit and its definition should state a disjunction LU ‘. . . �

�’ or

‘ �� �

. . . ’.

This first criterion will help make a decision about the occurrences of in-dividual in (3) and (4).

(3) In Charter cases, minimum standards of respect for individual au-tonomy must be specified with precision in analysis under section1, the omnibus limitations clause.

(4) Collectivity necessarily circumscribes individuality and the morecomplex and sophisticated the collective structures become, thegreater the threat to individual liberty in the sense protected by 8.

INDIVIDUAL can combine with the two cooccurrents found in these sen-tences. Since these cooccurrents are compatible, INDIVIDUAL will notbe divided into two separate lexical units. We will see further that othercooccurrents of INDIVIDUAL are not compatible.

� Respect for individual autonomy and liberty

The same criterion is applied to EXECUTE whose cooccurents appear in(5), (6) and (7). Since the cooccurrents found in these sentences are com-patible, all three occurrences of execute refer to the same LU.

(5) The MSDOS.SYS file works with the BIOS to manage files,execute programs, and respond to signals

(6) . . . which measure the time a PC takes to execute a set of applica-tion scripts . . .

(7) The trouble is that these applications will also execute any macroswithin the received file, thus enabling the virus to infect.

� Execute a set of application scripts, programs, and macros.

EXECUTE does not always combine with compatible cooccurrents; cf. (8).The next criterion will be used to define this situation.

(8) We will describe how to execute a typical installation (it shouldwork in every Linux version, but it has been only completely testedwith Debian).

� *execute a program and an installation

30 L’HOMME

b. Differential cooccurrence (Mel’cuk et al., 1995:66): If, for a given po-tential lexical unit (= LU) ‘ �

�,. . . , �

� �,. . . ’, two different sets of cooccurrents

can be identified, the first one corresponding to ‘ ��’; the second to ‘ �

�’,

then LU must be divided.

This second criterion will show that the two uses of INDIVIDUAL in (9)and (10) and FORMAT in (11) and (12) refer to two different LUs sincetheir cooccurrents are not compatible.

(9) In Charter cases, minimum standards of respect for individual au-tonomy must be specified with precision in analysis under section1, the omnibus limitations clause.

(10) Specific circumstances of each individual case must be examined. . .

� *Individual autonomy and case

(11) Often, teachers can format tests and spreadsheets . . .You can type, edit and format documents just by talking.HTML uses a set of standard codes to format text.

� Format tests, spreadsheets, documents and text

(12) You can either install Windows into an existing partition, or firstdelete a partition, create a partition, format a partition . . .Now to format a disk left click on Format.

� format a disk, and a partition, *format a document and a partition

Differential cooccurence also defines the situation shown in examples (5),(6), and (7) versus (8). Hence, we obtain: INDIVIDUAL1, INDIVIDUAL2,FORMAT1, FORMAT2, EXECUTE1 and EXECUTE2.

c. Differential derivation (Mel’cuk et al., 1995:68): If, for a given potentiallexical unit (= LU) ‘. . . �

�,. . . , �

� �,. . . ’, two different sets of derivatives can

be identified, the first one corresponding to ‘ ��’; the second to ‘ �

� �’, then

LU must be divided.

This criterion shows that address in texts on computing refers to two differ-ent LUs. The senses exemplified in (13) cannot be related to addressable,whereas the senses displayed in (14) can be related to addressable. Thememory is adressable, but not the location.

(13) You can enter the address of the location you wish to visit and thebrowser . . .

ECL IN TERMINOLOGY 31

� *addressable location

(14) At the machine level that location has a memory address. The fourbytes at that address are known to you, the programmer, as I, andthe four bytes can hold one integer value.

The 386 has a huge amount of addressable memory compared tothe 286.

Similarly, two LUs for INSTALL can be distinguished since one(15) can be related to UNINSTALL whereas the other cannot (16).

(15) Select the directory to which you want to install Windows.Don’t install software just because you can.

� Uninstall Windows, Uninstall software

(16) Now you need to install the motherboard into the case.The first step is to install the I/O connectors, such as your paralleland serial ports.Also, some cases give room under the power supply to install ahard drive.

� *uninstall the hard drive

The application of these last two criteria (differential cooccurrence anddifferential derivation) will lead to: (a) the rejection of LUs that do notappear to be relevant in the domain under examination; or (b) the prepa-ration of separate terminological descriptions for LUs that can be relatedto the domain. For example, FORMAT corresponds to two different LUs,numbered below as FORMAT1 and FORMAT2. The same applies to install.Each LU will be associated with different sets of LUs, cooccurrents, syn-onyms, antonyms, etc. (Figure 11).

This approach prevents the type of vagueness that Frawley (1988) has crit-icized and that is exemplified in the following definition of CONFIGURATION

found in a term bank (Figure 12). Does it refer to an activity (the act of config-uring) or a result?

4.2 Describing Predicative Terms

Most work described in Section 3 show that parts of speech other that nouns,such as adjectives, verbs and adverbs, can be taken into account in specializeddictionaries � . However, these categories cannot be defined in terms of analytical

32 L’HOMME

FORMAT1 test INSTALL1 motherboardspreadsheet I/O connectorsdocument hard drivetext

FORMAT2 disk INSTALL2 Windowspartition softwareC drive Anti uninstalldisquette

Figure 11: Lexical units associated with FORMAT1 and FORMAT2

CONFIGURATION

the arrangement of a data processing system or network as definedby the nature, number, the interconnection and chief characteristicsof its functional units (EURODICAUTOM)

Figure 12: Definition of CONFIGURATION

definitions typically used in terminology that are listings of intrinsic characteris-tics. As I have already pointed out, terminology is much more comfortable withnon-predicative units (or units considered as being non-predicative), and muchless with predicative units such as verbs and adjectives.

Thus, I will define these categories as “predicative lexical units” and resortto definitions in which the participants are explicited, as did Frawley who ap-plied ECL-principles (Figure 1). However, it must be remembered that certaincategories of nouns are also predicative lexical units.

For example (17), in order to define INSTALL2 cited in the previous section,three variables corresponding to three semantic actants are provided: X, a per-son, installs Y, a program, on W, a permanent storage device.

(17) INSTALL2: X installs Y on Z

Then, a defining statement giving a description of the lexical unit can beprovided in a semi-controlled vocabulary (18).

�

(18) INSTALL2: X installs Y on ZX puts Y on Z and makes a certain number of settings so Z will functioncorrectly.

ECL IN TERMINOLOGY 33

In addition, certain regularities displayed by actants can be taken into accountin the actant structure of a specialized lexical unit. In a certain number of studiesof specialized lexical units in different domains (philosophy and air traffic con-trol) and computing, we have shown that specialized predicative units combinewith terms sharing semantic properties. Examples (19) illustrate this observa-tion: INSTALL2 combines with OPERATING SYSTEM, WINDOWS, PACKAGE,WORD, SURFER, terms that denote a “piece of software”.

(19) Once the operating system is installed, you can install the drivers.Users install Windows 98 on their portable computers.This package cannot be installed on 80486 computers.You can install Word on your computer from this CD-ROM.This routine will assist you in installing your web surfer.

In fact, this property does not only apply to the second actant of INSTALL2

(typically, its direct object) but also to its other actants, as shown in (20).

(20) user INSTALL driver ON portableprogrammer routine laptopetc. program disk

Word hard disketc. etc.

Thus, when accounting for specialized LUs, the variables given in the propo-sitional form and the definition can be explicited with a set of labels (similarto those used by Descamps et al. 1992, refer to Figure 2) representing concep-tual classes. For example (21), the defintion of INSTALL2 will contain ‘user X’,‘executable information Y’, and ‘permanent storage Z’.

(21) INSTALL2: X installs Y on ZA X(user) puts Y(executable information) on Z(permanent storage) andmakes a certain number of settings so Z(executable information) willfunction correctly.

The actant structure of over 200 French verbs related to computing and theircorresponding nominalizations was described using conceptual classes, thusdemonstrating the productivity of this approach. Concept classes were furtherorganized into a hierarchy that accounted for the fact that certain verbs combinewith a small number of actants (e.g. click with a mouse), while other com-bine with several actants (e.g. configure program/mouse/operating system/ key-board/computer/printer/. . . ).

34 L’HOMME

4.3 Capturing Relationships between Terms

Terminologists have been interested in representing relationships between termsfor a long time. Although the design of solid representations can be extremelycomplex, it is obvious why this issue has attracted so much interest. Terminolog-ical descriptions bear on finite sets of terms related to the same body of knowl-edge. Hence, it appears possible to identify subsets of semantically-related termsand to account for these relationships in a formal manner.

Up to now, the focus has been on hyperonymic (apple � fruit) andmeronymic relationships (apple � core), which appear to be more compatiblewith the organization of knowledge in scientific domains. More recently, syn-tagmatic relationships have attracted interest (cf. Subsection 3.2.3). In otherstudies, new sets of relationships are highlighted: for example, Binon et al.(2000), Descamps (1973, 1976) and Descamps et al. (1992) give lists ofmorphologically-related terms (see Subsections 3.2.1 and 3.2.2).

Several problems arise when trying to account for this variety of relationshipsin a formal representation. First, there seems to be a confusion between concep-tual endeavors that are attempts to capture the knowledge structure of a subjectfield and lexical representations that focus on relationships between lexical unitssuch as synonyms, derivatives or cooccurrents. Secondly, some representationsare designed to capture a given set of relationships and cannot account for others(e.g., a model that captures hyperonymic relationships, is not normally suited toaccount for co-occurrence) leading to a problem of compatibility between rep-resentations.

Even printed dictionaries are not consistent in their listing of relationships.For example, Binon et al. (2000), even though they establish different typesof relationships, disperse them throughout the article. Morphologically-relatedterms are listed under the headword (no mention is made of the nature of therelationship); near-synonyms, and hyperonyms are stated by means of cross-references; cooccurrents are provided in a different section; etc. (see Subsec-tion 3.2.2).

Lexical functions (LFs) appear to be a solution to this problem. The main ad-vantage of lexical functions, when applied to the representation of semantic re-lationships between terms, is that they can capture a wide range of relationshipswithout the need to turn to different models. They have been used by Fraw-ley (1988) in an exhaustive description of the unit MASS. However, the authorwas more concerned with the principles and did not apply them to large-scalevocabularies.

I hope to demonstrate how lexical functions can be implemented when deal-ing with semantic relationships between terms through the description of aproject on the computerization of a printed dictionary on retailing (Dancette &

ECL IN TERMINOLOGY 35

L’Homme, 2002). I will also discuss the repercussions of such an implementa-tion for terminological relations.

The printed dictionary (Dancette & L’Homme, 2002) contains 350 English-French articles that provide exhaustive descriptions of what are defined as keyconcepts of the field, such as “marketing”, “shopping center”, “auction”, “prod-uct”, “customer”. Articles include up to nine sections: (i) the English head-word and its synonyms followed by usage marks, (ii) French equivalent terms,(iii) a Definition, (iv) Semantic precisions (used to add information to the oneprovided in the definition), (v) Semantic relations between the terms belongingto a single field (relationships of hyperonymy and hyponymy, quasi-synonymyand antonymy), (vi) Additional information providing extralinguistic informa-tion (historical notes or pragmatic information), (vii) Linguistic information, andfinally, a (viii) French and (ix) an English context (which are example sentencesextracted from the corpora used to compile the dictionary).

A number of statements disseminated throughout the article are provided toexplain semantic relationships between the headword and other terms referringto “related concepts” (typical sentences are given in (22)). During the comput-erization process, we had to provide an access using an apparatus that wouldsystematize the relationships and explain them. Due to the variety of seman-tic relationships to represent, lexical functions appeared to be the most obviouschoice.

�

Figure 13 shows how the information contained in examples (22) wasconverted.

(22) According to life span, there is a distinction between durable goods,semidurable goods and non-durable goods. (antonymy)A category-killer is a big box store operated as a self-service and spe-cializing in a non-food product category. (hyperonymy)Clientele = All the customers of a store or a service retailer. (meronymy)Before adopting a new product, the consumer goes through a series ofstages that define the adoption process. (nominalization)The adopter, namely the consumer who adopts a product, [. . . ]. (theagent of the event, which happens to be a morphologically related term)Arborer une enseigne, porter une enseigne = carry a banner (lexical unitsthat combine with the term BANNER (and its French equivalent, EN-SEIGNE))

Approximately 2000 semantic relationships were captured using standardLFs,

� � which shows that it is well-suited to represent relationships betweenterms. Of course, LFs can also represent semantic relationships in other domainsand not exclusively retailing. This is shown in Table 1 and was also demonstratedin other work mentioned in Section 3.

36 L’HOMME

Table 1: Lexical functions applied to terms in different fields

Domain Lexical functionsFinance Syn(devalue) = devaluateComputing Syn � (computer) = portableFinance Syn � (funds) = moneyComputing Syn � (execute) = accomplishComputing Anti(install) = uninstallLaw Conv � � (defend) = accuseComputing Gener(printer) = peripheralComputing S � (compatible) = compatibilityLaw S � (defend) = defenceRetailing V � (layaway) = to layawayLaw A � (constitution) = constitutionalLaw Adv � (constitutional) = constitutionallyLaw S � (defend) = defendantComputing S � (lauch) = applicationComputing S � (install) = diskLaw S � (defend) = courtRetailing S � ��� (labelling) = labelRetailing Sing(clientèle) = clientComputing Mult(program) = [ � ] libraryRetailing A � (banner) = under [ART � ]Computing A � Perf(install) = installedComputing Able � (address) = addressableLaw Able � (defend) = defendableLaw AntiAble � (defend) = undefendableFinance Magn(assets) = highComputing Bon(program) = powerfulRetailing Oper � (banner) = carry [ART � ]Retailing Labor � � (banner) = set up [N under ART � ]Computing Real � (program) = run, execute [ART � ]Computing Fact � (program) = runs

Fact � (computer) = processes [data]Law Labreal � � (constitution) = // constitutionalizeLaw S � Labreal � � (constitution) = // constitutionalizationStock exchange IncepPredMinus(capital) = decreases

ECL IN TERMINOLOGY 37

Anti(durable good) = non-durable goodGener(big-box store) = category killerMult(customer) = // clienteleS � (adopt) = adoptionS � (adopt) = adopterV � (counterfeiting) = counterfeitOper � (banner) = carry [ART � ]

Figure 13: Representation of semantic relationships with lexical functions

Even though standard lexical functions capture a wide range of semantic re-lationships between terms, the project on the vocabulary of retailing reveal anumber of incompatibilities between conventional terminological practice andECL. This does not invalidate the usefulness of lexical functions for terminolog-ical purposes. It simply points to the fact that some type of adaptation needs bemade in terminology and ECL.

(a) Taxonomic relationships and meronymic relationships are central in termi-nological representations. ECL provides means to capture these relation-ships only partly by standard LFs. Moreover, it is likely that the definitionsgiven in both frameworks do not correspond.

(aa) Hyperonymy can be captured using Gener or Syn � . In the project, weused Gener, which appeared to be more compatible with standard termi-nological practice, but a thorough examination of this issue needs to becarried out.

(ab) In a large number of cases, we needed to use the Spec-LF proposed byGrimes (1990) to capture hyponymic relationships mentioned in articles(e.g. Spec(service) = self-selection). In addition, hyponymic relationshipscan be captured with the Syn � -LF.

(ac) Co-hyponyms are represented in a number of ways. They can be captureddirectly by means of a synonymic function, that is Syn � (this LF translatesthe relationship existing between LUs that have common and different se-mantic components). Hyponyms that have a relationship of opposition canbe described with the Anti-LF.

(ad) The synonymy LF (Syn) does not appear to correspond to the definitionterminology has traditionally given to this relationship. In terminology,two terms are considered as true synonyms when they are said to refer tothe same class of objects. ECL adopts an approach which is more com-patible with standard lexicographical practice in which synonyms are sel-dom completely “true”. Antonymic relationships also need to be examined

38 L’HOMME

from the point of view of terminology, since they have received very littleattention.

(ae) ECL provides only one pair of standard lexical functions to cap-ture meronymic relationships, Mult which refers to a collection (e.g.,Mult(client) = clientele); Sing(assortment) = item), the inverse function,which refers to the member within a collection. Other partitive relation-ships were found in the dictionary and are most likely to be present in otherspecialized vocabularies but could not be captured by means of a standardlexical function. The first one is the part-whole relationship (in which thepart plays a functional role). For this particular partitive relation, we re-sorted to the LF Part proposed by Fontenelle (1997). (An inverse functioncould also be used to capture the opposite relation) (e.g., Part(departmentstore) = aisle). Finally, we had to take into account a partitive relation-ship shared between a process and its phases (e.g., Phase(prospection) =canvassing).

(b) Terminology needs to reflect on the status is gives to complex terms. Ihave already pointed out, in Section 2, that specialized dictionaries list alarge number of complex terms that are completely compositional. Thischoice has a number of consequences on the way LFs are assigned. Letus examine a concrete example. In the dictionary on retailing, GOOD andDURABLE GOOD, are both listed as headwords. Should DURABLE GOOD

be represented as a hyponym of GOOD? Or should DURABLE be the valueof an adjectival lexical function applied to GOOD?

(c) Although they have not been designed for that purpose, lexical functionsare useful to reveal groups of semantically-related terms. This featureis particularly interesting to make use of in specialized domains, sincemany terms belong to conceptual families (e.g., in medecine, it is verylikely that terms referring to several types of diseases will be considered;in computing, different peripherals should be described).

(23) Gener(off-price store)Gener(category killer)Gener(liquidation center)Gener(liquidation store)Gener(factory outlet store)Gener(warehouse store)Gener(big-box store)Gener(hypermarket)Gener(deep discount store)

��������������������������

= discount store

Examples in (23) show how the Gener-LF, a paradigmatic LF, can reveala set of co-hyponyms in the field of retailing. If the Gener-LF is applied

ECL IN TERMINOLOGY 39

to different terms and yields the same value, these terms can be defined asLUs belonging to the same taxonomy.

Similarly, syntagmatic LFs can be exploited to find classes of terms col-locating with the same units. Example (24) shows how the terms thatcollocate with run can be found through the examination of the value ofthe LF Real � .

(24)Real � (program)Real � (application)Real � (word processor)

�= run [ART � ]

4.4 What This Work Tells Us about Terms

The studies described in Section 4 have first confirmed that some approachesbased on or compatible with Explanatory and Combinatorial Lexicology remainvalid even when they are applied to larger sets or data of the terms or differentdomains.

Predicative units can be included in specialized dictionaries provided they aredefined with a proper apparatus, i.e., one that clearly indicates semantic actants.In addition, specialized actants can be given with semantic labels that captureclasses of terms. This has been shown by the work on verbs related to the fieldof computing and to a lesser extent by the contextual patterns used in Descamps(1973, 1976) and Descamps et al. (1992).

The conversion of statements about semantic relationships into lexical func-tions carried out on the dictionary of retailing demonstrates that a wide variety ofsemantic relationships between terms can be represented with lexical functions.This is an application of Frawley’s ideas on a large scale vocabulary.

Finally, I have shown that ECL is not only useful to describe terms or termi-nological senses once they have been identified. Different criteria can be usedto make subtle semantic distinctions when potential specialized LUs are underexamination. Terminologists could resort to these criteria and base distinctionson linguistic tests rather than always rely exclusively on extra- linguistic criteria,namely knowledge of the domain.

5 Conclusion

In this paper, I have tried to show that studies based on Explanatory and Com-binatorial Lexicology and other studies with compatible methodologies are use-ful to study terms and consider them from a wide range of viewpoints. Some

40 L’HOMME

of the work presented here has also revealed various characteristics of termsthat have seldom attracted interest in terminology, since conventional terminol-ogy has been mainly concerned with standardizing lexical usage and consideredterms as a reflection of the organization of concepts within a field of knowledge.

Of course, more work needs to be carried out in order to find out the extentto which the ECL-framework can lead to richer terminological descriptions. Ihave listed a number of interesting contributions in that area but terminologistsinterested in this kind of endeavor must develop a methodology of their own.They can use the work presented in this article, but a systematic approach isnevertheless required.

This methodology will inevitably lead to a drastic change in focus when con-sidering terms. Terminologists will need to base their findings first and foremoston occurrences in specialized texts and associations between words as they canbe observed in texts and no longer reflect on the way concepts are named orshould be named.

With these observations, we return to the quote by Igor Mel’cuk I reproducedat the beginning of the article. It appears that a choice must be made betweenconceptual and linguistic analyses. In terminology, concept-oriented endeavorsfocus on aspects related to the organization of knowledge and this choice leadsto a concealment of the linguistic properties of terms. On the other hand, lin-guistic approaches distance themselves from conceptual matters since they areconcerned with the instantiations of terms and do not accept or simply ignore thefact that they reflect the organization of knowledge (in fact, in most linguistics-oriented studies, the issue is not raised or considered as relevant).

Both approaches are likely to be necessary to provide a better understandingof terms. However, they do not seem compatible. So it appears that, even interminology, a choice must be made.

Acknowledgements

I would like to thank Teresa Cabré, Alain Polguère and Juan C. Sager for theircomments and suggestions on the initial drafts of this article.

Notes�

I might not have reproduced the quote in its exact form (and I might not have translated eachword in the most accurate manner). Nevertheless, it was another very useful distinction I couldapply to terminological data and analysis.

�“Les termes étant des unités lexicales définies, ils ne représent, potentiellement, que cer-

taines acceptions de l’aire sémantique de l’unité lexicale, à savoir celles qui sont définies parles spécialistes dans les textes spécialisés, et non par les lexicographes dans les dictionnairesgénéraux” (Kocourek, 1991:180).

ECL IN TERMINOLOGY 41

�

Part of this work has been carried out with the help of undergraduate and graduate students.I gratefully acknowledge the contributions of Claudine Bodson, Claudine Bertrand, Anne-LaureJousse, Philippe Hanscom, and Isabelle Laporte. In addition, the project on the dictionary ofretailing was conducted in collaboration with my colleague Jeanne Dancette.

�

Most of this work was conducted before the principles of ECL were made available. Authorsrefer to (Tesniere, 1966) and (Gross, 1966).

�

Semantic labels appear to be a productive approach to the dscription of specialized vocabu-laries. This will be discussed further in Subsection 4.2 It should also be pointed out that semanticlabels are used extensively in the electronic version of the Explanatory and Combinatorial Dic-tionary, the DiCo, developed under the supervions of Alain Polguère. The definition of semanticlabels and their implementation in ECL are presented in (Milicevic, 1997).

�

This also depends on the nature of the corpus itself and the method that was used to assembleit.

�

I also claimed that verbs should be considered as terms (cf.L’Homme, 1998 for a list ofarguments).

�

A whole set of rules have been developed for definitions in ECL. I have not yet applied themto the definitions of terms and cannot comment on their application for terminological purposes.However, I can anticipate their usefulness in systematizing the explanations of lexical units in aspecialized domain, since many are semantically-related.

�

This approach was applied successfully by Fontenelle (1997) who converted the informa-tion contained in a bilingual dictionary using lexical functions. We applied it to a specializeddictionary.�

�

Some cases were set aside due to a lack of information in the dictionary.

42 L’HOMME

Appendix A: Lexical Functions Cited in the Article

LF Example Definition(according to Mel’cuk 1998,Mel’cuk et al. 1995)

A � A � (constitution) = constitutional adjectivalizationAble � Able � (address) = addressable passive potential adjectival

AntiAble � (defend) = undefendableAdv � Adv � (constitutional) = constitution-

allyadverbalization

A � A � (banner) = under [ART � ] passive adjectivalA � Perf(install) = installed

Anti Anti(install) = uninstall antonymBon Bon(program) = powerful positive modifierCentr Centr(mass) = center of mass noun for the center of [ � ]Conv � � Conv � � (defend) = being defended conversiveCulm Culm(mass) = point mass noun for the highest point of [ � ]Excess Excess(mass) = mass excess to function in an excessive man-

nerFact � Fact � (computer) = processes [data] realize, fulfil [the requirement of]

when the key word is subjectFin FinS � (mass) = annihilation, conver-

sion [into energy]the end of [ � ]

Func � Func � (mass) = react to, determine do, perform [support verb whenthe key word is subject]

Gener Gener(printer) = peripheral noun for genericIncep IncepPredPlus(capital) = increases the beginning of [ � ]Labor � � Labor � � (banner) = set up [N under

ART � ]do, perform [support verb whenthe key word is indirect

Labor � � (mass) = determine object]Labreal � � Labreal � � (constitution) = //constitu-

tionalizerealize, fulfil [the requirement of]when the key word is indirect ob-ject

Liqu LiquReal � (command) = cancel causation of non-existenceMagn Magn(assets) = high intense(ly), very [intensifier]Minus IncepPredMinus(capital) = decreases less of [ � ]Mult Mult(program) = [ � ] library collective (quantifier)Oper � Oper � (banner) = carry [ART � ] do, perform [support verb when

the key word is direct object]Plus PlusFunc � (mass) = increase more of [ � ]Pred Pred(mass) = be copula

ECL IN TERMINOLOGY 43

Propt Propt(mass) = because of, out of, from preposition or prepositionalphrase for “because of”

Real � Real � (program) = run, execute [ART

� ]realize, fulfil [the requirement of]when the key word is direct ob-ject

S � S � (compatible) = compatibility nominalizationS � (defend) = defence

S � S � (defend) = defendant agent nounS � S � (lauch) = application patient nounS � S � (licence) = licencee noun for third actantS � S � (defend) = court noun for fourth actantSing Sing(clientèle) = client singulative (quantifier)S ��� ��� � S ��� ��� � (pallet) = stacker noun for instrumentS ����� S ���� (retailing) = shopping center noun for placeS � � � S � � � (print) = printout noun for result

S � � � (defend) = acquittalSyn Syn(devalue) = devaluate synonymSyn � Syn � (computer) = portable synonym with poorer meaningSyn � Syn � (funds) = money synonym with richer meaningSyn � Syn � (execute) = accomplish synonym with intersecting mean-

ingV � V � (layaway) = to layaway verbalizationVer Ver(mass) = reste mass, proper mass confirming modifier

Subject Index

Dictionnaire d’apprentissage dufrançais des affaires(DAFA), 20, 23

Lexique de cooccurrents: Bourse etconjoncture économique,25

actantsemantic, 16

categoryconceptual, 20

co-hyponymy, 39collocation

specialized, 23concept, 14conceptual structure (of a domain),

14cross-reference, 20

definitionanalytical, 33

DiCo, 42dictionary

contextual, 18, 20specialized, 12–14, 16

domainconceptual structure of, 14specialized, 27

Explanatory CombinatorialDictionary (ECD), 12, 15,23

Explanatory CombinatorialLexicology (ECL), 12, 15,27, 40

Fr. ÉPARGNE, lexical entry for, 21Fr. FAVORABLE, lexical entry for,

19

hyperonymy, 37hyponymy, 37

labelsemantic, 20, 42

lexical function (LF), 16, 17, 35Part, 39Gener, 37, 40Mult, 39Phase, 39Real � , 40Sing, 39Spec, 37Syn � , 21, 39Syn � , 21Syn � , 21, 37Syn � , 37Syn, 39syntagmatic, 40

lexical unit (LU), 13predicative, 16, 33, 40specialized, 15, 28

MASS, lexical entry for, 18

patterncontextual, 19, 20

polysemy, 26, 29predicate

semantic, 17

sense

45

terminological, 27standard, linguistic, 13

term, 13, 14complex, 14family of, 21

terminology, 12, 13school of Vienna, 13

Vienna school (of terminology), 13

word, 13

Name Index

Binon, 20, 21, 25, 35

Cohen, 25

Dancette, 36Descamps, 19, 25, 34, 35, 40

Fontenelle, 39, 42Frawley, 16, 17, 25, 26, 32, 36Freibott, 23

Gentilhomme, 16Grimes, 37Gross, 42

Hausmann, 23Heid, 23

Kocourek, 12, 42

L’Homme, 11, 25, 36, 42Laporte, 25

Marcel, 16Mel’cuk, 11, 12, 17, 20, 23, 29, 31Milicevic, 42

Polguère, 42

Sager, 13

Tesnière, 42

Wanner, 17

47

Bibliography

Apresjan, Ju.D. 1992. “Brushstrokes for a Portrait”. Le mot, les mots, les bonmots ed. by André Clas. University of Montreal Press.

Binon, J., S. Verlinde, J Van Dyck & A. Bertels. 2000. Dictionnaired’apprentissage du français des affaires. Dictionnaire de compréhension et deproduction de la langue des affaires. Paris: Didier.

Cohen, B. 1986. Lexique de cooccurrents. Bourse–conjoncture économique.Montreal: Linguatech.

Dancette, J. & M.C. L’Homme. 2002. “The Gate to Knowledge in aMultilingual Specialized Dictionary: Using Lexical Functions for Taxonomicand Partitive Relations”. Proceedings of the Tenth EURALEX InternationalCongress ed. by A. Braasch & C. Povlsen. 597–605. Copenhagen: CST.

Descamps, J.L., M.A. Mochet, T. Lewin, B. Lamizet & D. Coste. 1992.Sémantique et concordances, suivi du Dictionnaire contextuel de françaispolitique. Paris: Klincksieck.

Descamps, J.L. 1973. “Présentationd’un dictionnaire contextuel de françaispour la géologie”. Langue française. 17:81–111.

Descamps, J.L. 1976. Dictionnaire contextuel de français pour la géologie:essai de classement d’un concordance de français scientifique et étudecritique. Paris: Didier.

Fontenelle, T. 1997. Turning a Bilingual Dictionary into a Lexical-SemanticDatabase. Tübingen: Niemeyer.

Frawley, William. 1988. “New Forms of Specialized Dictionaries.International Journal of Lexicography. 1-3:189–213.

Gentilhomme, Y. 1994. “L’éclatement du signifié dans les discourstechnoscientifiques”. Cahiers de lexicologie. 64.1:5–35.

Gentilhomme, Y. 1995. “Contribution à une réflection sur les locutionsmathématiques”. Cahiers de lexicologie. 66.1:5–37.

Grimes, J. 1990. “Inverse Llexical Functions”. Meaning-Text Theory:Linguistics, Lexicography and Implications ed. by J. Steele. 350–364. Ottawa:Ottawa University Press.

Gross, M. 1966. Grammaire transformationelle du français, syntaxe du verbe.Paris: Larousse.

Hausmann, F.J. 1979. “Un dicctionnaire des collocations est-il possible?”.Travaux de linguistique et de littérature. 17.1:187–195.

Heid, U. & G. Freibott. 1991. “Collocations dans une base de donnéesterminologique et lexicale”. Meta. 36.1:77–91.

Kocourek, R. 1991. La langue française de la technique et de la science. Versune linguistique de la langue savante. Wiesbaden: Oscar Brandstetter.

Laporte, I. & M.C. L’Homme. 1997. “Recensement et consignation descombinaisons lexicales spécialisées: exemple d’application dans le domainede la pharmacologie cardiovasculaire”. Termonologies nouvelles. 16:95–101.

L’Homme, M.C. 1998. “Définition du statut du verbe en langue de spécialitéet sa description lexicographique”. Cahiers de lexicologie. 73.2:61–84.

Marcel, É. 2000. “Quelques modalités d’application de la fonction lexicalegener en botanique”. Cahiers de lexicologie. 76.1:91–110.

Mel’cuk, I.A. & L. Wanner. 1996. Lexical Functions and Lexical Inheritancefor Emotion Lexemes in German. Lexical Functions in Lexicography andNatural Language Processing ed. by L. Wanner. 209–278. Amsterdam:Benjamins Academic Publishers.

Mel’cuk, Igor A., N. Arbatchewsky-Jumarie, L. Elnitsky & A. Lessard. 1984.Dictionnaire explicatif et combinatoire du français contemporain. Volume 1.Montréal, Canada: Presses de l’Université de Montréal.

Mel’cuk, Igor A., N. Arbatchewsky-Jumarie, L. Iordanskaja, S. Mantha &A. Polguère. 1984. Dictionnaire explicatif et combinatoire du françaiscontemporain. Volume 4. Montréal, Canada: Presses de l’Université deMontréal.

Mel’cuk, I., Clas A. & Polguère A. 1995. Introduction à la lexicologieexplicative et combinatoire. Louvain la Neuve: Duculot.

Mel’cuk et al., I.A. 1984, 1988, 1992, 1999. Dictionnaire explicatif etcombinatoire du français contemporain. Montréal, Canada: Presses del’Université de Montréal. Volumes I-IV.

Mel’cuk, I. 1988. Dependency Syntax: Theory and Practice. Albany: StateUniversity of New York Press.

Mel’cuk, I. 1997. Vers une linguistique Sens-Texte. Leçon inaugurale. Paris:College de France.

Mel’cuk, I. 1998. “Collocations and Lexical Functions”. Phraseology. Theory,Analysis and Applications ed. by A. P. Cowie. 23–53. Oxford: OxfordUniversity Press.

Milicevic, J. 1997. Étiquettes sémantiques dans un dictionnaire formalisé dutype Dictionnaire explicatif et combinatoire, Mémoire de maîtrise. Montréal:Département de linguistique et de traduction, Université de Montréal.

Polguère, A. 1998. “pour un model stratifié de la lexicalisation en générationde texte”. T.A.L.. 39.2:57–76.

Sager, I. 1990. A Practical Course in Terminology Processing. Amsterdam &Philadelphia: Benjamins Academic Publishers.

Tesniere, L. 1966. Élélements de syntaxe structurale. Paris: Klincksieck.

Wanner, L (ed.). 1996. Lexical Functions in Lexicography and NaturalLanguage Processing. Amsterdam & Philadelphia: Benjamins AcademicPublishers.