Meta-knowledge Annotation for Efficient Natural-Language Question-Answering

Abstract

A recent trend in the exploitation of unstructured text content is the use of natural language question answering (NLQA) systems. NLQA is an elaboration of traditional information retrieval techniques for satisfying a user’s information needs, where the goal is not simply to retrieve relevant documents but to additionally extract specific passages and semantic entities from these documents as candidate answers to a natural language question. NLQA is thus a tight integration of natural language processing (NLP), information retrieval (IR) and information extraction (IE) designed to circumvent the deep and brittle analysis of questions in favor of shallow but robust comprehension, to ultimately achieve a broad domain question-answering competence. It is argued here that the key to achieving good quality answers in a high-throughput setting lies in a system’s ability to construct rich queries that incorporate knowledge from multiple sources.