Translating Similar Languages: Role of Mutual Intelligibility in Multilingual Transformers

We investigate different approaches to translate between similar languages\nunder low resource conditions, as part of our contribution to the WMT 2020\nSimilar Languages Translation Shared Task. We submitted Transformer-based\nbilingual and multilingual systems for all language pairs, in the two\ndirections. We also leverage back-translation for one of the language pairs,\nacquiring an improvement of more than 3 BLEU points. We interpret our results\nin light of the degree of mutual intelligibility (based on Jaccard similarity)\nbetween each pair, finding a positive correlation between mutual\nintelligibility and model performance. Our Spanish-Catalan model has the best\nperformance of all the five language pairs. Except for the case of\nHindi-Marathi, our bilingual models achieve better performance than the\nmultilingual models on all pairs.\n

Paper

Similar papers

© 2026 NYSGPT2525 LLC