The train of argument that leads Fodor to his extreme conclusion begins innocently enough. Almost everyone in the various nature-nurture debates acknowledges that people have to be born with an ability to represent certain elementary concepts (if only "red", "loud", "round", and so on) and an ability to assemble new ones from this inborn inventory as a result of experience (if only by associating them with one another). For instance, the complex concept "red square" is learned by connecting the simple concepts "red" and "square". The key question is, which concepts are part of the innate inventory, and which are assembled out of them (or at least get their meaning from the way they connect to them)? One way to answer the question is to distinguish the concepts that are patently decomposable into simpler ones (such as the meaning of "the man in the gray flannel suit" and other combinations of words) from the concepts that are patently atomic, without containing anything smaller or more basic (such as "red" or "line," which are triggered directly by the eyes and visual system). On the nurture side, empiricists tend to make do with an abstemious inventory of sensori-motor features, invoking only the process of association to build more abstract set of concepts, such as "cause", "number", "living thing," "exchange", "kin", and danger, come to us ready-made, rather than being assembled onsite.
Both sides, if pressed, have to agree that the simple building blocks of cognition - like the keys on a piano, the alphabet in a typewriter, or the crayons in a box - must themselves be innate. Type on a standard typewriter all you want; though you can bang out any number of English words and sentences and paragraphs, you'll never see a single character of Hebrew or Tamil or Japanese. As Leibniz said in amending the slogan of empiricism, "There is nothing in the intellect that was not first in the senses... except the intellect itself."
What about the concepts that underlie the meanings of words? Both an empiricist and a not-so-extreme nativist would be satisfied with the claim that most of these concepts are built out of more elementary units - perhaps mother is mentally represented as "female parent," perhaps kill is conceptualized as "cause to become not alive." These units are innate, or perhaps in turn are decomposable into even more elementary units that are innate. (The buck has to stop with something innate, so we can explain why children, but not chickens or rhubarb or bricks, can learn words and concepts in the first place.) If a unit can't be decomposed into a combination of more basic units, it must be innate, just as the letter A, which can't be built out of anything simpler, is innate in a typewriter.
But, Fodor claims, the meaning of most words can't be decomposed into simpler units. Definitions always leak. Kill, for example, doesn't really mean "cause to become not alive." As we saw in the previous chapter, you can cause someone to become not alive on Wednesday by poisoning him on Tuesday, but you can't kill someone on Wednesday by poisoning him on Tuesday. Nor you can kill someone by slamming the door in his face as he flees a mad dog, though you can cause him to become not alive that way. Moreover, Fodor says, philosophers who have tried to reduce complex concepts - such as "know", "science", "good", "explain", and "electron" - into definitions made of more elementary concepts have failed abjectly in their efforts. Finally, he argues, when we turn to the psychology of people using language in real time, we see no sign that they have more difficulty with putatively complex concepts than with putatively simple ones. For example, intuitively we feel it's no more difficult to understand the word father than to understand the word parent, even though "father" is sometimes held to be a complex concept defined out of the simpler concepts "male" and "parent."