On the Difficulty of Translating Free-Order Case-Marking Languages

Abstract Identifying factors that make certain languages harder to model than others is essential reach language equality in future Natural Language Processing technologies. Free-order case-marking languages, such as Russian, Latin, or Tamil, have proved more challenging fixed-order for the tasks of syntactic parsing and subject-verb agreement prediction. In this work, we investigate whether class also difficult translate by state-of-the-art Neural Machine Translation (NMT) models. Using a variety synthetic newly introduced translation challenge set, find word order flexibility source only leads very small loss NMT quality, even though core verb arguments become impossible disambiguate sentences without semantic cues. The latter issue indeed solved addition case marking. However, medium- low-resource settings, overall quality remains unmatched.