MultiMET: A Multimodal Dataset for Metaphor Understanding

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 283–292 August 1–6, 2021. ©2021 Association for Computational Linguistics 283 A Bidirectional Transformer Based Alignment Model for Unsupervised Word Alignment Jingyi Zhang1 and Josef van Genabith1,2 1German Research Center for Artificial Intelligence (DFKI), Saarland Informatics Campus, Saarbr¨ucken, Germany 2Department of Language Science and Technology, Saarland University, Saarland Informatics Campus, Saarbr¨ucken, Germany Jingyi.Zhang@dfki.de,Josef.Van Genabith@dfki.de Abstract Word alignment and machine translation are two closely related tasks. Neural transla- tion models, such as RNN-based and Trans- former models, employ a target-to-source at- tention mechanism which can provide rough word alignments, but with a rather low accu- racy. High-quality word alignment can help neural machine translation in many different ways, such as missing word detection, anno- tation transfer and lexicon injection. Existing methods for learning word alignment include statistical word aligners (e.g. GIZA++) and re- cently neural word alignment models. This pa- per presents a bidirectional Transformer based alignment (BTBA) model for unsupervised learning of the word alignment task. Our BTBA model predicts the current target word by attending the source context and both left- side and right-side target context to produce accurate target-to-source attention (alignment). We further fine-tune the target-to-source atten- tion in the BTBA model to obtain better align- ments using a full context based optimization method and self-supervised training. We test our method on three word alignment tasks and show that our method outperforms both previ- ous neural word alignment approaches and the popular statistical word aligner GIZA++. 1 Introduction Neural machine translation (NMT) (Bahdanau et al., 2014; Vaswani et al., 2017) achieves state- of-the-art results for various translation tasks (Bar- rault et al., 2019, 2020). Neural translation models, such as RNN-based (Bahdanau et al., 2014) and Transformer (Vaswani et al., 2017) models, gen- erally have an encoder-decoder structure with a target-to-source attention mechanism. The target- to-source attention in NMT can provide rough word alignments but with a rather low accuracy (Koehn and Knowles, 2017). High-quality word alignment can be used to help NMT in many different ways, such as detecting source words that are missing in the translation (Lei et al., 2019), integrating an external lexicon into NMT to improve translation for domain-specific terminology or low-frequency words (Chatterjee et al., 2017; Chen et al., 2020), transferring word-level annotations (e.g. under- line and hyperlink) from source to target for docu- ment/webpage translation (M¨uller, 2017). A number of approaches have been proposed to learn the word alignment task, including both statis- tical models (Brown et al., 1993) and recently neu- ral models (Zenkel et al., 2019; Garg et al., 2019; Zenkel et al., 2020; Chen et al., 2020; Stengel- Eskin et al., 2019; Nagata et al., 2020). The pop- ular word alignment tool GIZA++ (Och and Ney, 2003) is based on statistical IBM models (Brown et al., 1993) which learn the word alignment task through unsupervised learning and do not require gold alignments from humans as training data. As …

Content truncated to 500 words