I found a dataset built on top of Wikipedia dump, which comes in Huggingface Dataset library. The Wikipedia dump is licensed under CC BY-SA and the Huggingface Dataset is licensed under Apache-2.0, but there is no license specified for the dataset I want to use. My question is, can the dataset be licensed under more restrictive license? Or can I assume, that it has the same license as Wikipedia dump?
In general it is certainly possible, but "viral" licenses, such as the SA ("share-alike") CC licenses, require derived works to be provided under the same license, so the derived work cannot be more restrictive (or less restrictive, for that matter).
With datasets it might be a little more complicated in that creating a dataset from a copyrighted work might not require permission from a copyright holder, in which case the dataset author does not have to follow the requirements of the license and can license the dataset however they wish. Whether that's the case depends on the jurisdiction and the way the original work was used to create the dataset.