Section: New Results

Automatic meronymy discovery

Participants : Emmanuel Lassalle, Pascal Denis.

Bridging descriptions are a special kind of anaphora whose interpretation requires not only identifying an antecedent, but also inferring a specific relation linking it to the anaphor. The resolution of bridging anaphora represents a very challenging task in discourse processing. It is considerably much harder than standard coreferential anaphora resolution for which shallow predictors (like distance, string matching, or morphosyntactic agreement) have been shown to be rather effective. Part of the challenge is due to an important information bottleneck. Lexical resources like WordNet are still too poor and uneven in coverage to provide a realistic solution. In turn, more recent approaches to bridging resolution have turned to web-based extraction methods. To date, the most complete and best-performing approach combines focus and lexical distance predictors using machine learning techniques [105] .

We have focused on mereological bridging anaphora (that is, cases wherein the inferred relation is a part-whole relation).(An illustrative English example is is the following discourse: The car will not move. The engine is broken.) Moreover, we have worked on French, a language for which current lexical resources have a very low coverage. The system, presented in [32] is similar to a system developed for English [105] , but it was enriched to integrate meronymic information extracted automatically from both web queries and raw text using syntactic patterns. Through various experiments on the DEDE corpus [78] , we show that although still mediocre the performance of our system compare favorably to those obtained for English by the above-mentioned system. In addition, our evaluation indicates that the different meronym extraction methods have a cumulative effect, but that the text pattern-based extraction method is more robust and leads to higher accuracy than the Web-based approach.