Compositional Phrase Alignment and Beyond

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 1611–1623, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 1611 Compositional Phrase Alignment and Beyond 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 Phrase alignment is the basis for modelling sentence pair interactions, such as paraphrase and textual entailment recognition. Most phrase alignments are compositional processes such that an alignment of a phrase pair is con- structed based on the alignments of their child phrases. Nonetheless, studies have revealed that non-compositional alignments involving long-distance phrase reordering are prevalent in practice. We address the phrase alignment problem by combining an unordered tree map- ping algorithm and phrase representation mod- elling that explicitly embeds the similarity dis- tribution in the sentences onto powerful con- textualized representations. Experimental re- sults demonstrate that our method effectively handles compositional and non-compositional global phrase alignments. Our method sig- nificantly outperforms that used in a previous study and achieves a performance competitive with that of experienced human annotators. 1 Introduction Phrase alignment is a fundamental problem in mod- elling the interactions between a pair of sentences, such as paraphrase identification, textual entail- ment recognition, and question answering (Das and Smith, 2009; Heilman and Smith, 2010; Wang and Manning, 2010). Phrase alignment generally adheres to compositionality, in which a phrase pair is aligned based on the alignments of their child phrases. Nonetheless, non-compositional align- ments involving long-distance phrase reordering are prevalent in practice (Burkett et al., 2010; Heil- man and Smith, 2010; Arase and Tsujii, 2017). Fig- ure 1 shows an example of phrase alignment in which phrases of the same colours are alignable, i.e. they are phrasal paraphrases. The alignment of ‘an- tivirus vaccines’ and ‘vaccines against the virus’ is compositional, as supported by alignments of their child nodes although their orderings are reversed. team spirit ,the research group developed the antivirus vaccine Relying on The scientific team created a vaccin Relying on team spirit , the research group developedantivirus vaccines The scientific team created vaccines against the virus through teamwork 𝜏𝜏1 𝑠𝑠 𝜏𝜏1 𝑡𝑡 𝜏𝜏2 𝑠𝑠 𝜏𝜏2 𝑡𝑡 lca(𝜏𝜏1 𝑡𝑡, 𝜏𝜏2 𝑡𝑡) 𝜏𝜏3 𝑠𝑠=lca(𝜏𝜏1 𝑠𝑠, 𝜏𝜏2 𝑠𝑠) 𝜏𝜏3 𝑡𝑡 Figure 1: Phrase alignments by the proposed method (phrases of the same colour are paraphrases) Similarly, the alignment of their parents τ s 2 and τ t is compositional. By contrast, the alignment of τ s2 1 and τ t 1 is non-compositional in relation to the align- ment of τ s 2 and τ t 2; although τ t 1 and τ t 2 are siblings, τ s 1 is not a sibling of τ s 2, i.e. not in the scope of the parent node of τ s 2. To treat such a long-distance correspondence in non-compositional alignment, one has to consider candidate phrases outside the local scope and potentially the entire sentence. In this study, we address the phrase alignment problem by …

Content truncated to 500 words