Unified Dependency Parsing of Chinese Morphological and Syntactic Structures

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 1445–1454, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Unified Dependency Parsing of Chinese Morphological and Syntactic Structures Zhongguo Li Guodong Zhou Natural Language Processing Laboratory School of Computer Science and Technology Soochow University, Suzhou, Jiangsu Province 215006, China {lzg, gdzhou}@suda.edu.cn Abstract Most previous approaches to syntactic pars- ing of Chinese rely on a preprocessing step of word segmentation, thereby assuming there was a clearly defined boundary between mor- phology and syntax in Chinese. We show how this assumption can fail badly, leading to many out-of-vocabulary words and incom- patible annotations. Hence in practice the strict separation of morphology and syntax in the Chinese language proves to be untenable. We present a unified dependency parsing ap- proach for Chinese which takes unsegmented sentences as input and outputs both morpho- logical and syntactic structures with a single model and algorithm. By removing the inter- mediate word segmentation, the unified parser no longer needs separate notions for words and phrases. Evaluation proves the effective- ness of the unified model and algorithm in parsing structures of words, phrases and sen- tences simultaneously. 1 1 Introduction The formulation of the concept of words has baf- fled linguists from ancient to modern times (Hock- ett, 1969). Things are even worse for Chinese, partly due to the fact that its written form does not delimit words explicitly. While we have no doubt that there are linguistic units which are definitely words (or phrases, for that matter), it’s a sad truth that in many cases we cannot manage to draw such a clear bound- ary between morphology and syntax, for which we now give two arguments. 1Corresponding author is Guodong Zhou. The first argument is that many sub-word linguis- tic units (such as suffixes and prefixes) are so pro- ductive that they can lead to a huge number of out- of-vocabulary words for natural language process- ing systems. This phenomenon brings us into an awkward situation if we adhere to a rigid separa- tion of morphology and syntax. Consider charac- ter 者 ‘someone’ as an example. On the one hand, there is strong evidence that it’s not a word as it can never be used alone. On the other hand, taking it as a mere suffix leads to many out-of-vocabulary words because of the productivity of such characters. For instance, Penn Chinese Treebank (CTB6) contains 失败者 ‘one that fails’ as a word but not 成功者 ‘one that succeeds’, even with the word 成功 ‘suc- ceed’ appearing 207 times. We call words like 成 功者 ‘one that succeeds’ pseudo OOVs. By defini- tion, pseudo OOVs are OOVs since they do not occur in the training corpus, though their components are frequently-seen words. Our estimation is that over 60% of OOVs in Chinese are of this kind (Section 2). Of course, the way out of this dilemma is to parse the internal structures of these words. That …

Content truncated to 500 words