Which model should I use for machine translation?

[maurus56]

❓ Questions & Help

I’m interested in training a model for translating articles from Spanish to English. There is too little information (Tutorials) about MT, should I use BERT, XLM or any other one? Also could you explain how to train the proposed model feeding the data, and output the predicted translation.
And is there a way to use XLNet so when translating chapters of a book it can remember the context of the previous ones and better translate?

There is even a model by Microsoft (MASS) that looks simple to use, would you recommend it?

[thomwolf]

Hi I recommend using XLM from Facebook for MT currently: https://github.com/facebookresearch/XLM
We may add some models for MT in the mid-term though.

[Bachstelze]

MASS reports higher BLEU-scores than XLM which is good in pretraining an encoder, but lacks in the training description of the decoder. So we could try to extend the XML-R #1769 encoder with MASS.

[Bachstelze]

In which languages and domains are you interested?

[maurus56]

I'm looking for translation mainly of Spanish and Chinese to English, mainly books and articles so maintaining an overall consistency of the terms and words is crucial.

Just a single way translation is enough and also it should be possible to further train the model on the already translated works.

[mahlettaye]

Please help me on how to use xmlr for summarization? Also if there is any example based on xmlr.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.