中英文的文字蘊涵與閱讀測驗的初步探索

An Exploration of Textual Entailment and Reading Comprehension for Chinese and English

文字蘊涵是研究文字敘述之間的邏輯關係的工作，本文利用詞彙、語法、詞彙語意的相關語文資訊，建構經驗法則公式與機器學習模型，檢驗自動推測文字蘊涵關係的效果。本文所報告的效果在NTCIR-10的RITE競賽中的簡體與繁體中文的文字蘊涵都有相當好的表現。我們同時延伸文字蘊涵的推論技術，企圖以語文處理技術自動回答國小及國中的中英文閱讀測驗試題，這一部份的工作仍在發展之中，對於比較簡易的四選一的試題，如果相關的基礎技術成熟，可以達到超過五成的答對率。

Research on text entailment studies the logical relationships between statements. We employed linguistic information at the lexical, syntactic, and semantic levels to build heuristics and machine-learning based models for algorithmic judgment of text entailment relationships. Methods proposed in this paper achieved relatively very good performances in the RITE task for both traditional and simplified Chinese entailment problems in NTCIR-10. We extended our work and attempted to automatically answer questions in reading comprehension tests in Chinese and English used in elementary and middle schools. To make the automatic answering more feasible, we manually selected statements which were relevant to the test items before we ran the text entailment component. Experimental results indicated that it was then possible to find the answers better than 50% of the time for one out of four multiple-choice items.