Multilingual Dependency Parsing Using Global Features

Proceedings of the CoNLL Shared Task Session of EMNLP-CoNLL 2007, pp. 952–956, Prague, June 2007. c⃝2007 Association for Computational Linguistics Multilingual Dependency Parsing using Global Features Tetsuji Nakagawa Oki Electric Industry Co., Ltd. 2−5−7 Honmachi, Chuo-ku, Osaka 541−0053, Japan nakagawa378@oki.com Abstract In this paper, we describe a two-stage multi- lingual dependency parser used for the mul- tilingual track of the CoNLL 2007 shared task. The system consists of two compo- nents: an unlabeled dependency parser us- ing Gibbs sampling which can incorporate sentence-level (global) features as well as token-level (local) features, and a depen- dency relation labeling module based on Support Vector Machines. Experimental re- sults show that the global features are useful in all the languages. 1 Introduction Making use of as many informative features as pos- sible is crucial to obtain high performance in ma- chine learning based NLP. Recently, several meth- ods for incorporating non-local features have been investigated, though such features often make mod- els complex and thus complicate inference. Collins and Koo (2005) proposed a reranking method for phrase structure parsing with which any type of global features in a parse tree can be used. For dependency parsing, McDonald and Pereira (2006) proposed a method which can incorporate some types of global features, and Riedel and Clarke (2006) studied a method using integer linear pro- gramming which can incorporate global linguistic constraints. In this paper, we study dependency parsing using Gibbs sampling which can incorpo- rate any type of global feature in a sentence. The parser determines unlabeled dependency structures only, and we attach dependency relation labels us- ing Support Vector Machines afterwards. We participated in the multilingual track of the CoNLL 2007 shared task (Nivre et al., 2007), and evaluated the system on data sets of 10 languages (Hajiˇc et al., 2004; Aduriz et al., 2003; Mart´ı et al., 2007; Chen et al., 2003; B¨ohmov´a et al., 2003; Marcus et al., 1993; Johansson and Nugues, 2007; Prokopidis et al., 2005; Csendes et al., 2005; Mon- temagni et al., 2003; Oflazer et al., 2003). The rest of the paper describes the specification of the system and the evaluation results. 2 Unlabeled Dependency Parsing using Global Features 2.1 Probabilistic Model Rosenfeld et al. (2001) proposed whole-sentence ex- ponential language models which can incorporate arbitrary features in a sentence, and we consider here a similar probabilistic model for dependency pars- ing which can incorporate any sentence-level fea- ture. Let w = w1 · · · w|w| denote an input sentence consisting of |w| tokens, and h = h1 · · · h|w| denote the sequence of the indices of each token’s head. Root nodes of a sentence do not have heads, and we regard the index of a root node’s head as zero, i.e., hi ∈ {0, 1, · · · , |w|} \ {i}. We define the probabil- ity distribution of the dependency structure h given a sentence w using exponential models as follows: PΛ,M(h|w)= 1 ZΛ,M(w)QM(h|w)exp � K � k=1 λkfk(w, h) � ,(1) ZΛ,M(w)= …

Content truncated to 500 words