Relaxed Cross-lingual Projection of Constituent Syntax

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1192–1201, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Relaxed Cross-lingual Projection of Constituent Syntax Wenbin Jiang and Qun Liu and Yajuan L¨u Key Laboratory of Intelligent Information Processing Institute of Computing Technology Chinese Academy of Sciences {jiangwenbin, liuqun, lvyajuan}@ict.ac.cn Abstract We propose a relaxed correspondence as- sumption for cross-lingual projection of con- stituent syntax, which allows a supposed constituent of the target sentence to corre- spond to an unrestricted treelet in the source parse. Such a relaxed assumption fundamen- tally tolerates the syntactic non-isomorphism between languages, and enables us to learn the target-language-specific syntactic idiosyn- crasy rather than a strained grammar di- rectly projected from the source language syn- tax. Based on this assumption, a novel con- stituency projection method is also proposed in order to induce a projected constituent tree- bank from the source-parsed bilingual cor- pus. Experiments show that, the parser trained on the projected treebank dramatically out- performs previous projected and unsupervised parsers. 1 Introduction For languages with treebanks, supervised models give the state-of-the-art performance in dependency parsing (McDonald and Pereira, 2006; Nivre et al., 2006; Koo and Collins, 2010; Martins et al., 2010) and constituent parsing (Collins, 2003; Charniak and Johnson, 2005; Petrov et al., 2006). To break the restriction of the treebank scale, lots of works have been devoted to the unsupervised methods (Klein and Manning, 2004; Bod, 2006; Seginer, 2007; Co- hen and Smith, 2009) and the semi-supervised meth- ods (Sarkar, 2001; Steedman et al., 2003; McClosky et al., 2006; Koo et al., 2008) to utilize the unan- notated text. In recent years, researchers have also conducted many investigations on syntax projection (Hwa et al., 2005; Ganchev et al., 2009; Smith and Eisner, 2009; Jiang et al., 2010), in order to borrow syntactic knowledge from another language. Different from the bilingual parsing (Smith and Smith, 2004; Burkett and Klein, 2008; Zhao et al., 2009; Huang et al., 2009; Chen et al., 2010) that improves parsing performance with bilingual con- straints, and the bilingual grammar induction (Wu, 1997; Kuhn, 2004; Blunsom et al., 2008; Snyder et al., 2009) that induces grammar from parallel text, the syntax projection aims to project the syntac- tic knowledge from one language to another. This seems especially promising for the languages that have bilingual corpora parallel to resource-rich lan- guages with large treebanks. Previous works mainly focus on dependency projection. The dependency relationship between words in the parsed source sen- tences can be directly projected across the word alignment to words in the target sentences, follow- ing the direct correspondence assumption (DCA) (Hwa et al., 2005). Due to the syntactic non- isomorphism between languages, DCA assumption usually leads to conflicting or incomplete projection. Researchers have to adopt strategies to tackle this problem, such as designing rules to handle language non-isomorphism (Hwa et al., 2005), and resorting to the quasi-synchronous grammar (Smith and Eis- ner, 2009). For constituency projection, however, the lack of isomorphism …

Content truncated to 500 words