Tilde at WMT 2020: News Task Systems
This is an incremental improvement for machine translation systems in a specific competition setting.
The paper describes Tilde's submission to the WMT2020 news translation task for English-Polish, building on previous years with Transformer models and experimenting with data selection and back-translation, resulting in final ensemble models with re-ranking.
This paper describes Tilde's submission to the WMT2020 shared task on news translation for both directions of the English-Polish language pair in both the constrained and the unconstrained tracks. We follow our submissions from the previous years and build our baseline systems to be morphologically motivated sub-word unit-based Transformer base models that we train using the Marian machine translation toolkit. Additionally, we experiment with different parallel and monolingual data selection schemes, as well as sampled back-translation. Our final models are ensembles of Transformer base and Transformer big models that feature right-to-left re-ranking.