ILQUA at TREC 2006

This year, we made changes to the passage/sentence retrieval component of ILQUA in handling factoid and list questions. All the other components remain same. ILQUA is an IE-driven QA system. To answer “Factoid” and “List” questions, we apply our answer extraction methods on NE-tagged passages or sentences. The answer extraction methods adopted here are surface text pattern matching, n-gram proximity search, and syntactic dependency matching. Although surface text pattern matching has been applied in some previous TREC QA systems, the patterns used in ILQUA are better since they are automatically generated by a supervised learning system and represented in a format of regular expressions which contain multiple question terms. In addition to surface pattern matching, we also adopt n-gram proximity search and syntactic dependency matching. N-grams of question terms are matched around every named entity in the candidate sentences or passages and a list of named entities are generated as answer candidate. These named entities then go through a multi-level syntactic dependency matching component until a final answer is generated. To answer “Other” questions, we parsed the answer sentences of “Other” questions in previous main task and built syntactic patterns combined with semantic features. These patterns are later applied to the parsed candidate sentences to extract answers of “Other” questions. Figure 1 shows the diagram of the ILQUA architecture.