It's not clear to me how they are using Friedman's test in that paper.*
Friedman's test is applicable only in the case of unreplicated complete block design. It doesn't test for an "interaction" in any sense I would use the term.
Why not explain what you are trying to accomplish? And maybe someone can help you find the right approach?
Thank you for your answer! I am trying to replicate their results, so I would be interested to find out if they used the wrong test or used another test (not Friedman´s) and just reported a wrong test name... Would you or anyone have an idea about how to test for an interaction between the task type and feature of the task, as in the paper (if it´s at all possible)?
You might try emailing the authors. That's your best bet.
One way to assess interactions with a nonparametric test is with an aligned ranks approach. There is an aligned ranks method attached to Friedman's name, so that's a possibility. (Search for Friedman:).