Clarity on licenses
Hei, thanks for sharing the data.
I am looking to include some of the datasets here in the Danish Dynaword. A public collection of openly available data.
However some of the license seems incorrectly specified. I was hoping that you could help me find clarity:
- the dataset seems to be cc-by-4.0 licensed but contain data with unknown licenses according to the paper?
- you say in the paper that the Twitter data is licensed under MIT? I can’t seem to find the source of this license though?
An aside from these licensing questions. I would also be more than happy to invite you to collaborate on Dynaword and in an upcoming publication. Let me know if this might be something that you might be interested in.
Best,
Kenneth
Hi Kenneth!
Thanks for the interest.
- Yes, we could only choose one license as metadata, but we can keep it unknown and refer to the paper for licensing.
- For the Twitter data, we just made an assumption that the data has been extracted using a version of the Twitter API which is usually MIT-licensed, to consider it as "an upper-bound". Though, we understand that the software/code is usually MIT-licensed. We are no experts, so feel free to leave the Twitter data out. Other Twitter datasets here on HF seem to have a flavor of cc-by*.
Appreciate the involvement! Happy to collaborate. Feel free to invite the Snakmodel authors: https://arxiv.org/pdf/2412.12956
Best,
Mike
Thanks for the clarification.
- I would keep it unknown, but it is a grey area you could technically keep it cc as you can re-license the collection (the work surrounding the data, metadata, links, filtering) but not the data itself.
- This also somewhat relates here. For e.g., a CC license on HF does not necessarily mean that the data itself is permissibly licensed. For E.g., Dolma is doc-by licensed data, but the data it contains is still copyrighted. Confusing, I know :) This paper gives quite a good introduction to it: https://arxiv.org/abs/2501.08365.
What does this mean for dynaword? As we go for openly licensed data (meeting that the text itself should not be under copyright), we will probably not include e.g. the Twitter data as I don't think the text is MIT (only the things surrounding it). I do think the bookshop data probably works though so will probably take a closer look at that one :)
Re. the invitation; glad to hear that you are interested. I will do an initial write-up of the paper (a minimal paper, if you will) and then reach out to contributors who might wish to contribute. I have already talked with Rob