Dependency-based empty category detection via phrase structure trees

Proceedings of NAACL-HLT 2013, pages 1051–1060, Atlanta, Georgia, 9–14 June 2013. c⃝2013 Association for Computational Linguistics Dependency-based empty category detection via phrase structure trees Nianwen Xue Brandeis University Waltham, MA, USA xuen@brandeis.edu Yaqin Yang Brandeis University Waltham, MA, USA yaqin@brandeis.edu Abstract We describe a novel approach to detecting empty categories (EC) as represented in de- pendency trees as well as a new metric for measuring EC detection accuracy. The new metric takes into account not only the position and type of an EC, but also the head it is a dependent of in a dependency tree. We also introduce a variety of new features that are more suited for this approach. Tested on a sub- set of the Chinese Treebank, our system im- proved significantly over the best previously reported results even when evaluated with this more stringent metric. 1 Introduction In modern theoretical linguistics, empty categories (ECs) are an important piece of machinery in repre- senting the syntactic structure of a sentence and they are used to represent phonologically null elements such as dropped pronouns and traces of dislocated elements. They have also found their way into large- scale treebanks which have played an important role in advancing the state of the art in syntactic parsing. In phrase-structure treebanks, ECs have been used to indicate long-distance dependencies, discontinuous constituents, and certain dropped elements (Marcus et al., 1993; Xue et al., 2005). Together with la- beled brackets and function tags, they make up the full syntactic representation of a sentence. The use of ECs captures some cross-linguistic commonalities and differences. For example, while both the Penn English TreeBank (PTB) (Marcus et al., 1993) and the Chinese TreeBank (CTB) (Xue et al., 2005) use traces to represent the extraction site of a dislocated element, dropped pronouns (rep- resented as *pro*s) are much more widespread in the CTB. This is because Chinese is a pro-drop lan- guage (Huang, 1984) that allows the subject to be dropped in more contexts than English does. While detecting and resolving traces is important to the in- terpretation of the syntactic structure of a sentence in both English and Chinese, the prevalence of dropped nouns in Chinese text gives EC detection added sig- nificance and urgency. They are not only an impor- tant component of the syntactic parse of a sentence, but are also essential to a wide range of NLP appli- cations. For example, any meaningful tracking of entities and events in natural language text would have to include those represented by dropped pro- nouns. If Chinese is translated into a different lan- guage, it is also necessary to render these dropped pronouns explicit if the target language does not al- low pro-drop. In fact, Chung and Gildea (2010) re- ported preliminary work that has shown a positive impact of automatic EC detection on statistical ma- chine translation. Some ECs can be resolved to an overt element in the same text while others only have a generic ref- erence that cannot be linked to any specific …

Content truncated to 500 words