<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-1801">
  <Title>A Lexico-semantic Approach to the Structuring of Terminology</Title>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
4 Implications for computational
</SectionTitle>
    <Paragraph position="0"> terminology and other corpus-based work The previous discussion has a number of implications for computational terminology (as well as other corpus-based terminology-related applications). I will examine a few in this section. First, both approaches will focus on different types of units when selecting terms in corpora. In conceptual approaches, a selection is made among linguistic units that can refer to a concept. The focus is on nouns and noun phrases. Even though concepts can be expressed in a variety of linguistic forms, synonyms considered will invariably be nouns of noun phrases. Work on terminological CompuTerm 2004 - 3rd International Workshop on Computational Terminology 11 variation (Daille, 2003; Jacquemin, 2001) has shown the variety of forms that terms can take in corpora (morphological derivation, insertion, elision, anaphora, etc.), but these are taken into account only if they can be associated with an admitted term.</Paragraph>
    <Paragraph position="1"> In a lexico-semantic approach as that presented in section 2.3, units considered will be those that convey a meaning that can be related to the field of computing (the subject field is delimited prior to the selection). Lexical units selected can pertain to different parts of speech as long as their meaning can be related to the field under examination: nouns (program, byte); verbs (debug, to program), adjectives (user-friendly, programmable). Even adverbs can convey a specialized meaning (e.g., digitally, dynamically).</Paragraph>
    <Paragraph position="2"> Secondly, any terminological work based on corpora will run into polysemy, even though it focuses on a small set of terms. The manner in which the distinctions between senses are made has important consequences on way terms will be processed afterwards.</Paragraph>
    <Paragraph position="3"> Polysemy can be dealt with using a conceptual approach, which considers this property to be an accidental problem. Hence, distinctions depend on decisions made during the classification process or the construction of conceptual representations.</Paragraph>
    <Paragraph position="4"> In lexico-semantic approaches, polysemy is viewed as a natural property of lexical units.</Paragraph>
    <Paragraph position="5"> Senses are delimited prior to the representation of semantic relationships and this delimitation is based on the observations of interactions between the term under examination and other lexical units.</Paragraph>
    <Paragraph position="6"> Sense delimitation and distinction is a necessary step before anything else can be done.</Paragraph>
    <Paragraph position="7"> Thirdly, regarding terminology structuring, conceptual methods, such as the one discussed in section 2.1, are useful as far as classification is concerned. Hence, they can be used for describing concepts that correspond to entities (concrete objects, substances, artefacts, animates, etc.).</Paragraph>
    <Paragraph position="8"> Moreover, the focus is on hierarchical relations (hyperonymy and meronymy) which is again valid for entities, and, as far as part-whole relations are concerned, more specifically concrete objects.</Paragraph>
    <Paragraph position="9"> Many non-hierarchical relationships, such as those listed in section 2.3 are disregarded, either because they involve units that do not refer to entities, or because they are relationships between lexical units and not concepts.</Paragraph>
    <Paragraph position="10"> Also, relationships between synonyms are considered from the point of view of true synonymy. Choosing a unique linguistic identifier for a concept and considering competing linguistic forms as true synonyms has implications for the variety of relationships that can be considered.</Paragraph>
    <Paragraph position="11"> Some relationships can be valid for one synonym but not for another.</Paragraph>
    <Paragraph position="12"> In lexico-semantic approaches, semantic relationships are attached to senses that have been distinguished previously. In addition, a wide variety of semantic relationships can be taken into account. These relationships can apply to terms that designate entities, as well as activities, and properties. Hypernymy and meronymy represent only a small part of the semantic relationships terms can share with other terms. Other relationships, such as argument relations, entityactivity relations, can be expressed by different parts of speech.</Paragraph>
    <Paragraph position="13"> Fourthly, conceptual approaches lead to representations that distance themselves from data collected in corpora. Many decisions are made during the construction of the representation. On the one hand, many meanings that would appear to be relevant in other approaches are not considered. On the other hand, things are added in order to build the representation. Consider, for example, Figure 1. Some subdivisions are created but do not correspond to lexical units (e.g., according to the interface); this sort of classification of units will result in considering several complex sequences that have a compositional meaning (hence, that are not true lexical units).</Paragraph>
    <Paragraph position="14"> Terminology structuring in conceptual approaches is often carried out in order to represent knowledge and not linguistic units. Problems arise when this work is done using corpora as a starting point, since linguistic units (such as terms) do not behave in a way that reflects perfectly a given knowledge structure. When analyzing terms, considerations regarding knowledge structure will constantly interfere with factors related to the behaviour of linguistic units in text.</Paragraph>
    <Paragraph position="15"> On the other hand, lexico-semantic approaches are much more compatible with data gathered from corpora. Of course, terminologists will make decisions since they must interpret data and synthesize their findings, but these are based on the observation of interactions between lexical units that appear in corpora.</Paragraph>
  </Section>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
5 Concluding remarks
</SectionTitle>
    <Paragraph position="0"> The point in my discussion, is not to say that an approach is much better than the other for terminology, regardless of the application at hand.</Paragraph>
    <Paragraph position="1"> This topic has been dealt with extensively by authors and even placed in a theoretical CompuTerm 2004 - 3rd International Workshop on Computational Terminology12 perspective. Rather, I wanted to demonstrate that an approach is probably better suited that the other as far as terms considered in corpora are concerned. I also wanted to point out the methodological consequences of choosing an approach over another.</Paragraph>
    <Paragraph position="2"> Conceptual approaches will account for consensual representations of knowledge, based on a predefined set of hierarchical relationships.</Paragraph>
    <Paragraph position="3"> However, in must be kept in mind that resulting representations distance themselves from corpus data and necessitate a lot of hand-crafted changes. Often, the ideal knowledge structure is formulated beforehand entirely or partly, and the difficulty consists in trying to find lexical units that fit into it. Lexico-semantic approaches will provide terminologists with a framework for interpreting data related to terms and the contexts in which they appear. However, one must accept, when using this kind of approach, that terminological structures are discovered gradually through semantic relations and that some of these relations will even contradict assumed knowledge structures.</Paragraph>
  </Section>
  <Section position="5" start_page="0" end_page="0" type="metho">
    <SectionTitle>
6 Acknowledgements
</SectionTitle>
    <Paragraph position="0"> I would like to thank Elizabeth Marshman for her comments on a preliminary version of this article.</Paragraph>
  </Section>
class="xml-element"></Paper>