Professors Kenneth Wexler (MIT) and Peter W Culicover (Ohio State University) in their book, Formal Principles of Language Acquisition (1983):
It is therefore necessary to ask if negative information is in fact available to a language learner. It is clear that parents do not simply labeled nonsentences to children in a systematic manner; no parents (or other speakers) say "Here is a sentence, and it is ungrammatical, and here is another one, and it is ungrammatical, and here is a third, which is grammatical." Even as a first step in looking for the existence of negative information we have to turn to the concept of correction, the concept that when the child produces an ungrammatical sentence, he is somehow informed that the sentence is somehow ungrammatical. The child will have to have some abilities of preanalysis in which some kind of event is translated into the information that the sentence he has spoken is ungrammatical. If some event can result in the learner's deciding that a sentence is ungrammatical , we can call this negative information. Of course this interpretation is consistent with the general need for preanalysis.
The question therefore becomes: Is the child corrected when he produces an ungrammatical sentence? (37) In the opinion of those who have studied the corpuses of of children's speech, there seems to be very little of this kind of feedback. [...] Brown and Hanlon (1970) analyzed a corpus of mother-child interactions and measured the proportion of nonapproval responses of the mother to "primitive" (ungrammatical) and to well-formed expressions of the child, and found no significant difference between the two proportions; mothers did not differentially disapprove grammatical and ungrammatical sentences. Brown and Hanlon also considered the possibility that a more subtle process of correction was going on than approval or disapproval, namely, that the child was understood more readily when he uttered a grammatical sentence than when he uttered a primitive sentence. In other words, the learner would have to translate noncomprehension by the listener into into an assumption that his sentence was ungrammatical. To examine this question Brown and Hanlon measured the proportion of times that mothers produced comprehending responses to grammatical and primitive sentences and again found no significant differences. Mothers seemed to understand ungrammatical sentences about as well as grammatical ones. The investigators pointed out that, in fact, what parents correct are semantic mistakes, not syntactic ones.
This evidence is only partial, of course. For example, it might be that mothers try especially hard to understand children's utterances, whereas other adults and children differentially understand grammatical and ungrammatical utterances. There is much room for empirical research on these questions, but Slobin (1972) claims that in the many societies the Berkeley group has studied there is no evidence that children are corrected for ungrammatical utterances. Pp.64-65.
37. Even if the child is corrected, there is no reason to believe that he takes note of the correction in a way useful for language acquisition. Consider the mother-child interaction reported in McNeill (1966, 69):
Child: Nobody don't like me. Mother: No, say "Nobody likes me." Child: Nobody don't like me. [... Eight repetitions of this dialogue ...] Mother: No, now listen carefully; say "nobody likes me." Child: Oh! Nobody don't like me.
R Brown & C Hanlon. 1970. "Derivational Complexity and the Order of Acquisition of Child Speech" in JR Hayes Cognition and the Development of Language.
D McNeill. 1966. "Developmental Psycholinguistics" in F Smith a GA Millers The Gensis of Language.
D Slobin. 197. "Children and Language: They Learn the Same Way All Around the World" in Psychology Today 6, 71-82.
It seems obvious to me, but perhaps it is not to others, but corrective feedback cannot happen before at least some part of the L1 grammar has been learned. An important question is how much of the grammar needs to be learned before corrective feedback might become useful. My feeling, without evidence or study, suggests more than the babbling stage and yet before the complete grammar has been learned. Another important question is how faulty grammar rules are learned? I see several possibilities:
1. The child generates a normalized form to cover a situation where the grammar has an exception, e.g., the child says gooder rather than using the suppletive form better.
2. The child learns a rule of grammar by being exposed to speakers whose dialect is not the standard which the parents wish the child to learn. That is, there are two different grammars, the one being learned by the child and the one the parents want the child to learn for matters of prestige or economics.
Some might believe that the child simply makes a sort of random mistake which is possible. Even fluent native speakers make grammatical errors or typos, slips of the tongue, etc. I just don't see any evidence (yet!) that infelicities of this sort would lead to verbal anarchy. To be able to communicate with one's peers one would learn the grammar that describes the language they are speaking. If this dialect differs slightly from the previous generation's or the standard (as embodied not in a reference grammar, but in the preferred texts of the speech community, i.e., its literature), there is very little chance of total confusion of the sort that the phrase "verbal anarchy" would suggest.