AfroTranslate 0.0.6

This package allows you to obtain translations from Masakhane JoeyNMT based models. Masakhane is a grassroots research community aiming to revive and strengthen African languages through AI.

AfroTranslate

This package allows you to obtain translations from Masakhane JoeyNMT based models with very few lines of code. Masakhane is a grassroots research community aiming to revive and strengthen African languages through AI.

Available models can be found here.

Note: Please, install the cuda supported version of pytorch to use the GPU. Ex: pip install torch==1.8.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html.

Here are some useful links:

Masakhane: Visit the Masakhane home page.

Masakhane web: Get translations using the Masakhane web application.

Masakhane Github: Take a look at the community main projects here!

JoeyNMT Github: Get hands on JoeyNMT here!

Test it right now on Colab!

Installation

pip install AfroTranslate

Example:

Translation using the English to Fon model

from afrotranslate import MasakhaneTranslate

translator = MasakhaneTranslate(model_name="en-fon")

translator.translate("I love you so much!", n_best=1)

'Un yí wǎn nú we tawun'

Loading a model from specified directory

translator = MasakhaneTranslate(model_path="<directory-where-your-model-resides>")

Translating several sentences at once

The models are trained on individual sentences, so we automatically detect sentence boundaries in inputs and translate them separately. The output shows alternatives for each of them.

translator.translate("I love you so much! Our love is very strong!", n_best=1)

{'Sentence1': 'Un yí wǎn nú we tawun',
 'Sentence2': 'Wanyiyi mǐtɔn ɖò taji tawun'}

Disclaimer

This is a community research project and as such, this service is not a production system. The models are only trained using religious data. Therefore, it should not be used for official translations.

Acknowledgement

I want to thank Julia Kreutzer for her precious feedback on this work.