This quote from the article suggests that mechanism involved is specifically linguistic:
"""
To determine if words were being automatically (and perhaps unconsciously) activated, the researchers added the following twist: they asked their Russian participants to perform a verbal task at the same time as making their perceptual discrimination. This condition eliminated the reaction time advantage of contrasting goluboy and siniy. However, a nonverbal task (a spatial task) could be done at the same time while retaining the goluboy/siniy advantage. The dual task variants indicated that the task of discriminating color patches was aided by silent activation of verbal categories.
"""
Thank you for posting this. It's always my biggest complaint about Linguistic Relativity... And that so many people believe it simply because it seems like it must be true. Boroditsky (who I associate with the strong push to push it on the public) and her ilk have done a lot of harm, as this does directly impact thinking in terms of things like AAVE, which we've already seen causes transcription errors in court leading to many issues.
I think one interesting example of where the linguistic hints is overridden by the cultural milieu is in the term ski mask. Picture someone wearing a ski mask; what is he about to do: ski down the mountain, or rob a bank? I suspect most people associate the ski mask more readily with a bank robber (particularly the three-hole variant) than a warm weather gear that you might use if the windchill is going to be below 0°F. (I personally use the term balaclava instead of ski mask, though I know there's a large segment of the population who doesn't know what a balaclava is). In that case, the linguistic clue of "ski" is less important than the fact that the main context people see it in is when people are concealing their identities to rob banks in movies.