EXTRANS, AN ANSWER EXTRACTION SYSTEM

Abstract

Answer Extraction (AE) systems retrieve phrases in textual documents that directly answer natural language questions. AE over technical manuals requires very high recall and precision, and yet small text units must be retrieved. It is therefore important to perform linguistic analysis in detail. We present ExtrAns, an AE system over Unix manuals that uses full parsing, partial disambiguation, and anaphora resolution to generate the minimal logical forms of the documents and the query. The search procedure uses a proof algorithm of the user query over the Horn clause representation of the minimal logical forms. Remaining ambiguities in the retrieved sentences are dealt with by graded highlighting.