Acquiring Synonyms from Monolingual Comparable Texts

Acquiring Synonyms from Monolingual Comparable Texts Mitsuo Shimohata1 and Eiichiro Sumita2 1 Oki Electric Industry Co., Ltd., 2-5-7, Honmachi, Chuo-ku, Osaka City, Japan shimohata363@oki.com 2 ATR Spoken Language Translation Research Laboratories, 2-2-2 Hikaridai, Keihanna Science City, Kyoto, Japan eiichiro.sumita@atr.jp Abstract. This paper presents a method for acquiring synonyms from monolingual comparable text (MCT). MCT denotes a set of monolin- gual texts whose contents are similar and can be obtained automatically. Our acquisition method takes advantage of a characteristic of MCT that included words and their relations are confined. Our method uses con- textual information of surrounding one word on each side of the target words. To improve acquisition precision, prevention of outside appear- ance is used. This method has advantages in that it requires only part-of- speech information and it can acquire infrequent synonyms. We evaluated our method with two kinds of news article data: sentence-aligned par- allel texts and document-aligned comparable texts. When applying the former data, our method acquires synonym pairs with 70.0% precision. Re-evaluation of incorrect word pairs with source texts indicates that the method captures the appropriate parts of source texts with 89.5% precision. When applying the latter data, acquisition precision reaches 76.0% in English and 76.3% in Japanese. 1 Introduction There is a great number of synonyms, which denote a set of words sharing the same meaning, in any natural language. This variety among synonyms causes difficulty in natural language processing applications, such as information re- trieval and automatic summarization, because it reduces the coverage of lexical knowledge. Although many manually constructed synonym resources, such as WordNet [4] and Roget’s Thesaurus [12], are available, it is widely recognized that these knowledge resources provide only a small coverage of technical terms and cannot keep up with newly coined words. We propose a method to acquire synonyms from monolingual comparable text (MCT). MCT denotes sets of different texts1 that share similar contents. MCT are appropriate for synonym acquisition because they share not only many 1 In this paper, “text” can denote various text chunks, such as documents, articles, and sentences. R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 233–244, 2005. c ⃝ Springer-Verlag Berlin Heidelberg 2005 234 M. Shimohata and E. Sumita synonymous words but also the relations between the words in a each text. Automatic MCT construction can be performed in practice through state-of- the-art clustering techniques [2]. News articles are especially favorable for text clustering since they have both titles and date of publication. Synonym acquisition is based on a distributional hypothesis that words with similar meanings tend to appear in similar contexts [5]. In this work, we adopt loose contextual information that considers only the surrounding one word from each side of the target words. This narrow condition enables extraction from source texts2 that have different structures. In addition, we use another con- straint, prevention of outside appearance, which reduces improper extraction by looking over outside places of other texts. This constraint eliminates many non- synonyms having the same surrounding words by …

Content truncated to 500 words