Mining the semantic information to facilitate reading comprehension

Abstract

The reading comprehension (RC) task- accepting arbitrary text input (a story) and answering questions about it. The RC system needs to draw upon external knowledge sources to achieve deep analysis of passage sentences for answer sentence extraction. This paper proposes an approach towards RC that attempts to utilize semantic information to improve performance beyond the baseline set by the bag-of-words (BOW) approach. Our approach emphasizes matching of linguistic features (i.e. verbs, named entities and base noun phrases) and semantic extending for answer sentence extraction. The approach gave improved RC performance in the Remedia corpus, attaining HumSent accuracies of 41.3%. In particular, performance analysis shows that a relative performance of 19.7% is due to the application of linguistic feature matching and a further 10.3% is due to the semantic extending. © Springer-Verlag Berlin Heidelberg 2007.