Improving Data Driven Dependency Parsing using Clausal Information

Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the ACL, pages 657–660, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Improving Data Driven Dependency Parsing using Clausal Information � Phani Gadde, Karan Jindal, Samar Husain, Dipti Misra Sharma, Rajeev Sangal Language Technologies Research Centre, IIIT-Hyderabad, India. phani.gadde@research.iiit.ac.in, karan_jindal@students.iiit.ac.in, {samar,dipti,sangal}@mail.iiit.ac.in Abstract The paper describes a data driven dependency parsing approach which uses clausal informa- tion of a sentence to improve the parser per- formance. The clausal information is added automatically during the parsing process. We demonstrate the experiments on Hindi, a lan- guage with relatively rich case marking sys- tem and free-word-order. All the experiments are done using a modified version of MSTParser. We did all the experiments on the ICON 2009 parsing contest data. We achieved an improvement of 0.87% and 0.77% in unla- beled attachment and labeled attachment accu- racies respectively over the baseline parsing accuracies. 1 Introduction Linguistic analysis of morphologically rich free- word-order languages (MoRFWO) using depen- dency framework have been argued to be more effective (Shieber, 1985; Mel’čuk, 1988, Bharati et al., 1993). Not surprisingly, most parsers for such languages are dependency based (Nivre et al., 2007a; Bharati et al., 2008a; Hall et al., 2007). In spite of availability of annotated treebanks, state- of-the-art parsers for MoRFWO have not reached the performance obtained for English. Some of the reasons stated for the low performance are small treebank size, complex linguistic phenomenon, long-distance dependencies, and non-projective structures (Nivre et al., 2007a, 2007b; Bharati et al., 2008a). Several approaches have been tried to handle these difficulties in such languages. For Hindi, Bharati et al. (2008a) and Ambati et al. (2009) used semantic features in parsing to reduce the negative impact of unavailable syntactic features and showed that use of minimal semantics can help in identifying cer- tain core dependency labels. Various attempts have proved to simplify the structure by dividing the sentence into suitable linguistic units (Attardi and Dell’Orletta 2008; Bharati et al., 1993, 2008b, 2009; Husain et al., 2009). These approaches han- dle complex structures by breaking the parsing process into several steps. Attardi and Dell'Orletta (2008) used chunk information as a feature to MaltParser (Nivre et al., 2007a) for parsing Eng- lish. Bharati et al., 1993 used the notion of local word groups, while Bharati et al., 2009 and Husain et al., 2009 used clauses. In this paper, we describe a data driven depen- dency parsing approach which uses clausal infor- mation of a sentence to improve the parser performance. Previous attempts at data driven parsing for Hindi have failed to exploit this feature explicitly. The clausal information is added auto- matically during the parsing process. We demon- strate the experiments on Hindi1. All the experiments are done using a modified version of MSTParser (McDonald et al., 2005a and the refer- ences therein) (henceforth MST) on the ICON 2009 parsing contest2 (Husain, 2009) data. We achieved an improvement of 0.87% and 0.77% in unlabeled attachment and labeled attachment accu- …

Content truncated to 500 words