Translation
Languages and regional variants
DeepReference supports 29 languages and 13 regional variants, with automatic source detection. Three languages are in beta. Variants are selected on demand when the difference matters.
DeepReference supports 29 languages and 13 regional variants: six Arabic dialects (Egyptian, Gulf, Levantine, Saudi, Moroccan, Tunisian), Spanish for Spain, Mexico and Argentina, European and Brazilian Portuguese, and French for France and Canada. You can choose both languages or leave the source on automatic detection and select only the target.
Languages in beta
Three languages are marked BETA in the language pickers: Telugu, Tamil and Hebrew. They work like every other language, with the same features and the same word count, but the model has seen less material in them than in the others, so expect more corrections. Have a fluent reviewer check the result before you deliver it.
The label is removed from a language once its quality matches the rest.
Regional variants
Some languages have regional variants where vocabulary and conventions differ: Brazilian vs. European Portuguese, Simplified vs. Traditional Chinese, Canadian vs. European French, and the Arabic dialects. The main language list stays uncluttered: variants don't appear until you ask for them. When you need one, select it and the model translates natively into that variant. This is handled by the model itself, not by a find-and-replace layer.
Why variants are selected on demand
The main picker shows the standard language choices first. Regional variants appear when you open the variant control, which keeps the list shorter without removing the distinction when a project requires it. Decimal separators, quotation marks, and punctuation spacing are not converted automatically; see Reviewing locale conventions.
Auto-detect works per request, so you can paste French one minute and Japanese the next without touching the source selector.
Keep exploring
Text translation
Translate text from the dashboard with language, tone, glossary, and Translation Memory controls. Input accepts up to 100,000 characters, subject to the model token budget.
Reviewing locale conventions
Review number formats, quotation marks, and punctuation against the target language and client requirements.
Domains
A domain steers terminology extraction toward the vocabulary of your field. It applies to termbase work, not to plain translation.