Human language is intrinsicallye creative -- in principle, an infinite number of different sentences could be produced. How do finite resources produce this creativity? Via mental grammar. "A human head is not big enought to store an infinite number of sentences, or even a hundred million trillion sentences. So what we know when we know a language is a program, or recipe, or set of rules, that can string words together in an unlimited number of systematic combinations."
Issue: How do children develop or acquire mental grammars? "Parents just talk to their children; they do not give them grammar lessons. Somehow, the children must distill the rules out of their parents' sentences. "This `somehow' must be constraints provided by an innate universal grammar. "It is a logical necessity that children's mental learning mechanisms be constrained in some way, for otherwise they could not generalize correctly beyond their parents' sentences to the rest of the language."
Overregularization: A Case Study Of Grammatical Creativity
Pinker now focusses on a specific example, acquiring the past-tense form of verbs. Part of this is revealed by errors that children produce when they produce something called "overregularization errors".
The Course Of Rule Development
Overregularization errors let us see the learning of the past-tense rule. "somewhere between the ages of late one and late two, the first `overregularization error' (as forms like goed and holded are called) appears -- a clear sigh that the child has acquired something like the past tense rule. Children are likely driven to abstract this rule because of their universal grammar. In terms of accuracy over time, a child's performance on verbs will be U-shaped --"the curve starts at 100 percent and then dips down (in one sense, the child is getting worse while growing older) before rising back to 100 percent by adulthood."
Explaining Rule Development Is Not So Easy
Big question -- what makes overregularization errors go away? Not a need to improve communication, nor the fact that adults have never heard these forms..."There must be a piece of adult psychology that causes the experience of hearing an irregular form like held to block the subsequent application of the regular `add -ed' rule to that item." Pinker calls this "piece of adult psychology" the blocking principle.
"So, maybe children lack the Blocking principle and have to learn it." But this begs the question of how the blocking principle is learned, which turns out to be tricky, because such learning does not apparently require negative evidence.
A Simple Explanation Of Overregularization -- That Works
Maybe children do not have to learn the Blocking principle, because it is already part of their universal grammar. But if this is the case, they should never produce overregularization errors! The way out of this conundrum is to appeal to memory -- memory of irregular forms does not start out strong enough to invoke the Blocking principle. "If children have heard held less often, their memory trace for it will be weaker, and retrieval less reliable."
"The cure for overregularization is living longer, hearing the irregulars more often, and consolidating them in memory, improving retrievability. Indeed, according to this account, we do not even need to posit any qualitative difference between the mind of the child and the mind of the adult, and so the account has parsimony on its side."
Evidence For The Blocking-Plus-Retrieval-Failure Theory
Pinker gives us 10 facts that support his theory:
Past Tense Overregularization And Connectionist Modeling
Alternative to Pinker's theory is that rules are not acquired. One example of this is a PDP model that gives a good account of the U-shaped performance as past-tense verbs are acquired. The performance of this network depends crucially on word frequency in the training set.
Testing Rules Versus Analogies
"Is there any way to tell whether children's overregularization errors are due to overapplication of a mental rule of the effect of automatically analogizing from the patterns found in regular verbs?
One point against PDP is that parental word frequency does not vary much during the U-shaped phase of children's learning. Also, there is no consistent relationship between stem similarity and errors, as the analogy model would predict. Finally, similar sounding words do not produce the errors, as is strikingly demonstrated by the fact that while overregularization is found for open-case version of verbs like be, it is never found for the close-case version, which sound identical! The only piece of info consistent with the PDP model is the fact that analogy seems to be used between different irregular verbs. "Connectionist pattern asscoiators may not be accurate models of the regular rule, but they may be accurate models of the memory system in which the irregulars are stored."
Lots of interesting facts about language can be seen to emerge from the study of this one phenomenon:
Pearl Street | "An Invitation To Cognitive Science" Home Page | Dawson Home Page |