A Fast, Accurate, Non-Projective, Semantically-Enriched Parser

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1257–1268, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics A Fast, Accurate, Non-Projective, Semantically-Enriched Parser Stephen Tratz and Eduard Hovy Information Sciences Institute University of Southern California Marina del Rey, California 90292 {stratz,hovy}@isi.edu Abstract Dependency parsers are critical components within many NLP systems. However, cur- rently available dependency parsers each ex- hibit at least one of several weaknesses, in- cluding high running time, limited accuracy, vague dependency labels, and lack of non- projectivity support. Furthermore, no com- monly used parser provides additional shal- low semantic interpretation, such as prepo- sition sense disambiguation and noun com- pound interpretation. In this paper, we present a new dependency-tree conversion of the Penn Treebank along with its associated fine-grain dependency labels and a fast, accurate parser trained on it. We explain how a non-projective extension to shift-reduce parsing can be in- corporated into non-directional easy-first pars- ing. The parser performs well when evalu- ated on the standard test section of the Penn Treebank, outperforming several popular open source dependency parsers; it is, to the best of our knowledge, the first dependency parser capable of parsing more than 75 sentences per second at over 93% accuracy. 1 Introduction Parsers are critical components within many natu- ral language processing (NLP) systems, including systems for information extraction, question answer- ing, machine translation, recognition of textual en- tailment, summarization, and many others. Unfortu- nately, currently available dependency parsers suf- fer from at least one of several weaknesses includ- ing high running time, limited accuracy, vague de- pendency labels, and lack of non-projectivity sup- port. Furthermore, few parsers include any sort of additional semantic interpretation, such as interpre- tations for prepositions, possessives, or noun com- pounds. In this paper, we describe 1) a new dependency conversion (Section 3) of the Penn Treebank (Mar- cus, et al., 1993) along with the associated de- pendency label scheme, which is based upon the Stanford parser’s popular scheme (de Marneffe and Manning, 2008), and a fast, accurate dependency parser with non-projectivity support (Section 4) and additional integrated semantic annotation modules for automatic preposition sense disambiguation and noun compound interpretation (Section 5). We show how Nivre’s (2009) swap-based reordering tech- nique for non-projective shift-reduce-style parsing can be integrated into the non-directional easy-first framework of Goldberg and Elhadad (2010) to sup- port non-projectivity, and we report the results of our parsing experiments on the standard test section of the PTB, providing comparisons with several freely available parsers, including Goldberg and Elhadad’s (2010) implementation, MALTPARSER (Nivre et al., 2006), MSTPARSER (McDonald et al., 2005; Mc- Donald and Pereira, 2006), the Charniak (2000) parser, and the Berkeley parser (Petrov et al., 2006; Petrov and Klein, 2007). The experimental results show that the parser is substantially more accurate than Goldberg and El- hadad’s original implementation, with fairly simi- lar overall speed. Furthermore, the results prove that Stanford-granularity dependency labels can be learned by modern dependency parsing systems when using our Treebank conversion, …

Content truncated to 500 words