Using semantics for paragraph selection in question answering systems

Ejiciency of term-based Question Answering systems is limited to answering questions whose answer is expressed in documents by using mainly the same terms appearing in questions. The system presented in this paper overcomes this fact by performing open domain Question Answering (QA) from a semantic perspective. For this purpose, we define a general semantic model that represents the concepts referenced into the questions as well as a relevance measure that allows locating and ranking fragments of documents fiom whose content is possible to infer the answer to specific questions. mth the purpose of evaluation, this model has been embedded into a full QA system. Comparison of performance between our model and term-based approaches shows that QA measures improve signiJicantly when this model is applied to paragraph selection process.