AI4D Yorùbá Machine Translation Challenge
Helping Nigeria
$2 000 USD
Completed (almost 5 years ago)
Machine Translation
683 joined
84 active
Start
Dec 04, 20
Close
May 30, 21
Reveal
May 30, 21
I wrote an article in which I used the data of the competition to test two different models, mT5 and M2M
Notebooks · 28 Jun 2021, 14:12 · edited 1 minute later · 14
I realize that winners in this competition have either used mT5 or M2M models to get the best possible result. For that reason, I wrote this article in which I compared the two models solely trained on the competition dataset (No JW300).
Please take a look: https://medium.com/@abdessalemboukil/comparing-facebooks-m2m-to-mt5-in-low-resources-translation-english-yoruba-ef56624d2b75
Github repo: https://github.com/maroxtn/mt5-M2M-comparison
Congratulations for the winners
@abdessalemboukil Great👍. I will check it out. Thanks.
Thanks, congratulations for the condirmed win
@abdessalemboukil Thanks👍
Well done! I am so amazed of how things are getting easier with simpletransformers. Coding this in pure pytroch will take at least 3 to 4 hours.
Agreed, the funny thing even when I coded it in pytorch, I didn't get the same level of performance as in simpletransformers for some reason
What was the LB score of this mt5 model?
I didn't use JW300 dataset to augment the data for my experiment, so both models were only trained on 10k sentence pairs.
mT5 : 0.3201 , M2M : 0.4013
This makes me believe that M2M has a great potential, and might be the first if trained on the JW300 dataset alongside the competition's dataset.
Amazing! I got my LB score using mT5 pretrained on JW300. I should have tried M2M but I didn't know about it lo!
Thanks for sharing this with the community!
I think nobody used it because there was no clear tutorial on how to use it, that's why I wrote the article 😅. Lets hope the organizers see my post and send me some money hahaha
just to check. @abdessalemboukil are our last comments deleted?
Yes they are, what ?? Apparament moula Zindi "ingéniure" and he got offended lmao #sayno2censorship
that's not good
hahaha, just kidding :D
hahahaha I see my rant went viral then, oh my god social media