Amazing work! Small request:

#3
by orendar - opened

Hey, wanted to say thank you for doing such an incredible work for the benefit of the community.

Going over some of the data, I noticed that sometimes some of the image captions find their way into the text (sometimes in the beginning and sometimes in the middle, before/after paragraphs).

For example, let's compare "Alan Turing" in wikipedia simple between the 2023 "official" dump and your updated version (Can be found here https://huggingface.co/datasets/wikimedia/wikipedia/viewer/20231101.simple?views%5B%5D=_20231101simple&row=6 and here https://huggingface.co/datasets/omarkamali/wikipedia-monthly/viewer/20250801.simple?views%5B%5D=_20250801simple_train&row=6), we can see that there is a caption artifact at the beginning of your version not present in the older "official" version. This caption can be clearly seen in the actual wikipedia article here (https://simple.wikipedia.org/wiki/Alan_Turing).

Do you by any chance have some idea on how to solve or mitigate this issue? I am assuming it is related to the processing pipeline?

Thank you again for all your amazing work!

Hey @orendar , thanks for the report!

I noticed the parsing issue as well. As I am using mwparserfromhell, this is either a misconfiguration on my side or a bug in the library. I am collecting test cases like the one you shared and see what I can do.

Sign up or log in to comment