The consensus in
Cognitive Psychology these days is that language is translated into a kind of
pseudo-language called "
mentalese" before it is used in thinking. Most people who think this also hold that mentalese is universal -- we are born with it, or the potential to develop it.
For example, evidence for this includes a surprising set of experiments that show that the logical modes available in your native language have no effect on the logical modes with which you can think. (This is an incontravertible empirical refutation of the Sapir-Whorf hypothesis) Chinese dialects, for instance, have no subjunctive case, but unilingual chinese speakers have identical performance to English on reasoning tests that involve "what if" scenarios.
An alternate explanation for ketaset's observations could that the words ketaset thinks ketaset thinks are merely epiphenomena -- by products of thinking that is going on in mentalese. But hey, it's just a theory.