[05 Aug 2005] This page somewhat out of date, there are now psycholinguists who understand that lanuage is learned like anything else. See papers by Michael Tomasello and also the Center for Research in Language in UCSD.
An important assumption underlying many psycholinguistics theories is that the brain, or at least the part that deals with language, is a symbolic system. See here what is wrong with this assumption.
Part of the "innatist" psycholinguistics (psyhcolinguistics which argue that langauge is innate) argument for the immateness of language is that all human languages follow some rules, so these must be built-in. This is an example of the 'wrong null hypothesis' error (see in Reasoning errors, [3.20]), because their 'null hypothesis', i.e. how would language look if there weren't any innate rules, is that it would not have any common rules. That is an obvious nonsense, because language is used for communication, so all languages have to follow those rules that are needed for communication. Thus, we first have to find what these are before starting to talk about innate universals.
I tried to find a discussion of what are the rules that languages have to follow because they are used in communication, and couldn't find any. Trying to raise the question in the mailing list LINGUIST resulted in quite amazing answers.
Since nobody else discusses this, I wrote down my ideas of Language universals from its role in communication . The main problem with this text is that most of it looks trivial, almost tautological. This is because the ideas are really almost trivial, and the fact that "innatists" psycholinguistics ignore these trivial ideas is the most important justification for regarding "innatist" psycholinguistics as nonsense. These ideas can either be rejected (by some argument) or taken into consideration, but currently psycholinguists simply ignore them.
The 'critical period' that "innatist" psycholinguistics use in their arguments has a simple explanation, given in the myths  page.
Pinker's book The Language Instinct is probably the most popular book about the subject, because he is the best user of demagogical tricks of the lot. Here I analyze a short segment of Pinker's book, showing some of the demagogical tricks he uses.
After several decades of research, it is clear that there isn't a theory that can explain all human languages. Therefore "innatist" psycholinguists start to rely on fudge factors. To make it more acceptable, they rename 'fudge factors' to 'parameters', and the 'fudge your way' approach is officially called 'Principles and Parameters'. This is currently the leading approach in psycholinguistics
A common error of psycholinguistics is to assume that when children acquire language, they acquire the same formal rules that linguistics use to describe the language. This baseless assumption is the basis for several lines of arguments, For example, the 'learnability' theory of Pinker.
Jackendoff (Patterns in the mind, 1993) bases his logic on a similar error. He argues that linguistics looked for years for the exact rules of the grammar, while each child learns them easily, which shows that the child has something innate to help him. Jackendoff fails to realize that linguistics look for a formal definition of the language that is globally applicable, while the child doesn't bother about formality or globality of his/her 'rules'. He takes it for granted that what the child is learning is the formal system that the linguistics look for.
Most of the arguments for innateness are nonsense, and here there a discussion of the worse of it. Of course, the primary nonsenser is Chomsky, which worth his own page.
A separate page about the "Language Gene" hype
The 'cognitive linguistics' (cognitive semantics, cognitive grammar), a branch of linguistics which is represented by Lakoff, Johnson, Langacker and others is an example of 'detached models' (see in Reasoning errors). For example, Lakoff's rejection of the formalist approach in favor of experience-based approach is a step forward, but his 'theory' of Idealized Cognitive Models (ICM) is not a theory at all. There are no restrictions of what the ICMs can be or do, so they offer an infinite model space which can fit everything, but cannot tell us how other systems (including the brain) behave.
Deacon's book The Symbolic Species got many good reviews, and was hailed as 'destroying Chomsky', but the analysis shows it does not contain a serious discussion.
In case you think that philosophy of language may help us to understand anything, here are my comments on one of Putnam's books, which is a reasonable reflection of the current state of philosophy of language. .