CharLevelMT

This repository will contain the results from Are Character-level Translations Worth the Wait? Comparing ByT5 and mT5 for Machine Translation, presented at EACL2024, to be published in the next edition of TACL.

Models

The models used in this work can be found at: https://huggingface.co/leukas

All of the models have a name as follows: {model_name}-{dataset}-{amount_of_data}-{lang_pair}

• model_name = byt5 or mt5
• dataset = wmt14 or nc16
• amount_of_data = 400, 2k, 10k, 50k, 250k, 1250k, or not specified for full wmt14
• lang_pair = ende, deen, enru, ruen, enes, ptes

Not all combinations exist, for more details check out the paper.

To use the models properly, be sure to start with the prompt "translate X to Y:" e.g. "translate English to German:"

Outputs

Outputs and scores will be added soon.

Questions

If you have any questions don't hesitate to contact me (preferably via email).