Deep-Syntactic Parsing

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1402–1413, Dublin, Ireland, August 23-29 2014. Deep-Syntactic Parsing Miguel Ballesteros1, Bernd Bohnet2, Simon Mille1, Leo Wanner1,3 1Natural Language Processing Group, Pompeu Fabra University, Barcelona, Spain 2School of Computer Science, University of Birmingham, United Kingdom 3Catalan Institute for Research and Advanced Studies (ICREA) 1,3{name.lastname}@upf.edu 2bohnetb@cs.bham.ac.uk Abstract “Deep-syntactic” dependency structures that capture the argumentative, attributive and coordi- native relations between full words of a sentence have a great potential for a number of NLP- applications. The abstraction degree of these structures is in-between the output of a syntactic dependency parser (connected trees defined over all words of a sentence and language-specific grammatical functions) and the output of a semantic parser (forests of trees defined over indi- vidual lexemes or phrasal chunks and abstract semantic role labels which capture the argument structure of predicative elements, dropping all attributive and coordinative dependencies). We propose a parser that delivers deep syntactic structures as output. 1 Introduction Surface-syntactic structures (SSyntSs) as produced by data-driven syntactic dependency parsers are per force idiosyncratic in that they contain governed prepositions, determiners, support verb constructions and language-specific grammatical functions such as, e.g., SBJ, OBJ, PRD, PMOD, etc. (Johansson and Nugues, 2007). For many NLP-applications, including machine translation, paraphrasing, text simpli- fication, etc., such a high idiosyncrasy is obstructive because of the recurrent divergence between the source and the target structures. Therefore, the use of more abstract “syntactico-semantic” structures seems more appropriate. Following Mel’ˇcuk (1988), we call these structures deep-syntactic structures (DSyntSs). DSyntSs are situated between SSyntSs and PropBank- (Palmer et al., 2005) or Semantic Frame-like structures (Fillmore et al., 2002). Compared to SSyntSs, they have the advantage to ab- stract from language-specific grammatical idiosyncrasies. Compared to PropBank and Semantic Frame stuctures, they have the advantage to be connected and complete, i.e., capture all argumentative, attribu- tive and coordinative dependencies between the meaningful lexical items of a sentence, while PropBank and Semantic Frame structures are not always connected, may contain either individual lexical items or phrasal chunks as nodes, and discard attributive and coordinative relations (be they within the chunks or sentential). In other words, they constitute incomplete structures that drop not only idiosyncratic, func- tional but also meaningful elements of a given sentence and often contain dependencies between chunks rather than individual tokens. Therefore, we propose to put on the research agenda the task of deep- syntactic parsing and show how a DSyntS is obtained from a SSynt dependency parse using data-driven tree transduction in a pipeline with a syntactic parser.1 In Section 2, we introduce SSyntSs and DSyntSs and discuss the fundamentals of SSyntS–DSyntS transduction. Section 3 describes the experiments that we carried out on Spanish material, and Section 4 discusses their outcome. Section 5 summarizes the related work, before in Section 6 some conclusions and plans for future work are presented. 2 Fundamentals of SSyntS–DSyntS transduction Before we set out to discuss the principles of the SSyntS–DSynt transduction, we must specify the DSyntSs and …

Content truncated to 500 words