An In-depth Study on Internal Structure of Chinese Words

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 5823–5833 August 1–6, 2021. ©2021 Association for Computational Linguistics 5823 An In-depth Study on Internal Structure of Chinese Words Chen Gong1, Saihao Huang1∗, Houquan Zhou1, Zhenghua Li1, Min Zhang1, Zhefeng Wang2, Baoxing Huai2, Nicholas Jing Yuan2 1Institute of Artificial Intelligence, School of Computer Science and Technology, Soochow University, China; 2 Huawei Cloud, China 1{cgong,shhuang1999,hqzhou}@stu.suda.edu.cn 1{zhli13,minzhang}@suda.edu.cn 2{wangzhefeng, huaibaoxing, nicholas.yuan}@huawei.com Abstract Unlike English letters, Chinese characters have rich and specific meanings. Usually, the meaning of a word can be derived from its con- stituent characters in some way. Several previ- ous works on syntactic parsing propose to an- notate shallow word-internal structures for bet- ter utilizing character-level information. This work proposes to model the deep internal struc- tures of Chinese words as dependency trees with 11 labels for distinguishing syntactic re- lationships. First, based on newly compiled annotation guidelines, we manually annotate a word-internal structure treebank (WIST) con- sisting of over 30K multi-char words from Chinese Penn Treebank. To guarantee qual- ity, each word is independently annotated by two annotators and inconsistencies are han- dled by a third senior annotator. Second, we present detailed and interesting analysis on WIST to reveal insights on Chinese word for- mation. Third, we propose word-internal struc- ture parsing as a new task, and conduct bench- mark experiments using a competitive depen- dency parser. Finally, we present two simple ways to encode word-internal structures, lead- ing to promising gains on the sentence-level syntactic parsing task. 1 Introduction Unlike English, Chinese adopts a logographic writ- ing system and contains tens of thousands of dis- tinct characters. Many characters, especially fre- quently used ones, have rich and specific meanings. However, words, instead of characters, are often considered as the basic unit in processing Chinese texts. We believe the reason may be two-fold. First, usually a character may have many meanings and usages. Word formation process greatly reduces such char-level ambiguity. Second, by definition, ∗Chen Gong and Saihao Huang make equal contributions to this work. Zhenghua is the corresponding author. coordinate left 法 设 left 方 想 root think plan design method right 法 coordinate 姻 婚 marriage marriage law coordinate coordinate 老 法 Pharaoh (a) Zhang et al. (2014): labels mark head positions. 法 设 方 想 root v-v-n v-v-v v-v-n think plan design method 法 姻 婚 n-n-n n-n-n root marriage marriage law 老 法 n-n-n root Pharaoh (b) Li et al. (2018): labels correspond to POS tag triples. 法 设 方 想 root obj coo obj think plan design method 法 姻 婚 attcoo root marriage marriage law 老 法 frag root Pharaoh (c) Ours: fine-grained structure with 11 labels. Figure 1: Three example words with internal struc- ture under different annotation paradigms. “想(think of) 方(plan) 设(design) 法(method)” is a verb and means “find ways or means to do”. “婚(marriage) 姻(marriage) 法(law)” is a noun. “法老” is phonetic transliteration of “Pharaoh”. The three words …

Content truncated to 500 words