Section: New Results

Towards a morpho-semantic resource for French designed for Word Sense Disambiguation

Participant : Lucie Barque.

The most promising WSD methods are those relying on external knowledge resources [78] but semantic resources for French are scarse. Moreover, existing resources offer fine grained sense dictinctions that do not fit to WSD. Our aim is to provide the NLP community with a broad-coverage morpho-semantic lexicon for French that relies on coarse-grained sense distinctions for polysemic units. Preliminary results concern nouns, on which we have first focused because their semantic description, compared to verbs, crucially lacks (for information retrieval, for instance) and because the regular polysemy phenomenon (recurring cases of polysemy within semantic classes) mainly occurs in nominal semantic classes:

• We proposed a linguistically motivated description of general semantic labels for nouns, that will allow for coarse-grained sense distinctions [107]

• Regular polysemy of nouns that can denote an event or a participant of this event has also been described for a large number of French nouns in [46]

• From a morphological point of view, nouns denoting events in French are mostly deverbal nouns (eg. conversation 'conversation', promenade 'stroll'), but there are also underived event nouns (eg. guerre 'war', séisme 'earthquake'). We compared their semantic properties in [35] .

• Some lexical meanings are not easily captured by ontological semantic classes and a closer look has to be taken at them. Relational meanings in relational nouns are one of them [15] .