Rule-based focus extraction in Turkish question answering systems

This study describes a rule-based approach that is employed in the question analysis module of the Turkish question answering system we design for high-school students to assist their education. The technique is used to extract parts of the question (the focus) that indicate the type and the character of the answer. Evaluation is performed on a set of geography question data that are collected and annontated by human experts, and the results are provided. Manually tailored rules are based on general Turkish question patterns, thereby not dependent on the domain. Therefore, this technique can be used by any Turkish question answering system, regardless of the domain. Both the source codes and the annontated data set are electronically provided for both reproducability and future work.