Character-Level Dependencies in Chinese: Usefulness and Learning

Proceedings of the 12th Conference of the European Chapter of the ACL, pages 879–887, Athens, Greece, 30 March – 3 April 2009. c⃝2009 Association for Computational Linguistics Character-Level Dependencies in Chinese: Usefulness and Learning Hai Zhao Department of Chinese, Translation and Linguistics City University of Hong Kong Tat Chee Avenue, Kowloon, Hong Kong, China haizhao@cityu.edu.hk Abstract We investigate the possibility of exploit- ing character-based dependency for Chi- nese information processing. As Chinese text is made up of character sequences rather than word sequences, word in Chi- nese is not so natural a concept as in En- glish, nor is word easy to be defined with- out argument for such a language. There- fore we propose a character-level depen- dency scheme to represent primary lin- guistic relationships within a Chinese sen- tence. The usefulness of character depen- dencies are verified through two special- ized dependency parsing tasks. The first is to handle trivial character dependencies that are equally transformed from tradi- tional word boundaries. The second fur- thermore considers the case that annotated internal character dependencies inside a word are involved. Both of these results from character-level dependency parsing are positive. This study provides an alter- native way to formularize basic character- and word-level representation for Chinese. 1 Introduction In many human languages, word can be naturally identified from writing. However, this is not the case for Chinese, for Chinese is born to be written in character1 sequence rather than word sequence, namely, no natural separators such as blanks ex- ist between words. As word does not appear in a natural way as most European languages2, it 1Character here stands for various tokens occurring in a naturally written Chinese text, including Chinese charac- ter(hanzi), punctuation, and foreign letters. However, Chi- nese characters often cover the most part. 2Even in European languages, a naive but necessary method to properly define word is to list them all by hand. Thank the first anonymous reviewer who points this fact. brings the argument about how to determine the word-hood in Chinese. Linguists’ views about what is a Chinese word diverge so greatly that multiple word segmentation standards have been proposed for computational linguistics tasks since the first Bakeoff (Bakeoff-1, or Bakeoff-2003)3 (Sproat and Emerson, 2003). Up to Bakeoff-4, seven word segmentation stan- dards have been proposed. However, this does not effectively solve the open problem what a Chi- nese word should exactly be but raises another is- sue: what a segmentation standard should be se- lected for the successive application. As word often plays a basic role for the further language processing, if it cannot be determined in a uni- fied way, then all successive tasks will be affected more or less. Motivated by dependency representation for syntactic parsing since (Collins, 1999) that has been drawn more and more interests in recent years, we suggest that character-level dependen- cies can be adopted to alleviate this difficulty in Chinese processing. If we regard traditional word boundary as a linear representation for neighbored characters, then character-level dependencies …

Content truncated to 500 words