UCAM Biomedical translation at WMT19: Transfer learning multi-domain ensembles
This work addresses biomedical translation for researchers and practitioners, but it is incremental as it builds on existing ensemble and transfer learning methods.
The authors tackled the WMT19 Biomedical translation task by using transfer learning to create multi-domain ensembles of neural models, achieving the best submitted results for English-Spanish translation.
The 2019 WMT Biomedical translation task involved translating Medline abstracts. We approached this using transfer learning to obtain a series of strong neural models on distinct domains, and combining them into multi-domain ensembles. We further experiment with an adaptive language-model ensemble weighting scheme. Our submission achieved the best submitted results on both directions of English-Spanish.