Relation Extraction for Open and Closed Domain Question Answering

Abstract

One of the most accurate methods in Question Answering uses off-line information extraction to find answers for frequently asked questions. It requires automatic extraction from text of all relation instances for relations that users frequently ask for. In this chapter, we present two methods for learning relation instances for relations relevant in a closed and open domain (medical) question answering system. Both methods try to learn automatically dependency paths that typically connect two arguments of a given relation. The first (lightly supervised) method starts from a seed list of argument instances, and extracts dependency paths from all sentences in which a seed pair occurs. This method works well for large text collections and for seeds which are easily identified, such as named entities, and is well-suited for open domain question answering. In a second experiment, we concentrate on medical relation extraction for the question answering module of the IMIX system. The IMIX corpus is relatively small and relation instances may contain complex noun phrases that do not occur frequently in the exact same form in the corpus. In this case, learning from annotated data is necessary. We show that dependency patterns enriched with semantic concept labels give accurate results for relations that are relevant for a medical question answering system. Both methods improve the performance of the Dutch question answering system Joost.