A Baseline System for Chinese Near-Synonym Choice

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1366–1370, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP A Baseline System for Chinese Near-Synonym Choice Liang-Chih Yu1, Wei-Nan Chien1,2 and Shih-Ting Chen1 1Department of Information Management, Yuan Ze University, Taiwan, R.O.C. 2Information Technology Center, National Taiwan Normal University, Taiwan, R.O.C. Contact: lcyu@saturn.yzu.edu.tw Abstract Near-synonym sets represent groups of words with similar meaning, which are useful knowl- edge resources for many natural language ap- plications such as query expansion for infor- mation retrieval (IR) and computer-assisted language learning. However, near-synonyms are not necessarily interchangeable in contexts due to their specific usage and syntactic con- straints. Previous studies have developed vari- ous methods for near-synonym choice in Eng- lish sentences. To our best knowledge, there is no such evaluation on Chinese sentences. Therefore, this paper implements two baseline systems: pointwise mutual information (PMI) and a 5-gram language model that are widely used in previous work for Chinese near- synonym choice evaluation. Experimental re- sults show that the 5-gram language model achieves higher accuracy than PMI. 1 Introduction Lexical semantics plays an important role in many natural language applications. For instance, knowing that the word “arm” has (at least) the senses weapon and bodypart enables systems to perform word sense disambiguation. Knowing in addition the near-synonyms of a word can fur- ther improve the applications’ effectiveness. For instance, knowing that the weapon sense of “arm” corresponds to the weapon sense of “weapon” and of “arsenal” means that systems can in addition perform term expansion for in- formation retrieval (Moldovan and Mihalcea, 2000; Bhogal et al., 2007), (near-)duplicate de- tection for summarization, alternative word se- lection for writing support systems (Inkpen and Hirst, 2006; Inkpen, 2007; Wu et al., 2010), as well as computer-assisted language learning (Cheng et al., 2004; Ouyang et al., 2009). Recent studies have shown that near- synonyms are not necessarily interchangeable in practical use due to their specific usage and col- locational constraints, as shown in the following examples. (1) {strong, powerful} coffee (Pearce, 2001) (2) ghastly {error, mistake} (Inkpen, 2007) (3) {bridge, overpass, tunnel} under the bay (Yu et al., 2010) Example (1) and (2) present an example of col- locational constraints in given contexts. In (1), the word “strong” in the near-synonym set {strong, powerful} is more suitable than “power- ful” to fit the given context “coffee”, since “powerful coffee” is an anti-collocation. Simi- larly, in (2), “mistake” is more suitable than “er- ror” because “ghastly mistake” is a collocation and “ghastly error” is an anti-collocation. In (3), the near-synonym set {bridge, overpass, tunnel} represents the meaning of a physical structure that connects separate places by traversing an obstacle. Suppose that the original word in the given context “under the bay” is “tunnel”. It can be found that the word “tunnel” cannot be substi- tuted by the other words in the same set because all the substitutions are semantically implausible. The above examples indicate that not all words in a near-synonym set can …

Content truncated to 500 words