The Source-Target Domain Mismatch Problem in Machine Translation

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 1519–1533 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 1519 The Source-Target Domain Mismatch Problem in Machine Translation Jiajun Shen⋄† Peng-Jen Chen⋄† Matt Le† Junxian He•∗ Jiatao Gu† Myle Ott† Michael Auli† Marc’Aurelio Ranzato† †Facebook AI Research •Carnergie Mellon University {jiajunshen,pipibjc,mattle,jgu}@fb.com {myleott,michaelauli,ranzato}@fb.com junxianh@cs.cmu.edu Abstract While we live in an increasingly intercon- nected world, different places still exhibit strik- ingly different cultures and many events we experience in our every day life pertain only to the specific place we live in. As a result, people often talk about different things in dif- ferent parts of the world. In this work we study the effect of local context in machine translation and postulate that this causes the domains of the source and target language to greatly mismatch. We first formalize the con- cept of source-target domain mismatch, pro- pose a metric to quantify it, and provide em- pirical evidence for its existence. We con- clude with an empirical study of how source- target domain mismatch affects training of machine translation systems on low resource languages. While this may severely affect back-translation, the degradation can be allevi- ated by combining back-translation with self- training and by increasing the amount of target side monolingual data. 1 Introduction The use of language greatly varies with the ge- ographic location (Firth, 1935; Johnstone, 2010). Even within places where people speak the same language (Britain, 2013), there is a lot of lexical variability due to change of style and topic distribu- tion, particularly when considering content posted on social media, blogs and news outlets. For in- stance, while a primary topic of discussion between British sport fans is cricket, American sport fans are more likely to discuss other sports such as base- ball (Leech and Fallon, 1992). The effect of local context in the use of language is even more extreme when considering regions where different languages are spoken. Despite the ∗Work done during an internship at Facebook AI Research. ⋄Equal contribution. increasingly interconnected world we live in, peo- ple in different places tend to talk about different things. There are several reasons for this, from cultural differences due to geographic separation and history, to the local nature of many events we experience in our every day life. This phenomenon has not only interesting socio- linguistic aspects but it has also strong implica- tions in machine translation (MT) (Bernardini and Zanettin, 2004). In particular, machine translation aims at automatically translating content in two languages that are often spoken in very distant ge- ographic locations by people with rather different cultures. As of today, most MT research has been based on the often implicit assumption that content in the two languages is comparable. Sentences com- prising the parallel dataset used for training are assumed to cover the same topic distribution, re- gardless of the originating language. Even when there exist a mismatch between the source and the target domain, the …

Content truncated to 500 words