A RULE-BASED APPROACH TO UNDERSTAND QUESTIONS IN ARABIC QUESTION ANSWERING

摘要

Research on Arabic Natural Language Processing (NLP) is facing a lot of problems due to languagecomplexity, lack of machine readable resources and lack of interest among Arab researchers. One of thefields that research has started to appear in is the field of Question Answering. Although some research hasbeen done in this area, few have proved to be effective in producing exact relevant answers. One of the issuesthat affected the accuracy of producing correct answers is proper tagging of entities and proper analysis of auser’s question. In this research, a set of 60+ tagging rules, 15+ Question Analysis rules and 20+ QuestionPatterns were built to enhance the answer generation of Natural Language Questions posed over somecorpora collected from different sources. A QA system was built and experiments showed good results with anaccuracy of 78%, a recall of 97% and an F-Measure of 87%.