Using Verb Dependency Matching in a Reading Comprehension System

In this paper, we describe a reading comprehension system. This system can return a sentence in a given document as the answer to a given question. This system applies bag-of-words matching approach as the baseline and combines three technologies to improve the result. These technologies include named entity filtering, pronoun resolution and verb dependency matching. By applying these technologies, our system achieved 40% HumSent accuracy on the Remedia test set. Specifically, verb dependencies applied in our system were not used in previous reading comprehension systems. In addition, we have developed a new bilingual corpus (in English and Chinese) - the ChungHwa corpus. The best result is 68% and 69% HumSent accuracy when the system is evaluated on the ChungHwa English and Chinese corpora respectively.

• Type: Conference
• Year: 2004
• Where: AIRS
• Authors: Kui Xu, Helen Meng
• Keywords: AIRS 2004, Bag-of-words Matching Approach, HumSent Accuracy, Information Retrieval, Reading Comprehension Systems