Structural Correspondence Learning for Dependency Parsing

Proceedings of the CoNLL Shared Task Session of EMNLP-CoNLL 2007, pp. 1166–1169, Prague, June 2007. c⃝2007 Association for Computational Linguistics Structural Correspondence Learning for Dependency Parsing Nobuyuki Shimizu Information Technology Center University of Tokyo Tokyo, Japan shimizu@r.dl.itc.u-tokyo.ac.jp Hiroshi Nakagawa Information Technology Center University of Tokyo Tokyo, Japan nakagawa@dl.itc.u-tokyo.ac.jp Abstract Following (Blitzer et al., 2006), we present an application of structural correspondence learning to non-projective dependency pars- ing (McDonald et al., 2005). To induce the correspondences among dependency edges from different domains, we looked at ev- ery two tokens in a sentence and examined whether or not there is a preposition, a de- terminer or a helping verb between them. Three binary linear classifiers were trained to predict the existence of a preposition, etc, on unlabeled data and we used singu- lar value decomposition to induce new fea- tures. During the training, the parser was trained with these additional features in ad- dition to these described in (McDonald et al., 2005). We discriminatively trained our parser in an on-line fashion using a vari- ant of the voted perceptron (Collins, 2002; Collins and Roark, 2004; Crammer and Singer, 2003). 1 Introduction We have recently seen growing popularity of depen- dency parsing. It is no longer rare to see dependency relations used as features, in tasks such as machine translation (Ding and Palmer, 2005) and relation ex- traction (Bunescu and Mooney, 2005). However, there is one factor that prevents the use of depen- dency parsing: sparseness of annotated corpora out- side Wall Street Journal. In many situations we need to parse sentences from a target domain with no la- beled data, which is a different distribution from a source domain where plentiful labeled training data is available. In this paper, we investigate the effectiveness of structural correspondence learning (SCL) (Blitzer et al., 2006) in the domain adaptation task given by the CoNLL 2007. They hypothesize that a model trained in the source domain using this common fea- ture representation will generalize better to the tar- get domain, and focus on using unlabeled data from both the source and target domains to learn a com- mon feature representation that is meaningful across both domains. The paper is structured as follows: in section 2, we review the decoding and learning aspects of (McDonald et al., 2005), in section 3, structural cor- respondence learning applied to dependency pars- ing, and in section 4, we describe the experiments and the features needed for the CoNLL 2006 shared task. 2 Non-Projective Dependency Parsing 2.1 Dependency Structure Let us define x to be a generic sequence of input to- kens together with their POS tags and other morpho- logical features, and y to be a generic dependency structure, that is, a set of edges for x. A labeled edge is a tuple ⟨DEPREL, i → j⟩ where i is the start point of the edge, j is the end point, and DEPREL is the label of the edge. The token at i is the head of the token at …

Content truncated to 500 words