Samuel Jay Keyser
A friend of mine likes to start the first lecture of his introductory course in linguistics by demonstrating a talking doll. When you pull a string in its back, the doll says things like, "Hello. My name is Sandy," or "I want to play. Don't you?" Whenever you pull the string, the doll begins a new sentence; it never starts in the middle, even if you only pull the string out part way. There is no strict sequence of sentences. Their number is very small, perhaps ten or so, and the order in which they are heard is random.
Suppose we ask the question: How is the doll able to talk? From the brief description just given, it would not be easy to answer this question. Several possible answers might come to mind and it is very possible that the one crucial clue to help us choose between answers will never occur.
The talking doll problem is very much like the talking human problem. Linguists and psychologists try to figure out how people talk. They must do so without looking inside. For one thing, if they could look inside the human brain safely, it is not clear what they could learn by looking. Instead, researchers listen to what people do when they speak and try to make intelligent guesses about the inside organization on the basis of what happens on the outside.
My friend considers the talking doll to be a practical demonstration that determining how people speak is impossible since the problem cannot be solved even for a doll.
Yet linguists and psychologists think that there might be a solution to the problem of what people do when they speak and