Multilingual word translation using auxiliary languages

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 1330–1335, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 1330 Multilingual Word Translation using Auxiliary Languages Hagai Taitelbaum Faculty of Engineering Bar-Ilan University, Israel hagait62@gmail.com Gal Chechik The Gonda Brain Research Center Bar-Ilan University, NVIDIA gal.chechik@biu.ac.il Jacob Goldberger Faculty of Engineering Bar-Ilan University, Israel jacob.goldberger@biu.ac.il Abstract Current multilingual word translation methods are focused on jointly learning mappings from each language to a shared space. The actual translation, however, is still performed as an isolated bilingual task. In this study we pro- pose a multilingual translation procedure that uses all the learned mappings to translate a word from one language to another. For each source word, we first search for the most rel- evant languages. We then use the auxiliary translations to these languages to form an im- proved representation of the source word. Fi- nally, this representation is used for the ac- tual translation to the target language. Exper- iments on a standard multilingual word trans- lation benchmark demonstrate that our model outperforms state of the art results. 1 Introduction Monolingual continuous word embeddings are standard building blocks of many natural language tasks. The embedding spaces can exhibit simi- lar structures across languages. Several studies (Mikolov et al., 2013; Klementiev et al., 2012) proposed to exploit this similarity by learning a linear mapping from a source to a target embed- ding space, and demonstrated this approach on a word translation task. Xing et al. (2015) showed that using orthogonality matrices can significantly improve performance. Bilingual embedding can be extended to a mul- tilingual setup by jointly learning mappings from each monolingual space to a shared word space. In recent years several studies have proposed aligning multiple languages simultaneously in a shared space by enforcing (or at least encour- aging) transitive relations between the mappings (Chen and Cardie, 2018; Kementchedjhieva et al., 2018; Alaux et al., 2019; Jawanpuria et al., 2019; Taitelbaum et al., 2019). Dealing with multiple languages simultaneously has been shown to im- prove performance on some bilingual tasks by us- ing knowledge learned from other languages (Am- mar et al., 2016; Duong et al., 2017). Once the multilingual mappings are learned, one can infer word correspondences for words that are not in the initial lexicons. Previous multilin- gual methods have focused on training procedures that benefit from the multilingual setting. The ac- tual translation, however, is still done as an iso- lated bilingual task. It makes eminent sense how- ever to go beyond this stage and utilize the rela- tions between all the languages at the inference phase as it is done in the mapping learning phase. In this study, we propose a new inference method for multilingual word translation that uses all the learned mappings to translate a word from one language to another. For each source word we first search for the most relevant languages that help translate the source …

Content truncated to 500 words