Noam Chomsky gave the name "Plato's Problem" to the general problem of how we come to know things in conditions of sparse evidence. As Chomsky has shown over the years, this problem arises sharply in the case of language. In linguistics, Plato's Problem is understood to be the problem of explaining how a grammar can be acquired under conditions of poverty of the stimulus, that is, a seeming lack of sufficient positive and negative evidence.

Plato's own take on this problem is found in his dialogue, The Meno. Meno raises what appears to be an intractable paradox. How, he asks, can one investigate what one does not know? Which of the things you do not know will you propose as the object of your search? Even if you stumble across it, how will you know it is the thing you did not know? Let us call this "Meno's Paradox."

Plato's Problem refers to the gap between experience and knowledge. To close the gap, we need to either show that learners have more experience than we thought, or that they have some knowledge from another source. Plato took the second approach: learning is but recollection of knowledge we acquired in a previous life. Generative linguists have adopted a version of this solution; that is, to provide the learner with innate knowledge in the form of principles of Universal Grammar (UG). By restricting the set of hypotheses that a learner can formulate, it overcomes, at least to some extent, the poverty of the stimulus, and in this fashion provides a partial solution to Plato's Problem

Meno's Paradox is a sharper form of Plato's Problem. Whereas Plato's Problem requires us to reduce the gap between knowledge and evidence, Meno's Paradox raises the frightening prospect that we could be immersed in an ocean of evidence and would not recognize it for what it is. How does the process of recollection work? In the story of the slave boy "recalling" the Pythagorean Theorem, the boy required some skillful prompting from Socrates in an ordered sequence. The basic concepts may indeed be innate - the notions of lines and triangles, and so on - these are plausibly provided by universal principles of cognition. But supplying these concepts does not in itself give us the Pythagorean Theorem. In Plato's story, something further has to happen, a systematic error-driven procedure in which a series of questions of increasing complexity are posed and answered in order.

In linguistics, Meno's Paradox takes the form of the question: How are learners able to relate their experience, however copious, to their abstract principles of UG? Meno's Paradox has been the central problem confronting learning algorithms for generative grammar. I will review a number of proposed solutions to this problem. I will conclude that the best current approach is one that has some of the characteristics of Plato's procedure.