CUNI Systems for the Unsupervised and Very Low Resource Translation Task in WMT20

• DOI: 10.48550/arxiv.2010.11747
• Publication Date: 2020-01-01
• Authors: Ivana Kvapilíková; Tom Kocmi; Ondřej Bojar
• Keywords: FOS: Computer and information sciences; Computer Science - Computation and Language; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; Computation and Language (cs.CL)

This paper presents a description of CUNI systems submitted to the WMT20 task on unsupervised and very low-resource supervised machine translation between German Upper Sorbian. We experimented with training synthetic data pre-training related language pair. In fully scenario, we achieved 25.5 23.7 BLEU translating from into Sorbian, respectively. Our relied transfer learning German-Czech parallel 57.4 56.1 BLEU, which is an improvement 10 points over baseline trained only available small German-Upper Sorbian corpus.