Page 89 - Encyclopedia of Philosophy of Language

learning to the discovery of the regular rule. Once the regular rule is discovered, it applies to all verbs unless there is an irregular past tense form available in the child's lexicon, causing blocking of the general rule. Overregularization is attributed to a lack of memory strength for a 'blocking' irregular form. When an irregular form is unavailable, the regular past tense rule is applied to the irregular stem as a default.
The Marcus, et al. study is a contribution to the debate about the acquisition of morphology, and ulti- mately about acquisition generally. It represents a perspective on learning which, in common with the majority of studies, assumes that the child's language development depends on the organization and reor- ganization of rules and representations. The alter- native view, often referred to as 'connectionism,' sees statements such as the rule for past tense formation as merely descriptions of features of the language. The connectionist view sets out to demonstrate how a model of language acquisition could avoid reliance on mechanisms using rules that manipulate discrete symbols, but still account for what happens in the child's language learning. The battleground for the competing theories has been past tense formation in English, and the connectionists' hypothesis testing has implemented computer simulations of learning, using parallel distributed processing (PDF) models (Rumel- hart and McLelland 1986; Plunkett and Marchman 1991). Such models are constructed in the form of networks (claimed to be analogous to neural net- works) which are 'trained' on sets of past tense forms from the language, and which 'learn' from successive sweeps through the input. Successive outputs from the network can then be checked for their approximation to what is known of successive stages of the child's development. In the rather restricted area of the devel- opment of past tense, the simulations have been rela- tively successful, though they are still the subject of extensive debate.
2.2 Linguistic Theory and Language Acquisition
As seen above, linguistic theory was a major influence on language acquisition studies at the outset. After a period of estrangement, the relationship was renewed during the 1980s as the reformulation of Chomskyan theory (which is referred to as 'Principles and Par- ameters Theory'—PPT) offered the prospect of testing predictions against language acquisition data, in a way that had not been possible before. It is obviously not feasible to deal with the full complexity of PPT here (see Atkinson 1992 for a book-length treatment. However, one of its crucial dimensions, parameter- ization, can be introduced via a specific example.
The original linkage between linguistic theory and language acquisition depended on commonalities between languages at a rather abstract level. PPT maintains this view that there are universal features of language but also acknowledges cross-linguistic
differences by specifying, within modules of the gram- mar, 'parameters'—dimensions of variation from which languages select possible values. The most widely discussed such dimension in the child language literature is the 'null subject parameter.' One of the differences between English and a Romance language like Italian is that in the latter subjects of sentences do not need to be explicitly realized by a noun phrase or a pronoun, as verb paradigms are inflected for person and number, as well as indicating tense. So an Italian hearing a sentence which consists only of a verb and an object noun will be able to identify the subject of the sentence from the verb form. In English, by contrast, subjects must be expressed. There are exceptions to this rule, but they are limited to certain well-defined contexts such as responses to questions, e.g.,
Q. what did you do?
A. finished my drink and left.
Faced with the problem of learning their language, Italian and English children have to determine which way the null subject parameter is set, on the basis of the input evidence they hear. More generally, for each parameter that is made available within the theory to account for linguistic variation, the child has to determine which setting his particular language selects. The PPT theory is still very much an innateness hypothesis. As before, the child is seen as coming to the task of acquisition 'hard-wired' with the principles of the theory, and with the parameters. Input (earlier called CDS) is more significant, in this view of the linguistic theory-language acquisition relationship, but only to provide just enough evidence to set the relevant parameters. And input to the child is restric- ted to what is called 'positive evidence.' An important part of the argumentation for the new innateness hypothesis is that the child does not receive any overt evidence about the structure of the language. In par- ticular, he receives no 'negative evidence' when he makes errors (such as overregularizations). This view does seem to be borne out by studies of CDS, at least so far as clearly explicit parental correction of syntactic or morphological error is concerned. Mothers and fathers do not generally take any notice of grammatical errors on the part of their preschool children. They seem to regard them, rightly, as a nor- mal part of development, which the child will grow out of. Furthermore, even if an adult does try to correct overtly a child's error, the attempt isunlikely to be successful, unless the child is ready to make the change to the more adult-like form. As a consequence, any incorrect hypotheses about the structure of the language which are made by the child have to be eliminated by his own efforts, without any direct inter- vention by adults. This can only be achieved, it is argued, if the 'hypothesis space' for language learning is heavily constrained from the outset. The principles
Language Acquisition in the Child
67