Semantic retrieval approach to factoid question answering for Bulgarian

Abstract

With the abundance of information available today, we need efficient tools to explore it. Search engines attempt to retrieve the most relevant documents for a given query, but still require users to look for the exact answer. Question Answering (Q&A) systems go one step further by trying to answer users' questions posed in natural language. In this paper we describe a semantic approach to Q&A retrieval for Bulgarian language. We investigate how the usage of named entity recognition, question answer type detection and dependency parsing can improve the retrieval of answer-bearing structures compared to the bag-of-words model. Moreover, we evaluate nine different dependency parsing algorithms for Bulgarian, and a named entity recognizer trained with data automatically extracted from Wikipedia.