Lexical Gaps and Lexicalization: Implications for Word Segmentation Systems for Chinese NLP

Copyright 2012 by Chan-Chia Hsu 26th Pacific Asia Conference on Language,Information and Computation pages 191–198 Lexical Gaps and Lexicalization: Implications for Word Segmentation Systems for Chinese NLP Chan-Chia Hsu Graduate Institute of Linguistics, National Taiwan University No. 1, Sec. 4, Roosevelt Road, Taipei, 10617 Taiwan chanchiah@gmail.com Abstract This paper is motivated by the observation that not all adjectives in Chinese have a canonical antonym. For example, most Chinese speakers choose to translate the English word dishonest into a word string bu chengshi ‘not honest’ instead of any antonym candidates of chengshi suggested in antonym dictionaries. Our discourse evidence from corpus data suggests that bu chengshi is evolving into a word in discourse at a faster pace than some other ‘bu + adjective’ strings, and this may result from the lexical gap for a canonical antonym of chengshi and the communicative need for such a word. As a consequence, it is proposed that if the lexicalization process of bu chengshi continues in the future, the string may need to be considered a single word in a segmentation system (i.e., buchengshi ‘dishonest’). For a segmentation system to distinguish between words and phrases, discourse factors should be taken into consideration. 1 Introduction The issue of antonym canonicity has been empirically investigated in English and other languages (Paradis et al., 2009; Willners and Paradis, 2010). This paper is motivated by the observation that not all adjectives in Chinese have a generally accepted antonym. For example, although the antonym of chengshi ‘honest’, according to antonym dictionaries (e.g., Han and Song, 2001; Xu, 2000), can be xuwei ‘hypocritical’, xujia ‘unreal’, and jiaohua ‘cunning’, none of them are canonical for most native speakers. Intriguingly, in translating dishonest into Chinese, most Chinese speakers also choose to translate the English word into a word string bu chengshi ‘not honest’ instead of any antonym candidates of chengshi. The aim of this paper is thus to address the question: Will the lexical gap for a canonical antonym of chengshi enable the string bu chengshi to evolve into a word? To answer the question, we adopt a corpus-based approach and see how bu chengshi behaves in discourse. The results will have implications for word segmentation systems for Chinese NLP in that if bu chengshi functions like a word both linguistically and conceptually, then the string may not need to be further segmented into ‘bu + chengshi’ in a segmentation system. This line of study can shed light on the segmentation task from a discourse perspective. This paper is organized as follows. Section 2 reviews different views of what a word is. Section 3 introduces the data examined in the present study, and Section 4 presents the results. Section 5 discusses the implications of the results for Chinese wordhood and the segmentation task in Chinese. Section 6 offers the conclusion and some suggestions for future research. 2 What Is a Word? Generally, a word is defined as “a unit which has universal intuitive recognition by native-speakers, in both spoken and written language” (Crystal, …

Content truncated to 500 words