Transition-Based Deep Input Linearization

Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers, pages 643–654, Valencia, Spain, April 3-7, 2017. c⃝2017 Association for Computational Linguistics Transition-Based Deep Input Linearization Ratish Puduppully †∗, Yue Zhang ‡ and Manish Shrivastava † †Kohli Center on Intelligent Systems (KCIS), International Institute of Information Technology, Hyderabad (IIIT Hyderabad) ‡Singapore University of Technology and Design ratish.surendran@research.iiit.ac.in, yue zhang@sutd.edu.sg, m.shrivastava@iiit.ac.in Abstract Traditional methods for deep NLG adopt pipeline approaches comprising stages such as constructing syntactic input, pre- dicting function words, linearizing the syntactic input and generating the sur- face forms. Though easier to visual- ize, pipeline approaches suffer from er- ror propagation. In addition, informa- tion available across modules cannot be leveraged by all modules. We construct a transition-based model to jointly per- form linearization, function word predic- tion and morphological generation, which considerably improves upon the accuracy compared to a pipelined baseline system. On a standard deep input linearization shared task, our system achieves the best results reported so far. 1 Introduction Natural language generation (NLG) (Reiter and Dale, 1997; White, 2004) aims to synthesize nat- ural language text given input syntactic, seman- tic or logical representations. It has been shown useful in various tasks in NLP, including machine translation (Chang and Toutanova, 2007; Zhang et al., 2014), abstractive summarization (Barzilay and McKeown, 2005) and grammatical error cor- rection (Lee and Seneff, 2006). A line of traditional methods treat the problem as a pipeline of several independent steps (Bohnet et al., 2010; Wan et al., 2009; Bangalore et al., 2000; H. Oh and I. Rudnicky, 2000; Langkilde and Knight, 1998). For example, shown in Fig- ure 1b, a pipeline based on the meaning text the- ory (MTT) (Melˇcuk, 1988) splits NLG into three ∗Part of the work was done when the author was a vis- iting student at Singapore University of Technology and De- sign. Deep graph Shallow graph String Morph (a) Sem graph Synt tree String Morph (b) Deep graph Morphological String with Syntactic tree (c) Figure 1: Linearization pipelines (a) NLG pipeline with deep input graph, (b) pipeline based on the meaning text theory, (c) this paper. be meanwhile . price think have increase ADV P SBJ VC C-A1 VC A1 A1 AM-TMP Figure 2: Sample deep graph for the sentence: meanwhile, prices are thought to have increased. Note that words are replaced by their lemmas. The function word to and comma are absent in graph. independent steps 1. syntactic generation: gen- erating an unordered and lemma-formed syntactic tree from a semantic graph, introducing function words; 2. syntactic linearization: linearizing the unordered syntactic tree; 3. morphological gener- ation: generating the inflection for each lemma in the string. In this paper we focus on deep graph as input. Exemplified in Figure 2, the deep input type is intended to be an abstract representation of the meaning of a sentence. Unlike semantic input, where the nodes are semantic representations of input, deep input is more surface centric, with lem- 643 …

Content truncated to 500 words