Section: New Results

Natural Language Processing

In [7] and [3] , we develop a new algorithm for drastically improving a pairwise coreference classification system. Specifically, this algorithm works by learning the best partition over mention type pairs by training different pairwise coreference models for each pair type. In effect, our algorithm finds the optimal feature space (from a base feature set and set of types) for separating coreferential mention pairs, but it remains tractable by exploiting the structure of the hierarchies built from the pair types. In [6] , we propose a new approach for the automatic identification of so-called implicit discourse relations. Our system combines hand labeled examples and automatically annotated examples (based on explicit relations) using different methods inspired by work on domain adapation. Our system is evaluated empirically and yields important performance gains compared to only using hand-labeled data. This paper has received the best paper award at the TALN 2013 conference, the national NLP conference.