lurauch
commented
10 months ago CGPT on ND
The key point of the NoDerivatives clause is to prevent the creation and distribution of a version of the work that is different in a meaningful or substantial way from the original. The intention is to prevent alterations to the artistic or intellectual content of the work.
In the case of simply changing a file format:
-
If the conversion is purely technical and does not materially alter the content (like changing from .wav to .mp3 without changing the actual audio content), it might not be considered a derivative work in the strict sense.
-
However, because the license is strict about no modifications, it's often safest to assume that even format changes could be seen as creating a derivative work.
JonasLange
raphaelschwinger
licence //creativecommons.org/licenses/by-nc-nd/2.5/ - 67554 //creativecommons.org/licenses/by-nc-nd/3.0/ - 7379 //creativecommons.org/licenses/by-nc-nd/4.0/ - 118984 //creativecommons.org/licenses/by-nc-sa/3.0/ - 68896 //creativecommons.org/licenses/by-nc-sa/3.0/us/ - 1 //creativecommons.org/licenses/by-nc-sa/4.0/ - 415453 //creativecommons.org/licenses/by-nc/4.0/ - 128 //creativecommons.org/licenses/by-sa/3.0/ - 679 //creativecommons.org/licenses/by-sa/4.0/ - 6614 //creativecommons.org/licenses/by/4.0/ -199 //creativecommons.org/publicdomain/zero/1.0/ - 706
From CC, definitions:
Adapted Material means material subject to Copyright and Similar Rights that is derived from or based upon the Licensed Material and in which the Licensed Material is translated, altered, arranged, transformed, or otherwise modified in a manner requiring permission under the Copyright and Similar Rights held by the Licensor. For purposes of this Public License, where the Licensed Material is a musical work, performance, or sound recording, Adapted Material is always produced where the Licensed Material is synched in timed relation with a moving image.
Problems with publishing Datasets:
licence compatibility problem: Many of the CC-licenses are incompatible (see table). This can only be circumvented, by using multiple licenses for different parts of the dataset. Another question is about how to license the derivatives (i.e. the classification models we produce) when those contain multiple conflicting licenses.
Non-Derivative (nd): As stated under "Adapted Material", we are not allowed to translate, alter, arrange or transform the original data. I think it is impossible to comply with the Non-Derivative license, when uploading our data to Hugging-Face
Share-Alike (sa):
Possible Solutions
Instead of pushing to HF, just publish a script, that pulls the original data, converts it and adds it to the local HF-repository. Fair Use under copyright law: When publishing models that were trained on the datasets there should not be any problems. This is because we made transformative changes (from a dataset to a classification model).
We could also ask the authors for permission.