Datasets:
At a quick glance it looks like these pairs are misaligned?
The very first Tibetan field doesn't have a corresponding English entry and the following Tibetan fields each match the English field of the previous row. Not sure if this pattern continues.
Very cool stuff you are doing here though! Thanks!
Holy smokes! Good catch, thank you!
Oh, dude, I am sorry, I should have checked more thoroughly. Now the final rows are misaligned in the opposite direction... Not sure how to properly check 100,000+ rows...
Hm, yeah. I'll have to think about this. I can check them programmatically but it'll take some doing. I'll reopen the issue and update things here as I make progress.
I really appreciate your help here!
Ok, I think the issue is resolved. In the re-formatting for Hugging Face, empty lines were dropped before parallel sentences were paired which threw off alignment. There are still problematic entries in the data, but it looks like these are mostly the result of translations not actually being literal line-for-line translations. This a broader issue though that I won't be able to resolve here. Let me know if you spot any other problems, and thanks again!
Awesome, thanks!