Suppose I have speech impediment where whenever I try to say a sentence s of English, what comes out of my vocal apparatus is something that sounds just like a semantically unrelated sentence h(s) of Hittite. Suppose that h is a one-to-one map between sentences of English and what sound like Hittite sentences. Thus, when I try to say "Snow is white", it sounds just like Hittite for "I would like to sell you a square circle." I also have an associated hearing defect. If the sound of h(s) occurs in my environment, what I hear is the sound of s. Thus, if you speak Hittite around me and say "I would like to sell you a square circle" in Hittite, it sounds to me as if you said in English "Snow is white". As a result of the hearing defect, it sounds to me as if I were just speaking English normally—my hearing defect cancels out my speech defect.
Suppose, now, that you have the same pair of defects. Then we can communicate with one another just fine. Question: What language are we speaking with one another? Are we speaking Hittite or English or something else?
Surely, we are speaking English, but mispronouncing. But this means that we had better not understand languages in terms of the sounds made, but at most in terms of the sounds intended to be made.
We may further suppose that as it happens, the sentence h(s) of Hittite has the property that it is appropriate to assert whenever s is appropriate to assert, but has a different meaning. If that hypothesis is correct, then one cannot read off what people are saying from their behavior.
