Introduction

In this chapter, we survey a number of areas of lexical semantics which are often refered to and used with some adaptations in NLP applications. One of the aims of this section is to provide the reader with a background of linguistic issues in lexical semantics so that he/she can have a better view of what the main definitions, the hypothesis, the foundations and the limitations of each area are. These linguistic issues are often not used directly in NLP systems and in the description of lexical resources, they are usually adapted in various degrees to make the descriptions e.g. more comprehensive, and better suited for the problem or the domain addressed.

The elements presented here are basic issues in lexical semantics and have in most cases an inter-theoretical perspective. They have been judged to be sufficiently stable to be presented here. They can therefore be used in a number of frameworks; they can also be integrated or combined with a large number of syntactic systems. A choice criterion has been that lexical semantic descriptions should be as easy to use as possible and non-ambiguous. They should also be as language independent as possible and easy to encode in a lexicon. It should be noted that we do not claim to be comprehensive, and the absence of a framework is certainly not an a priori judgement of its value. Lexical semantics is indeed a vast area with many ramifications, impossible to cover in such a document.

The areas surveyed here are the following:

• Aspect in the lexicon, the aim is to define a common notation for actionality/Akstionart and Aspect. The concrete use in lexicons of aspect is to mark verbs, in syntactic contexts, in order to identify the nature of the action: e.g. perfective/imperfective,
• Lexicalization, the aim is to identify and to organize lexicalization patterns across languages. Uses are numerous, e.g.: the study of the cooperation syntax-semantics, and its use in generation of NL, the investigation of mismatches between 2 languages, and the taking into account of lexicalization preferences. Questions addressed are: What is a lexicalization pattern, what are the meaning components at stake, what are the combinatorial elements, and what is used in projects and how can they be normalized ?

• Semantic roles (or thematic roles), the aim is to classify the arguments of predicates into a small and closed set of types of `participants' modelling the role played by the argument w.r.t. the predicate. This approach provides a level of semantic representation. Semantic roles can also be viewed as a bridge between syntax and semantics.

• Verb semantic classes and the semantics of verbs, the aims are to identify and to compare different classifications methods for verbs, to outline their properties, advantages and limitations, and to isolate semantic universals. This study contributes to the identification of senses and sense extensions. Uses are quite numerous, e.g.: a better organization of semantic lexicons where verb classes share a number of semantic properties, and a contribution to lexical acquisition. The contents are: the verb syntactic alternation systems, classifications based on semantic roles and Lexical Conceptual Structure-based classifications.

• The semantics of nouns and compound nouns, the aims are the same as for verbs; in addition, the following points are studied: nouns and nominalizations, semantic properties: WordNet classes, quantification, semantic alternations. Different forms of representations (features, networks) are presented. An analysis and a specification of noun compounds, relational nouns and deverbal nouns is also introduced

• Adjectives, the aim is to classify different types of adjectives, and to normalize syntactic alternations involving adjectives. A second aim is to identify generic properties modified by adjectives and to normalize them.

• Prepositions, the aims are to classify prepositions w.r.t. uses, to identify general semantic classes, prepositions being very polysemic. Another goal is to relate prepositions to semantic roles, and to selectional restrictions. The main use considered here is the introduction of generic classifications in verb subcategorization frames.

For each area, a description of the main approaches is first given, followed by comparisons aimed at identifying basic notions and components. Then relations to other areas of lexical semantics are emphasized. Finally, as a form of pointer towards the two next chapters, indications are given on how lexical semantics data is encoded in lexical knowledge bases and used in large-size NL applications 4.

Lexical semantics being a very vast domain, it has not been possible to include every approach. Among the missing frameworks that we foresee to add in the future are the Mel'cuk functional system (mentioned in passing in some paragraphs), the Script/Frame systems, frame nets and recent developments in ontology description.