This model generates translations from English to colloquial Arabic, conditioning the translations on dialect. Supported dialects are: Palestinian/Jordanian, Syrian, Lebanese and Egyptian. Dialect is specified via the first input token which should be either P/S/L/E.

For example, to translate to Palestinian/Jordanian, use the input:

And for Syrian:

The model was trained by fine-tuning the opus-mt-ar-en (MSA to English) model on ~85K parallel sentences in four dialects of colloquial Arabic.