From the corpus to the lexicon: the example of data models for verb subcategorization.

This paper describes the integration of corpus-based syntactic subcategorization frames and correlated semantic<br />information into a large-scale, cross-theoretically informed lexical database for French<br />(Romary et al. (2004)). This database is the first to implement the Lexical Markup Framework (LMF), an international<br />initiative towards ISO standards for lexical databases (ISO TC 37/SC 4). The subcategorization<br />frames have been acquired via a dependency-based parser (Bick (2003)), whose verb lexicon is currently incomplete<br />with respect to subcategorization frames. Therefore, we have implemented probabilistic filtering as<br />a post-parsing treatment using the binomial distribution. Building on our discussion of what semantic information,<br />e.g., participant roles, to include in the database, we describe how we plan to exploit our findings on<br />subcategorization frames to derive this information via unsupervised learning techniques