Monolingual Phrase Alignment on Parse Forests

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 1–11 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Monolingual Phrase Alignment on Parse Forests Yuki Arase1⋆ and Junichi Tsujii⋆2 1Osaka University, Japan ⋆Artificial Intelligence Research Center (AIRC), AIST, Japan 2NaCTeM, School of Computer Science, University of Manchester, UK arase@ist.osaka-u.ac.jp, j-tsujii@aist.go.jp Abstract We propose an efficient method to con- duct phrase alignment on parse forests for paraphrase detection. Unlike previ- ous studies, our method identifies syntac- tic paraphrases under linguistically mo- tivated grammar. In addition, it allows phrases to non-compositionally align to handle paraphrases with non-homographic phrase correspondences. A dataset that provides gold parse trees and their phrase alignments is created. The experimental results confirm that the proposed method conducts highly accurate phrase alignment compared to human performance. 1 Introduction Paraphrase detection is crucial in various applica- tions, which has been actively studied for years. Due to difficulties caused by the non-homographic nature of phrase correspondences, the units of cor- respondence in previous studies are defined as se- quences of words like in (Yao et al., 2013) and not syntactic phrases. On the other hand, syn- tactic structures are important in modeling sen- tences, e.g., their sentiments and semantic simi- larities (Socher et al., 2013; Tai et al., 2015). In this paper, we present an algorithm to align syntactic phrases in a paraphrased pair of sen- tences. We show that (1) the problem of identify- ing a legitimate set of syntactic paraphrases under linguistically motivated grammar is formalized, (2) dynamic programing a la CKY (Cocke, 1969; Kasami, 1965; Younger, 1967) makes phrase alignment computationally feasible, (3) alignment quality of phrases can be improved using n-best parse forests instead of 1-best trees, and (4) non- compositional alignment allows non-homographic correspondences of phrases. Motivated by recent Source: Whenever I go to the ground floor for a smoke, I always come face to face with them. Target: Whenever I go down to smoke a cigarette, I come face to face with one of them. ⋯goto the ground floor for a smoke NP PP NP VP PP VP go smoke a cigarette NP VP VP to CP VP down ⋯ ⋯ ⋯ Figure 1: Example of phrase alignments findings that syntax is important for phrase embed- ding (Socher et al., 2013) in which phrasal para- phrases allow semantic similarity to be replicated (Wieting et al., 2016, 2015), we focus on the syn- tactic paraphrase alignment. Fig. 1 shows a real example of phrase align- ments produced by our method. Alignment pro- ceeds in a bottom-up manner using the compo- sitional nature of phrase alignments. First, word alignments are given. Then, phrase alignments are recursively identified by supporting relations be- tween phrase pairs. Non-compositional alignment is triggered when the compositionality is violated, which is common in paraphrasing. For systematic research on syntactic phrase alignment in paraphrases, we constructed a gold standard dataset of paraphrase sentences with phrase alignment (20, 678 phrases in 201 para- phrasal sentences). This dataset will be …

Content truncated to 500 words