crash-sv/scribe-translate-ct2

Opus-MT tc-big for Scribe SV (CTranslate2)

CTranslate2 conversions of two Helsinki-NLP Opus-MT tc-big models. They are used by Scribe SV, a Windows dictation and translation utility, and are published here so the application can download them on demand instead of shipping them inside the installer.

Nothing was retrained or fine-tuned: these are format conversions of the original models, quantized to int8_float16 with ctranslate2-converter.

Contents

folder	direction	source model
opus-mt-tc-big-zle-en-ct2	East Slavic (ru, uk, be) to English	Helsinki-NLP/opus-mt-tc-big-zle-en
opus-mt-tc-big-en-zle-ct2	English to East Slavic (ru, uk, be)	Helsinki-NLP/opus-mt-tc-big-en-zle

Each folder holds a complete CTranslate2 model: model.bin, config.json, shared_vocabulary.json, the SentencePiece models (source.spm, target.spm) and the tokenizer files.

Language token

opus-mt-tc-big-en-zle is a multi-target model. When translating into East Slavic, the source sentence must start with a target-language token, e.g. >>rus<< for Russian:

>>rus<< The build failed.

The other direction needs no token.

Attribution and license

Original models by the Helsinki-NLP group (Language Technology Research Group at the University of Helsinki), released under CC-BY-4.0. These conversions keep the same license, and attribution to Helsinki-NLP is required when using them.

• Tiedemann, J., Thottingal, S. OPUS-MT — Building open translation services for the World. EAMT 2020.
• Tiedemann, J. The Tatoeba Translation Challenge — Realistic Data Sets for Low Resource and Multilingual MT. WMT 2020.