Addressing the Data Sparsity Issue in Neural AMR Parsing

Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers, pages 366–375, Valencia, Spain, April 3-7, 2017. c⃝2017 Association for Computational Linguistics Addressing the Data Sparsity Issue in Neural AMR Parsing Xiaochang Peng∗1, Chuan Wang∗2, Daniel Gildea1 and Nianwen Xue2 1University of Rochester {xpeng, gildea}@cs.rochester.edu 2Brandeis University {cwang24, xuen}@brandeis.edu Abstract Neural attention models have achieved great success in different NLP tasks. How- ever, they have not fulfilled their promise on the AMR parsing task due to the data sparsity issue. In this paper, we de- scribe a sequence-to-sequence model for AMR parsing and present different ways to tackle the data sparsity problem. We show that our methods achieve significant improvement over a baseline neural atten- tion model and our results are also compet- itive against state-of-the-art systems that do not use extra linguistic resources. 1 Introduction Abstract Meaning Representation (AMR) (Ba- narescu et al., 2013) is a semantic formalism where the meaning of a sentence is encoded as a rooted, directed graph. Figure 1 shows an exam- ple of an AMR in which the nodes represent the AMR concepts and the edges represent the rela- tions between the concepts they connect. AMR concepts consist of predicate senses, named entity annotations, and in some cases, simply lemmas of English words. AMR relations consist of core se- mantic roles drawn from the Propbank (Palmer et al., 2005) as well as very fine-grained semantic re- lations defined specifically for AMR. These prop- erties render the AMR representation useful in ap- plications like question answering and semantics- based machine translation. The task of AMR graph parsing is to map nat- ural language strings to AMR semantic graphs. Recently, a sizable new corpus of English/AMR pairs (LDC2015E86) has been released. Different parsers have been developed to tackle this prob- lem (Flanigan et al., 2014; Wang et al., 2015b; *Both authors contribute equally. describe-01 person genius ARG0 ARG1 ARG2 name “Ryan” name op1 Figure 1: An example of AMR graph representing the meaning of: “Ryan’s description of himself: a genius.” Artzi et al., 2015; Pust et al., 2015; Peng et al., 2015). Most of these parsers have used external re- sources such as dependency parses, semantic lexi- cons, etc., to tackle the sparsity issue. Recently, Sutskever et al. (2014) introduced a neural network model for solving the general sequence-to-sequence problem, and Bahdanau et al. (2014) proposed a related model with an atten- tion mechanism that is capable of handling long sequences. Both models achieve state-of-the-art results on large scale machine translation tasks. However, sequence-to-sequence models mostly work well for large scale parallel data, usually in- volving millions of sentence pairs. Vinyals et al. (2015) present a method which linearizes parse trees into a sequence structure and therefore a sequence-to-sequence model can be applied to the constituent parsing task. Competitive results have been achieved with an attention model on the Penn Treebank dataset, with only 40K annotated sen- tences. AMR parsing is a much harder task in that …

Content truncated to 500 words