Enriched syntax-based meaning representation for answer extraction

In Question Answering a major challenge is the fact that similar meaning is very often expressed with different surface realizations in questions and in sentences containing the answer. In this paper we propose an enriched syntax-based representation which helps deal with this widespread variability and provides a degree of generalization. We encode uncertainty about the syntactic structure of the question by using multiple alternative dependency parse trees. We then augment the question meaning representation by including multiple paraphrases of each dependency path, derived from distributional analysis of a large corpus.