A Definitional Metalanguage for Explanatory Combinatorial Lexicography

How are lexical senses described? One can identify two main families of sense descriptions on the basis of their relation to word meaning: word paraphrases vs. word interpretations. The first family can itself be subdivided into three main subfamilies, according to the formal nature of sense descriptions. 1. Word paraphrases 1.1. " Standard " lexicographic definitions are, roughly speaking, linguistic paraphrases of individual word senses. Thanks to their relatively controlled structure (use of a " defining vocabulary "), the LDOCE's definitions (Longman, 2003) may claim to be one of the most representative example of this approach to sense description. 1.2. Analytical definitions are more formal descriptions of senses, to be used in the scientific study of language semantics. Such definitions (like the previous ones) find their origin in Aristotle's theory of concept definition and are used in " theoretical " dictionaries such as (Wierzbicka, 1987) or (Mel' č uk et al. 1.3. Logical definitions are, so to speak, translation of word senses into logical formulas, usually based on some form of predicate calculus (Dowty, 1979), or graphical encoding of logical configurations (conceptual graph à la Sowa, etc.). Feature-value matrices are formal descriptions tailored for unification, rather paraphrase processes. They can be based on simple sets of binary features (Pottier, 1974) or be based on richer, logic-based feature systems (Pustejovsky, 1998). We will present an approach to sense description that is somehow intermediary between analytical definitions and feature-value matrices (1.2 and 2 above). It has been developed in the process of translating the Explanatory Combinatorial Dictionary (ECD) definitions into a semantic database of French lexical meanings. The defining language we propose can function as some sort of " pseudo-code " for both reasoning on word senses and formally encoding their description in a machine-readable format. This new approach to sense description is applied to the construction of the BDéf database (Altman and Polguère, 2003) and is presently used for two types of research: 1) exploration of how lexical function relations (semantic derivations and collocational links) can be hooked up to semantic components of definitions, and 2) identification and modeling of regular patterns of polysemy. In order to illustrate our approach in this short abstract, we will look at two senses of the French vocable MAISON (Eng. HOUSE or HOME). There is no entry for MAISON in the published ECDs. However, MAISON is accounted for in the DiCo database of French collocations …