Chunking-based Question Type Identification for Multi-Sentence Queries

This paper describes a technique of question type identification for multi-sentence queries in open domain question-answering. Based on observations of queries in real question-answering services on the Web, we propose a method to decompose a multi-sentence query into question items and to identify their question types. The proposed method is an efficient sentence-chunking based technique by using a machine learning method, namely Conditional Random Fields. Our method can handle a multi-sentence query comprising multiple question items, as well as traditional single sentence queries in the same framework. Based on the evaluation results, we discuss possible enhancement to improve the accuracy and robustness.