Skip to Main Content
 

Global Search Box

 
 
 
 

ETD Abstract Container

Abstract Header

Recovering Chinese Nonlocal Dependencies with a Generalized Categorial Grammar

Abstract Details

2019, Doctor of Philosophy, Ohio State University, Linguistics.
Correctly resolving nonlocal dependencies is challenging (Rimell et al., 2009) but crucial for natural language processing tasks such as question-answering and information extraction. It is especially important for Mandarin Chinese, which makes a heavy use of nonlocal dependencies (Kummerfeld et al., 2013). This thesis describes a Generalized Categorial Grammar (GCG) which has a slightly larger set of inference rules tailored specifically for Mandarin Chinese and a much smaller set of categories compared with other more lexicalized categorial grammars. The -g category in the grammar provides an intuitive account for the filler-gap constructions in Mandarin Chinese. The thesis introduces a systematic reannotating process which converts the Penn Chinese Treebank annotations (Xue et al., 2005) into GCG annotations. The resulting broad-coverage GCG annotations can be used to train a probabilistic grammar using an automatic constituent parser. The reannotation process takes full advantages of the trace information annotated in the Penn Chinese Treebank so that the resulting GCG annotations retain the information needed for resolving nonlocal dependencies. A syntactic parsing evaluation shows that a Berkeley latent-variable parser (Petrov and Klein, 2007) trained on the Chinese GCG annotations is significantly more accurate than the same parser trained on a more lexicalized categorial grammar in predicting a common test set of gold binary unlabeled trees in both grammars. A semantic dependency parsing system based on GCG syntactic features achieves comparable performance for the overall labeled dependency prediction and best results in predicting multi-headed dependencies in a shared task of Chinese semantic dependency parsing. In order to conduct a construction-specific nonlocal dependency recovery evaluation in Mandarin Chinese, a set of test sets is annotated for eight nonlocal constructions in Mandarin Chinese according to their annotations in the Penn Chinese Treebank. The coverage evaluation across di erent dependency formalisms shows that the Chinese GCG annotations have the best coverage of the nonlocal dependencies annotated in the test sets compared with other dependency formalisms such as Chinese Stanford dependencies or the Penn2Malt dependencies. The nonlocal dependency recovery evaluation across different parsers shows that a higher accuracy in general parsing evaluation such as evalb does not necessarily correlate to better performance in nonlocal dependency recovery. The construction-specific parsing evaluation shows that recovering nonlocal dependencies from frequent nonlocal constructions such as subject and object relative clauses, passive voice is easier than that from infrequent nonlocal constructions, such as topicalization or extraction from embedded clauses.
William Schuler (Advisor)
Carl Pollard (Advisor)
Micha Elsner (Committee Member)
Zhiguo Xie (Committee Member)
123 p.

Recommended Citations

Citations

  • Duan, M. (2019). Recovering Chinese Nonlocal Dependencies with a Generalized Categorial Grammar [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu154622673336324

    APA Style (7th edition)

  • Duan, Manjuan. Recovering Chinese Nonlocal Dependencies with a Generalized Categorial Grammar. 2019. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu154622673336324.

    MLA Style (8th edition)

  • Duan, Manjuan. "Recovering Chinese Nonlocal Dependencies with a Generalized Categorial Grammar." Doctoral dissertation, Ohio State University, 2019. http://rave.ohiolink.edu/etdc/view?acc_num=osu154622673336324

    Chicago Manual of Style (17th edition)