應用跳脫語言模型於同義詞取代之研究

Skip N-gram modeling for Near-Synonym choice

陳士婷(Shih-Ting Chen)

指導教授 ： 禹良治

元智大學/資訊學院/資訊管理學系/碩士(2012年)

https://doi.org/10.6838/YZU.2012.00226

摘要

同義詞(Near-Synonym)不只在自然語言應用中是重要的一環，也是對第二語言學習者很重要的部分。同義詞雖然是一群意思相近的單字集合，但在特定的情況與特殊用法下，選擇錯誤的同義詞會造成句意上的誤解，甚至是整個文法錯誤，因此我們希望能夠藉由上下文的訊息，再利用系統分辨出正確的同義詞，協助外語學習者做有效率的學習。
目前為止已有許多同義詞的相關研究，這些研究的方法包含：點式交互資訊(Pointwise Mutual Information, PMI)與N連詞(N-gram)模型都是常用的方法，我們想使用與以往不同的方法來提升正確率，因此我們使用跳脫語言模型(Skip N-gram)的方法參與SemEval-2007同義詞任務，結果顯示我們提出的方法是可行的，正確率也有明顯的提升。

關鍵字

同義詞 ； 點式交互資訊 ； N連詞 ； 跳脫語言模型

並列摘要

Near-synonym is not only an important thing in natural language applications, and also very important for the second language learner. Although, near-synonym represent a groups of words with similar meaning. But, in specific case and specific usage, we choice the wrong near-synonym may cause wrong meaning, even cause grammatical errors. Therefore, we hope system can use contextual information to differentiate near-synonym.

So far, there are many studies about near-synonym, the methods of these studies include: PMI and N-gram modeling. We want to use different method to improve the accuracy, so we use Skip N-gram modeling for near-synonym choice in SemEval-2007 task, the results show that our proposed method is feasible and the accuracy have improved significantly

並列關鍵字

Near-synonym ； PMI ； N-gram ； Skip N-gram

參考文獻

1.D. Moldovan and R. Mihalcea, "Using WordNet and Lexical Operators to Improve Internet Searches," IEEE Internet Computing, pp. 34-43, 2000.

2.J. Bhogal, A. Macfarlane, and P. Smith, "A Review of Ontology based Query Expansion," Information Processing & Management, pp. 866-886, 2007.

4.S. Ouyang, H. H. Gao, and S. N. Koh, "Developing a Computer-Facilitated Tool for Acquiring Near-Synonyms in Chinese and English," In Proc. of IWCS-09, pp. 316-319, 2009.

6.D. Inkpen, "A Statistical Model of Near Synonym Choice," ACM Trans. Speech and Language Processing, pp. 1-17, 2007.

7.L. C. Yu, C. H. Wu, R. Y. Chang, C. H. Liu, and E. H. Hovy, "Annotation and verification of sense pools in OntoNotes," Information Processing&Management 46(4), pp.436-447, 2010.