Question Answering System for Entrance Exams in QA4MRE.
Xinjian Li, Ran Tian, Ngan Luu-Thuy Nguyen, Yusuke Miyao, Akiko Aizawa · CLEF (Working Notes) · 2013
This paper describes our question answering system for Entrance Exams, which is a pilot task of the Question Answering for Machine Reading Evaluation at Conference and Labs of the Evaluation Forum (CLEF) 2013. We conducted experiments in which participants were provided with documents and multiple-choice questions. Their goals was to select one answer or leave it unanswered for each question. In our system, we developed a component to detect all story characters in the documents and tag all personal pronouns using coreference resolution. For each question, we extracted related sentences and combined them with candidate answers to create inputs for a Recognizing Textual Entailment (RTE) component. The answers were then selected based on the confidence scores from the Recognizing Textual Entailment component. We submitted five runs in the task and the run that ranked highest obtained a c@1 score of 0.35, which outperformed the baseline c@1 score of 0.25.