Less Grammar, More Features

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 228–237, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Less Grammar, More Features David Hall Greg Durrett Computer Science DivisionDan Klein University of California, Berkeley {dlwh,gdurrett,klein}@cs.berkeley.edu Abstract We present a parser that relies primar- ily on extracting information directly from surface spans rather than on propagat- ing information through enriched gram- mar structure. For example, instead of cre- ating separate grammar symbols to mark the definiteness of an NP, our parser might instead capture the same information from the first word of the NP. Moving context out of the grammar and onto surface fea- tures can greatly simplify the structural component of the parser: because so many deep syntactic cues have surface reflexes, our system can still parse accurately with context-free backbones as minimal as X- bar grammars. Keeping the structural backbone simple and moving features to the surface also allows easy adaptation to new languages and even to new tasks. On the SPMRL 2013 multilingual con- stituency parsing shared task (Seddah et al., 2013), our system outperforms the top single parser system of Bj¨orkelund et al. (2013) on a range of languages. In addi- tion, despite being designed for syntactic analysis, our system also achieves state- of-the-art numbers on the structural senti- ment task of Socher et al. (2013). Finally, we show that, in both syntactic parsing and sentiment analysis, many broad linguistic trends can be captured via surface features. 1 Introduction Na¨ıve context-free grammars, such as those em- bodied by standard treebank annotations, do not parse well because their symbols have too little context to constrain their syntactic behavior. For example, to PPs usually attach to verbs and of PPs usually attach to nouns, but a context-free PP symbol can equally well attach to either. Much of the last few decades of parsing research has therefore focused on propagating contextual in- formation from the leaves of the tree to inter- nal nodes. For example, head lexicalization (Eis- ner, 1996; Collins, 1997; Charniak, 1997), struc- tural annotation (Johnson, 1998; Klein and Man- ning, 2003), and state-splitting (Matsuzaki et al., 2005; Petrov et al., 2006) are all designed to take coarse symbols like PP and decorate them with additional context. The underlying reason that such propagation is even needed is that PCFG parsers score trees based on local configurations only, and any information that is not threaded through the tree becomes inaccessible to the scor- ing function. There have been non-local ap- proaches as well, such as tree-substitution parsers (Bod, 1993; Sima’an, 2000), neural net parsers (Henderson, 2003), and rerankers (Collins and Koo, 2005; Charniak and Johnson, 2005; Huang, 2008). These non-local approaches can actually go even further in enriching the grammar’s struc- tural complexity by coupling larger domains in various ways, though their non-locality generally complicates inference. In this work, we instead try to minimize the structural complexity of the grammar by moving as much context as possible onto local surface fea- tures. …

Content truncated to 500 words