The Opus project provides translation models for many languages. This task identifies languages not supported by other translation services, not even those potentially supported by NLLB-200 (T326578). In addition, it also includes languages for which feedback suggests that Opus could significantly improve the translation quality compared to other options available.
This ticket proposes to provide support for the following languages (and specific pairs):
Language | Pair | BLEU | Status | Notes |
---|---|---|---|---|
Central Bikol (bcl) | en – bcl | 31.9 | ✅ | Enabled as part of T331836: Support multiple MT models in self hosted machine translation service. |
tl - bcl | ❌ | Model not available | ||
Cantonese (zh-yue) | zh – zh-yue | ❌ | Model not available NLLB-200 support is not valid for the language based on T333835: Disable machine translation for Cantonese. We may want to check whether Opus support is useful. | |
en – zh-yue | ❌ | Model not available. | ||
Moroccan arabic (ary) | en – ary | ❌ | Model not available NLLB-200 support is not valid for the language based on T339926: The NLLB-200 MT engine in MinT returns standard Arabic translation instead of Moroccan Darija in Moroccan Arabic Wikipedia. We may want to check whether Opus support is useful. | |
ar – ary | ❌ | Model not available. | ||
Gun (guw) | en – guw | 45.7 | ✅ | |
Cherokee (chr) | en – chr | 44.6 | ✅ | |
Sranan Tongo (srn) | en – srn | 34.6 | ✅ | |
Venda (ve) | en – ve | 40.5 | ✅ | |
Tahitian (ty) | en – ty | 46.8 | ✅ | |
fr – ty | 39.6 | ✅ | ||
Bislama (bi) | en – bi | 37.1 | ✅ | |
th – bi | ❌ | Model not available | ||
Tongan (to) | en – to | 59.1 | ✅ | |
Manx (gv) | pt – gv | ❌ | Model not available | |
en – gv | 70.1 | ❌ | Poor quality | |
Walloon (wa) | en - wa | 33.4 | ❌ | Sentence segmenter using old file format; Not possible to use this model as of now. |
fr - wa | ❌ | Model not available | ||
Western Frisian (fy) | nl - fy | ❌ | Model not available | |
en - fy | ❌ | Model not available | ||
Breton (br) | en - fr - br | ✅ | Available now. | |
fr - br | ❌ | Model not available | ||
Finnish (fi) | en - fi | 25.7 | ✅ | Supported by other services already, but feedback suggests Opus may improve quality). Low BLEU score. |
sv - fi | 45.2 | ✅ | ||