Almost every child learns how to speak and to understand his native language. At an appropriate stage of development a child learns vocabulary with amazing speed: typically a child learns many new words, and their correct usage, each day. The learning is efficient, in that a child does not need to hear the same words repeated over and over again or to be corrected very often. Thus learning language must be easy, but we do not have effective theories that explain the phenomenon.
The mystery deepens when we notice that children learn many new words without ever hearing them. In a classic experiment by Berko [2], a number of English-speaking children were shown representations of a fanciful being called a ``wug.'' When asked to say something about a situation with more than one of these beings, the children correctly pluralized the novel word to make ``wugz'' (not ``wugs''). In another experiment [9], Marcus et. al. showed that young children who first use an irregular verb properly (such as ``came'') would later err on the same verb (by supplementing ``came'' with ``comed'') before they use the verb correctly again. Thus children reliably exhibit behavior that indicates that they have made generalizations that linguists describe with rules.
If children do have knowledge of the generalizations, what is the form of such knowledge?