Not Low-Resource Anymore: Aligner Ensembling, Batch Filtering, and New Datasets for Bengali-English Machine Translation

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 2612–2623, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 2612 Not Low-Resource Anymore: Aligner Ensembling, Batch Filtering, and New Datasets for Bengali-English Machine Translation Tahmid Hasan∗, Abhik Bhattacharjee∗, Kazi Samin, Masum Hasan, Madhusudan Basak, M. Sohel Rahman and Rifat Shahriyar Bangladesh University of Engineering and Technology (BUET) {tahmidhasan, madhusudan, msrahman, rifat}@cse.buet.ac.bd, {abhik, samin, masum}@ra.cse.buet.ac.bd Abstract Despite being the seventh most widely spoken language in the world, Bengali has received much less attention in machine translation lit- erature due to being low in resources. Most publicly available parallel corpora for Bengali are not large enough; and have rather poor quality, mostly because of incorrect sentence alignments resulting from erroneous sentence segmentation, and also because of a high vol- ume of noise present in them. In this work, we build a customized sentence segmenter for Bengali and propose two novel methods for parallel corpus creation on low-resource setups: aligner ensembling and batch filter- ing. With the segmenter and the two methods combined, we compile a high-quality Bengali- English parallel corpus comprising of 2.75 mil- lion sentence pairs, more than 2 million of which were not available before. Training on neural models, we achieve an improvement of more than 9 BLEU score over previous ap- proaches to Bengali-English machine transla- tion. We also evaluate on a new test set of 1000 pairs made with extensive quality control. We release the segmenter, parallel corpus, and the evaluation set, thus elevating Bengali from its low-resource status. To the best of our knowl- edge, this is the first ever large scale study on Bengali-English machine translation. We be- lieve our study will pave the way for future research on Bengali-English machine transla- tion as well as other low-resource languages. Our data and code are available at https: //github.com/csebuetnlp/banglanmt. 1 Introduction Recent advances in deep learning (Bahdanau et al., 2015; Wu et al., 2016; Vaswani et al., 2017) have aided in the development of neural machine trans- lation (NMT) models to achieve state-of-the-art results in several language pairs. But a large num- ber of high-quality sentence pairs must be fed into ∗These authors contributed equally to this work. these models to train them effectively (Koehn and Knowles, 2017); and in fact lack of such a cor- pus affects the performance thereof severely. Al- though there have been efforts to improve machine translation in low-resource contexts, particularly using, for example, comparable corpora (Irvine and Callison-Burch, 2013), small parallel corpora (Gu et al., 2018) or zero-shot multilingual translation (Johnson et al., 2017), such languages are yet to achieve noteworthy results (Koehn et al., 2019) compared to high-resource ones. Unfortunately, Bengali, the seventh (fifth) most widely spoken language in the world by the number of (native1) speakers,2 has still remained a low-resource lan- guage. As of now, only a few parallel corpora for Bengali language are publicly available (Tiede- mann, 2012) and those too suffer from poor sen- tence segmentation, resulting in poor alignments. They also contain …

Content truncated to 500 words