FrameNet-based fact-seeking answer processing: a study of semantic alignment techniques and lexical coverage

In this paper, we consider two aspects which affect the performance of factoid FrameNet-based Question Answering (QA): i) the frame semantic-based answer processing technique based on frame semantic alignment between questions and passages to identify answer candidates and score them, and ii) the lexical coverage of FrameNet over the predicates which represent the main actions in question and passage events. These are studied using a frame semantic-based QA run over the TREC 2004 and TREC 2006 factoid question sets.