Chinese CCGbank: extracting CCG derivations from the Penn Chinese Treebank

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 1083–1091, Beijing, August 2010 Chinese CCGbank: extracting CCG derivations from the Penn Chinese Treebank Daniel Tse and James R. Curran School of Information Technologies University of Sydney {dtse6695,james}@it.usyd.edu.au Abstract Automated conversion has allowed the de- velopment of wide-coverage corpora for a variety of grammar formalisms without the expense of manual annotation. Analysing new languages also tests formalisms, ex- posing their strengths and weaknesses. We present Chinese CCGbank, a 760,000 word corpus annotated with Combinatory Categorial Grammar (CCG) derivations, in- duced automatically from the Penn Chi- nese Treebank (PCTB). We design parsimo- nious CCG analyses for a range of Chinese syntactic constructions, and transform the PCTB trees to produce them. Our process yields a corpus of 27,759 derivations, cov- ering 98.1% of the PCTB. 1 Introduction An annotated corpus is typically used to develop statistical parsers for a given formalism and lan- guage. An alternative to the enormous cost of hand-annotating a corpus for a specific formalism is to convert from an existing corpus. The Penn Treebank (PTB; Marcus et al., 1994) has been converted to HPSG (Miyao et al., 2004), LFG (Cahill et al., 2002), LTAG (Xia, 1999), and CCG (Hockenmaier, 2003). Dependency corpora, e.g. the German Tiger corpus, have also been con- verted (Hockenmaier, 2006). The Penn Chinese Treebank (PCTB; Xue et al., 2005) provides analy- ses for 770,000 words of Chinese. Existing PCTB conversions have targeted TAG (Chen et al., 2005) and LFG (Burke and Lam, 2004; Guo et al., 2007). We present Chinese CCGbank, a Chinese cor- pus of CCG derivations automatically induced from the PCTB. Combinatory Categorial Grammar (CCG; Steedman, 2000) is a lexicalised grammar formal- ism offering a unified account of local and non- local dependencies. We harness the facilities of CCG to provide analyses of Chinese syntax includ- ing topicalisation, pro-drop, zero copula, extrac- tion, and the 把 ba- and 被 bei-constructions. Pushing the boundaries of formalisms by sub- jecting them to unfamiliar syntax also tests their universality claims. The freer word order of Turk- ish (Hoffman, 1996) and the complex morphology of Korean (Cha et al., 2002) led to the development of extensions to the CCG formalism. We present our analysis of Chinese syntax un- der CCG, and provide an algorithm, modelled af- ter Hockenmaier and Steedman (2007), to incre- mentally transform PCTB trees into CCG derivations. The algorithm assigns CCG categories which di- rectly encode head and subcategorisation informa- tion. Instances of Chinese syntax demanding spe- cial analysis, such as extraction, pro-drop or topi- calisation, are pin-pointed and given elegant anal- yses which exploit the expressivity of CCG. Our conversion yields CCG analyses for 27,759 PCTB trees (98.1%). Coverage on lexical items, evaluated by 10-fold cross-validation, is 94.46% (by token) and 73.38% (by type). We present the first CCG analysis of Chinese syntax and obtain a wide-coverage CCG corpus of Chinese. Highly efficient statistical parsing using a CCGbank has recently been demonstrated for English (Clark and Curran, 2007). Our Chinese …

Content truncated to 500 words