The University of Edinburgh’s English-Tamil and English-Inuktitut Submissions to the WMT20 News Translation Task

Rachel Bawden (1) , Alexandra Birch (1) , Radina Dobreva (1) , Arturo Oncevay (1) , Antonio Valerio Miceli Barone (1) , Philip Williams (1)

1 School of Informatics [Edimbourg]

We describe the University of Edinburgh's submissions to the WMT20 news translation shared task for the low resource language pair English-Tamil and the mid-resource language pair English-Inuktitut. We use the neural machine translation transformer architecture for all submissions and explore a variety of techniques to improve translation quality to compensate for the lack of parallel training data. For the very low-resource English-Tamil, this involves exploring pretraining, using both language model objectives and translation using an unrelated high-resource language pair (German-English), and iterative backtransla-tion. For English-Inuktitut, we explore the use of multilingual systems, which, despite not being part of the primary submission, would have achieved the best results on the test set.

Mots clés: Machine translation; Tamil; Inuktitut; Shared task

Domaines: Informatique et langage [cs.CL]

Rachel Bawden, Alexandra Birch, Radina Dobreva, Arturo Oncevay, Antonio Valerio Miceli Barone, et al.. The University of Edinburgh’s English-Tamil and English-Inuktitut Submissions to the WMT20 News Translation Task. 5th Conference on Machine Translation, Nov 2020, Online, Unknown Region. ⟨hal-02981153⟩