Against LOT

I'm not a fan of the language of thought, since it's a theory that seems to have a number of prima facie absurd conditions built into its premises. Still, a Sunday morning is a good time to muse about it.

Fodor, in “Language of Thought: First Approximations," makes the following move. He is discussing “concept learning.” He says,

what distinguishes rote learning and sensory learning from concept learning is that, in the former cases, what is remembered of an experience typically exhausts what is learned from that experience. Whereas concept learning somehow ‘goes beyond’ the experiential data. ….concept learning is essentially a process of hypothesis formation and confirmation. (Fodor, 1975)

Since experiential data serves to strengthen the concept being learned by confirming instances of its application, and since this process of confirmation is to be a hypothesis confirmation process, and since it seems that the only way Fodor allows us to understand a hypothesis is in terms of first-order sentences, we see that the LOT is already built into our account.

Are there not simple counterexamples to this? Can we not learn the “concept” of, say, convergence at the horizon by having a picture of a landscape that matches more or less well with visual stimuli? A “confirmation” relation would obtain just in case the mental picture is (apparently) more or less congruent to the visual field. See also geometric “similarity”—but highly unusual shapes can fulfill the same mathematical “distance” under an averaging algorithm. A pretty mess.