I could see a sufficiently abstract interpretation of 'abstract' which might fit your (or Vygotsky’s) theory though. But then you should make it precise.

Abstraction can mean finding a concept common to multiple more specific concepts or entitites. Example: Apples and oranges are fruits are edible. Fruit is an abstraction of certain plant properties, edible even more so. I can form this abstraction without language (albeit not communicate it without). Abstraction in this sense is a special kind of pattern detection and can be performed by non-linguistic algorithms.

Abstraction as a process working on other abstract entities via language by naming properties or stripping away properties is linguistic by construction but then the result is trivially true.

I think that I can perform abstraction without a) recursing to others and b) without language. To try this out I had a look at a painting (art is a domain I have no prior knowledge in so no words for concepts I might encounter). I noticed certain color patterns, shapes reoccurring in the picture. Not the same shapes but patterns. I could have given names to these but refrained for test purposes. But I nonetheless could think about - notice, quantify - the reoccurrences. Isn't this abstraction? Isn't this non-linguistic (except for communicating it here after the fact)? I agree that this kind of abstraction is difficult to transport without language (I could paint the gestalt of the pattern and show you). I also agree that this kind of abstraction is likely not applicable to other domains (except other visual or motorical domains) without language.