tailieunhanh - Báo cáo khoa học: "Structural Correspondence Learning for Parse Disambiguation"

The paper presents an application of Structural Correspondence Learning (SCL) (Blitzer et al., 2006) for domain adaptation of a stochastic attribute-value grammar (SAVG). So far, SCL has been applied successfully in NLP for Part-of-Speech tagging and Sentiment Analysis (Blitzer et al., 2006; Blitzer et al., 2007). An attempt was made in the CoNLL 2007 shared task to apply SCL to non-projective dependency parsing (Shimizu and Nakagawa, 2007), however, without any clear conclusions. We report on our exploration of applying SCL to adapt a syntactic disambiguation model and show promising initial results on Wikipedia domains. . | Structural Correspondence Learning for Parse Disambiguation Barbara Plank Alfa-informatica University of Groningen The Netherlands Abstract The paper presents an application of Structural Correspondence Learning SCL Blitzer et al. 2006 for domain adaptation of a stochastic attribute-value grammar SAVG . So far SCL has been applied successfully in NLP for Part-of-Speech tagging and Sentiment Analysis Blitzer et al. 2006 Blitzer et al. 2007 . An attempt was made in the CoNLL 2007 shared task to apply SCL to non-projective dependency parsing Shimizu and Nakagawa 2007 however without any clear conclusions. We report on our exploration of applying SCL to adapt a syntactic disambiguation model and show promising initial results on Wikipedia domains. 1 Introduction Many current effective natural language processing systems are based on supervised Machine Learning techniques. The parameters of such systems are estimated to best reflect the characteristics of the training data at the cost of portability a system will be successful only as long as the training material resembles the input that the model gets. Therefore whenever we have access to a large amount of labeled data from some source out-of-domain but we would like a model that performs well on some new target domain Gildea 2001 Daume III 2007 we face the problem of domain adaptation. The need for domain adaptation arises in many NLP tasks Part-of-Speech tagging Sentiment Analysis Semantic Role Labeling or Statistical Parsing to name but a few. For example the performance of a statistical parsing system drops in an appalling way when a model trained on the Wall Street Journal is applied to the more varied Brown corpus Gildea 2001 . The problem itself has started to get attention only recently Roark and Bacchiani 2003 Hara et al. 2005 Daume III and Marcu 2006 Daume III 2007 Blitzer et al. 2006 McClosky et al. 2006 Dredze et al. 2007 . We distinguish two main approaches to domain adaptation that have .