JParaCrawl v3.0: A Large-scale English-Japanese Parallel Corpus

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6704–6710 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6704 JParaCrawl v3.0: A Large-scale English-Japanese Parallel Corpus Makoto Morishita, Katsuki Chousa, Jun Suzuki, and Masaaki Nagata NTT Communication Science Laboratories, NTT Corporation 2-4 Hikaridai, Seika-cho, Soraku-gun, Kyoto, 619-0237, Japan {makoto.morishita.gr, katsuki.chousa.bg, masaaki.nagata.et}@hco.ntt.co.jp jun.suzuki@tohoku.ac.jp Abstract Most current machine translation models are mainly trained with parallel corpora, and their translation accuracy largely depends on the quality and quantity of the corpora. Although there are billions of parallel sentences for a few language pairs, effectively dealing with most language pairs is difficult due to a lack of publicly available parallel corpora. This paper creates a large parallel corpus for English-Japanese, a language pair for which only limited resources are available, compared to such resource-rich languages as English-German. It introduces a new web-based English-Japanese parallel corpus named JParaCrawl v3.0. Our new corpus contains more than 21 million unique parallel sentence pairs, which is more than twice as many as the previous JParaCrawl v2.0 corpus. Through experiments, we empirically show how our new corpus boosts the accuracy of machine translation models on various domains. The JParaCrawl v3.0 corpus will eventually be publicly available online for research purposes. Keywords: parallel corpus, machine translation, English, Japanese 1. Introduction The current neural machine translation models are gen- erally trained by supervised approaches (Sutskever et al., 2014; Bahdanau et al., 2015; Luong et al., 2015; Vaswani et al., 2017), denoting reliance on parallel cor- pora. However, since publicly available parallel corpora remain limited, training a model for many language pairs is difficult. Thus, constructing a parallel corpus is crucial for expanding the applicability of machine translation. This paper introduces a new large-scale web-based par- allel corpus for English-Japanese for which only lim- ited parallel corpora are available. One of the cur- rent largest parallel corpora for this language pair is JParaCrawl (Morishita et al., 2020), which is con- structed by crawling the web and automatically align- ing parallel sentences. However, this corpus contains around 10 million sentence pairs, which is still limited compared to the other resource-rich language pairs, and it is somewhat outdated because it was created two years ago. We entirely re-crawled the web to update the cor- pus and applied a different approach to extract parallel sentences. We collected more than 21 million unique sentence pairs, which is more than twice as many as the previous JParaCrawl corpus. We experimentally show how the new crawled corpus increases the accuracy of machine translation for English-Japanese and Japanese- English. Our new corpus, named JParaCrawl v3.0, will be publicly available through our website1 for further researches. Our contributions can be summarized: • We constructed a large-scale English-Japanese par- allel corpus, which contains more than 21 million 1http://www.kecl.ntt.co.jp/icl/lirg/ jparacrawl/ sentence pairs, on top of the previous JParaCrawl corpus. • We empirically confirmed that our corpus boosted the accuracy of English-Japanese and Japanese- English machine translation in broad domains. • We plan to …

Content truncated to 500 words