SAPPHIRE: Simple Aligner for Phrasal Paraphrase with Hierarchical Representation

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6861–6867 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6861 SAPPHIRE: Simple Aligner for Phrasal Paraphrase with Hierarchical Representation Masato Yoshinaka†, Tomoyuki Kajiwara‡, Yuki Arase† †Graduate School of Information Science and Technology, Osaka University ‡Institute for Datability Science, Osaka University {yoshinaka.masato, arase}@ist.osaka-u.ac.jp, kajiwara@ids.osaka-u.ac.jp Abstract We present SAPPHIRE, a Simple Aligner for Phrasal Paraphrase with HIerarchical REpresentation. Monolingual phrase alignment is a fundamental problem in natural language understanding and also a crucial technique in various applications such as natural language inference and semantic textual similarity assessment. Previous methods for monolingual phrase alignment are language-resource intensive; they require large-scale synonym/paraphrase lexica and high-quality parsers. Different from them, SAPPHIRE depends only on a monolingual corpus to train word embeddings. Therefore, it is easily transferable to specific domains and different languages. Specifically, SAPPHIRE first obtains word alignments using pre-trained word embeddings and then expands them to phrase alignments by bilingual phrase extraction methods. To estimate the likelihood of phrase alignments, SAPPHIRE uses phrase embeddings that are hierarchically composed of word embeddings. Finally, SAPPHIRE searches for a set of consistent phrase alignments on a lattice of phrase alignment candidates. It achieves search-efficiency by constraining the lattice so that all the paths go through a phrase alignment pair with the highest alignment score. Experimental results using the standard dataset for phrase alignment evaluation show that SAPPHIRE outperforms the previous method and establishes the state-of-the-art performance. Keywords: phrase alignment, phrasal paraphrase 1. Introduction Monolingual phrase alignment is one of the fundamen- tal tasks in natural language understanding. It identifies the most plausible phrase alignments that are semantically equivalent in a monolingual sentence pair. The applications of monolingual phrase alignment are diverse. The most rel- evant application is the sentence pair modeling tasks (Lan and Xu, 2018), such as recognizing textual entailment (Da- gan et al., 2006) and assessing semantic textual similarity (Sultan et al., 2014). Besides, monolingual phrase align- ment is useful to identify parallel sentences automatically from a comparable corpus (Kajiwara and Komachi, 2016) and to generate paraphrases (Li et al., 2019). Previous studies on monolingual phrase alignment (Mac- Cartney et al., 2008; Yao et al., 2013; Arase and Tsu- jii, 2017; Ouyang and McKeown, 2019) depend on large- scale paraphrase dictionaries or assume the availability of high-quality parsers or chunkers, which severely restricts the applicability of alignment methods to corpora of spe- cific domains and different languages. On the other hand, bilingual phrase extraction that has been widely studied in the field of statistical machine translation (SMT) only as- sumes the availability of a large-scale parallel corpus. The standard approach is first identifying word alignment and then composing phrase alignments from the identified word alignments based on heuristics. However, the purpose of bilingual phrase extraction is the collection of a large scale bilingual phrase pairs, and thus, identification of plausible phrase alignments in a single sentence pair is out of their scope. In this study, we propose …

Content truncated to 500 words