Datasets:

AmazonScience

/

massive

like 50

Thanks for pointing this out. Sorry I reverted the changes back to the original for now because after the merge I realize that zh-CN and zh-TW are only represented a single time. Can you tell me where the referencing happens and how to handle languages where the xx is the same? Thanks.

Hi :)

The Hub PR functionality does not currently allow images to be included in responses so I will try to describe as best I can without images. Feel free to bounce back and forth on what follows if it is not clear.

Currently, if you go to https://huggingface.co/datasets, you can see on the left the different filters that can be applied (Task Categories, Tasks, Languages, Multilinguality, etc.) to find a dataset that meets a criterion we are interested in.
In the case of languages, you can see that the tags available for the filter are only based on letters (what I call xx). For example, if I apply the tag "fr" (https://huggingface.co/datasets?languages=languages:fr&sort=downloads), I find 137 data sets... but not yours.
I can't find it because you have set the filter to "fr-FR". Which is sad because your dataset is very interesting for French speakers (and more generally for all the other languages it takes into account).
So that's why I added for example the tag "fr" to your dataset via my PR.
Note however that I added "fr" but did not remove "fr-FR" because it is super important information in the case of different language varieties (you cite also the example of Chinese).
I am in favour of being able to filter on the different varieties of the same language (zh-CN and zh-TW for example). At https://huggingface.co/languages, for example, it would be possible to sort like this : https://user-images.githubusercontent.com/58078086/134772903-74d3221c-d73a-4fa5-99f2-894b085dfa66.png.
For the moment this is not possible (I intend to open a PR or at least an issue to try to get something similar from the HF team).

And so in the meantime, in order to have a better reference and for your dataset to be visible to everyone with the current tag system, I suggest you add the "xx" tags but not delete the "xx-XX" tags you already had filled in as they will be used one day.

For the moment this is not possible (I intend to open a PR or at least an issue to try to get something similar from the HF team).

Thanks for the suggestion, not sure how to do it UX-wise but it's a good idea

And yes @cperiz I agree with @lbourdois ' suggestion here (you can add the two-letter codes in addition to the more precise ones)

@lbourdois thank you for the clear explanation! :) That makes sense. I agree that it would be good to have the ability to filter on the different varieties of the same language (especially as we progress with datasets larger varieties of languages). @julien-c thank you for the confirmation. I updated the README again to contain what @lbourdois suggested in his PR. Thanks!