Automatic Acquisition of GL Resources, Using an Explanatory, Symbolic Technique

The Generative Lexicon (GL) theory (Pustejovsky 1995) has proved its usefulness in the analysis of numerous linguistic phenomena across languages. Moreover, elements from Generative Lexicons have been shown to be relevant in several natural language processing (NLP) applications (e.g. information retrieval, etc.). For instance, the qualia structure gives access to relational information, crucial for such applications. In particular, the qualia roles (namely the telic, agentive, constitutive and formal roles) express, in terms of predicative formulae, the basic features of the semantics of nouns. In a GL model, the noun is linked not only to other nouns via traditional lexical relations (such as meronymy and hyperonymy) but also to verbs. For example, the noun book is linked to the verbal predicate read via its telic role and to the predicate write via its agentive role. Hereafter, a noun(N)verb(V) pair in which V expresses one of the qualia roles of N (like book-read or book-write) is called a qualia pair. Previous work by Fabre and Sébillot (1999) has demonstrated that these N-V relations provide lexical resources that are found to be useful for information retrieval systems. Different studies (Grefenstette 1997; Pustejovsky et al. 1997, inter alia) also show that N-V pairs can feed indexes that help a user to select the most interesting occurrences of a given noun in a text. Moreover, a short survey (Vandenbroucke 2000) at the documentation center of the Banque Bruxelles Lambert (Brussels) shows that verbs that express a qualia relation seem to be more relevant than others for a document retrieval task; indeed, in this