'My present point is that there is only one kind of theory that has ever been proposed for concept learning – indeed, there would seem to be only one kind of theory that is conceivable – and this theory is incoherent unless there is a language of thought. In this respect, the analysis of concept learning is like the analysis of considered choice; we cannot begin to make sense of the phenomena unless we are willing to view them as computational and we cannot begin to make sense of the view that they are computational unless we are willing to assume a representational system of considerable power in which computations are carried out.'
(Fodor, Jerry A. (1975): The Language of Thought, 36)
'Learning a language (including, of course, a first language) involves learning what the predicates of the language mean. Learning what the predicates of a language mean involves learning a determination of the extension of these predicates. Learning a determination of the extension of the predicates involves learning that they fall under certain rules (i.e., truth rules). But one cannot learn that P falls under R unless one has a language in which P and R can be represented. So one cannot learn a language unless one has a language. In particular, one cannot learn a first language unless one already has a system capable of representing the predicates in that language and their extensions. And, on pain of circularity, that system cannot be the language that is being learned. But first languages are learned. Hence, at least some cognitive operations are carried out in languages other than natural languages.'
(Fodor, Jerry A. (1975): The Language of Thought, 63-64)
'If learning a language is literally a matter of making and confirming hypotheses about the truth conditions associated with its predicates, then learning a language presupposes the ability to use expressions coextensive with each of the elementary predicates of the language being learned. ... The upshot would appear to be that one can learn L only if one already knows some language rich enough to express the extension of any predicate of L.'
(Fodor, Jerry A. (1975): The Language of Thought, 80)
'It has been a main argument of this book that if you want to know what response a given stimulus is going to elicit, you must find out what internal representation the organism assigns to the stimulus. Patently, the character of such assignments must in turn depend upon what kind of representational system is available for mediating the cognitive processes of the organism. The present point, however, is that that’s not all that it depends on. On the Broadbent-Treisman model, it is attentional mechanisms which determine how the available representational capacities are exploited. On the Garrett-Lackner model, it is whatever mechanisms affect the transfer of information from computing memory to long-term memory.'
(Fodor, Jerry A. (1975): The Language of Thought, 163)
"To show that mental images violate 'the rules of images in general', one would have to show not just that they are indeterminate under some visual description or other, but rather that they are determinate under no visual descriptions at all. There may be a way of showing this, but I doubt it and the striped tiger argument doesn’t do it."
(Fodor, Jerry A. (1975): The Language of Thought, 191)