Syntactic Parse Fusion

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 1360–1366, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Syntactic Parse Fusion Do Kook Choe Brown University Providence, RI dc65@cs.brown.edu David McClosky IBM Research Yorktown Heights, NY dmcclosky@us.ibm.com Eugene Charniak Brown University Providence, RI ec@cs.brown.edu Abstract Model combination techniques have con- sistently shown state-of-the-art perfor- mance across multiple tasks, including syntactic parsing. However, they dramat- ically increase runtime and can be diffi- cult to employ in practice. We demon- strate that applying constituency model combination techniques to n-best lists in- stead of n different parsers results in sig- nificant parsing accuracy improvements. Parses are weighted by their probabilities and combined using an adapted version of Sagae and Lavie (2006). These accu- racy gains come with marginal computa- tional costs and are obtained on top of ex- isting parsing techniques such as discrim- inative reranking and self-training, result- ing in state-of-the-art accuracy: 92.6% on WSJ section 23. On out-of-domain cor- pora, accuracy is improved by 0.4% on average. We empirically confirm that six well-known n-best parsers benefit from the proposed methods across six domains. 1 Introduction Researchers have proposed many algorithms to combine parses from multiple parsers into one fi- nal parse (Henderson and Brill, 1999; Zeman and ˇZabokrtsk`y, 2005; Sagae and Lavie, 2006; Now- son and Dale, 2007; Fossum and Knight, 2009; Petrov, 2010; Johnson and Ural, 2010; Huang et al., 2010; McDonald and Nivre, 2011; Shindo et al., 2012; Narayan and Cohen, 2015). These new parses are substantially better than the originals: Zhang et al. (2009) combine outputs from mul- tiple n-best parsers and achieve an F1 of 92.6% on the WSJ test set, a 0.5% improvement over their best n-best parser. Model combination ap- proaches tend to fall into the following categories: hybridization, where multiple parses are combined into a single parse; switching, which picks a sin- gle parse according to some criteria (usually a form of voting); grammar merging where gram- mars are combined before or during parsing; and stacking, where one parser sends its prediction to another at runtime. All of these have at least one of the caveats that (1) overall computation is increased and runtime is determined by the slowest parser and (2) using multiple parsers in- creases the system complexity, making it more difficult to deploy in practice. In this paper, we describe a simple hybridization extension (“fu- sion”) which obtains much of hybridization’s ben- efits while using only a single n-best parser and minimal extra computation. Our method treats each parse in a single parser’s n-best list as a parse from n separate parsers. We then adapt parse combination methods by Henderson and Brill (1999), Sagae and Lavie (2006), and Fos- sum and Knight (2009) to fuse the constituents from the n parses into a single tree. We empir- ically show that six n-best parsers benefit from parse fusion across six domains, obtaining state- of-the-art results. These improvements are com- plementary to other techniques such as rerank- …

Content truncated to 500 words