Textual Entailment Recognition for Chinese and English

• Main Authors: Huang, Wei Jie, 黃瑋杰
• Other Authors: Liu, Chao Lin
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/54110700002815558996

碩士 === 國立政治大學 === 資訊科學學系 === 101 === Recognizing Inference in Text (RITE) has become a serious issue in several research areas, such as Information Retrieval (IR), Information Extraction (IE), Automatic Summarization, or Intelligent Tutoring Systems (ITS). The research topic is getting more important since the First Recognizing Textual Entailment Challenge (RTE-1) was held in 2005. For Asian languages, Recognizing Inference in Text (RITE-1) provides evaluation standards on recognizing entailment systems. In this research, we built a system based on textual analysis and construct several heuristic functions to compute entailment in text. Besides, we proposed a method to measure the similarity between two Chinese words based on E-HowNet and used it to enhance the system’s performance. Moreover, machine learning techniques, such as SVM, J48 and Linear Regression are used to train classification models. We extracted features based on heuristic functions and other syntactic features. The experimental results indicated that our systems achieved great performances and received second places in NTCIR-10 RITE-2. The analysis of machine learning approaches also showed Chinese and English shared different linguistic characteristics and effective features on recognizing textual entailments. Besides, the experimental results of reading comprehensions showed that we can develop intelligent tutoring system based on this research. The intelligent tutoring system is able to enhance students the ability of reading understandings and help on generating quality reading tests.