metadata

tags:
  - Levantine Arabic
  - Shami
  - English
  - Egyptian

English2Dialect

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:

P What time is it now?

And for Syrian:

S What time is it now?

Demo: https://huggingface.co/spaces/guymorlan/English2Shami
Version w/o conditional generation (Levantine only): https://huggingface.co/guymorlan/English2Shami

Training Data

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.

Model Description

Developed by: Guy Mor-Lan (guy.mor@mail.huji.ac.il)
Model type: MarianMT Seq2Seq
License: MIT
Finetuned from model: Helsinki-NLP/opus-mt-ar-en[https://huggingface.co/Helsinki-NLP/opus-mt-ar-en]