Converting Phrase Structures to Dependency Structures in Sanskrit

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1834–1843, Dublin, Ireland, August 23-29 2014. Converting Phrase Structures to Dependency Structures in Sanskrit Pawan Goyal Department of CSE Indian Institute of Technology Kharagpur, India – 721302 pawang@cse.iitkgp.ernet.in Amba Kulkarni Department of Sanskrit Studies University of Hyderabad Hyderabad, India – 500046 apksh@uohyd.ernet.in Abstract Two annotations schemes for presenting the parsed structures are prevalent viz. the constituency structure and the dependency structure. While the constituency trees mark the relations due to positions, the dependency relations mark the semantic dependencies. Free word order languages like Sanskrit pose more problems for constituency parses since the elements within a phrase are dislocated. In this work, we show how the enriched constituency tree with the information of displacement can help construct the unlabelled dependency tree automatically. 1 Introduction Sanskrit has a rich tradition of linguistic analysis with intense discussions and argumentations on various aspects of language analysis ranging from phonetics (´siks.¯a), grammar (vy¯akaran.a), logic (ny¯aya), ritual exegesis (karmam¯ım¯am. s¯a), and literary theory (alam. k¯ara´s¯astra) which is not only useful for analysing Sanskrit but it also has much to offer computational linguistics in these areas. The series of symposia in Sanskrit Computational Linguistics (Huet et al., 2009; Kulkarni and Huet, 2009), the consortium project sponsored by the Technology Development for Indian Languages (TDIL) and the research of individual scholars and the collaborations (Goyal et al., 2012) among them resulted into a) development of several tools ranging from segmenters (Huet, 2009), morphological analysers (Kulkarni and Shukl, 2009), parsers (Goyal et al., 2009; Hellwig, 2009; Kumar, 2012; Kulkarni, 2013) to discourse annotators, b) lexical resources ranging from dictionaries, WordNet (Kulkarni et al., 2010) to Knowledge-Nets (Nair, 2011), and c) annotated corpora [http://sanskrit.uohyd.ernet.in/scl]. P¯an.inian grammar, the oldest dependency grammar, provides a formalism for annotation of the sentences. While the Sanskrit consortium has annotated a few thousand sentences following the dependency grammar, we also came across a very valuable source of annotation of Sanskrit sentences following the constituency structure (Gillon, 1996). The constituency structure was enriched to suite the requirements of Sanskrit. This aroused our curiosity to study the equivalence of the two annotation schemes. The importance of dependency structure has been well recognised by several computational linguists (Culotta and Sorensen, 2004; Haghighi et al., 2005; Quirk et al., 2005) in the recent past. The dependency format is preferred over the constituency not only from evaluation point of view (Lin, 1998) but also because of its suitability (Marneffe et al., 2006) for a wide range of NLP tasks such as Machine Translation (MT), information extraction, question answering etc.. This has upsurged several works on converting a constituency structure into dependency. The parsers for English now produce the dependency parse as well. Xia and Palmer discuss three different algorithms to convert dependency structures to phrase structures for English (Xia and Palmer, 2001). Magerman gave a set of priority lists, in the form of a head percolation table to find heads of constituents (Magerman, 1994). Yamada and Matsumoto …

Content truncated to 500 words