The theory of Universal Grammar (UG), as proposed by Chomsky (e.g. Chomsky, 1986a, 1988), sees the acquisition of the first language as a process of setting the values of parameters in the system of language principles with which the human mind is endowed, in response to "triggering" evidence, these principles and parameters being couched in terms of the Go vernment/Binding (GB) theory of syntax (Chomsky, 198la, 1986a, 1986b). The aim of this paper is to examine how observations of actual children's language can be related to this Chomskyan model of UG. It does not therefore concern the use of such observational data within approaches that do not share the UG assumptions.

1. EVIDENCE IN THE UG MODEL

The claims that UG theory makes for language acquisition are largely based on the "poverty of the stimulus" argument; given that the adult knows X, and given that X is not acquirable from the normal language input the child hears, then X must have been already present in the child's mind. This crucial argument uses the comparison of the knowledge of language that the adult possesses with the initial state of the child to establish what could not have been acquired from the types of evidence available and must therefore be innate. Chomskyan UG theory would not be discomfited if other evidence from acquisition were not forthcoming. On the one hand, such research is not of prime importance, given the reliance on the poverty of the stimulus argument. On the other, evidence from language development in the child is related with difficulty to acquisition because of the other factors involved in language performance and development - production and comprehension processes, situation and use, the growth in other mental faculties, and so on - all of which are "non-stationary" (Morgan, 1986) and liable to change as the child grows older.

The attraction of the current model is that the aspects of language built-in to the mind are precisely the principles of GB theory - the Projection Principle, the Binding Principles, and so on; the aspects that have to be learnt are the settings for parameters of variation, and the properties of lexical items. Hence built-in principles of syntax can now be postulated in a rigorous form that has testable consequences; it is possible to start looking for evidence of the effects of UG in children's language development. And also the reverse; it is possible to start phrasing research into acquisition in ways that can affect issues of linguistic theory, the prime example being the work of Hyams (1986) and Radford (1986).

2. I-LANGUAGE AND E-LANGUAGE THEORIES

A starting point is the distinction made in Chomsky (1986a) between I-language (internalised language) and E-language (externalised language). I-Language is "a system represented in the mind/brain of an individual speaker" (Chomsky, 1986a, p.36); the task of I-language theories is to describe this mental possession; hence I-language syntax is closely related to the mind and to psychology. E-language is a collection of sentences "understood independently of the properties of the mind" (Chomsky, 1986a, p.20); it is the speech people have actually produced, the ways they use language to interact, and the description of the statistical properties of language events; E-language syntax is related to social interaction and to sociology. In I-language theories, acquisition research is seen as ex­plaining how the child acquires knowledge of language; in E-language theories, it is seen partly as describing the regularities in a corpus of the child's sentences, partly as describing how the child develops social interaction. UG theory is concerned with I-language knowledge rather than with the E-language regularities in a set of utterances. GB does not necessarily confine itself to a single form of data: "In principle evidence ... could come from many different sources ... perceptual experiments, the study of acquisition and deficit, or of partially invented languages such as Creoles, or of literary usage or language change, neurology, biochemistry, and so on" (Chomsky, 1986a, p. 36-37); its chief evidence is whether a sentence conforms to the speaker's knowledge of language. In practice GB theory has concentrated almost exclusively on what can be called "single-sentence" evidence; a single sentence, such as Is the man who is here tall? (Chomsky, 1980, p39) or ?Está el hombre, que está contento, en la casa? 'Is the man, who is happy, at home?' (Chomsky, 1988, p.42), is sufficient to attest to the native speaker's language knowledge, and hence to invoke the poverty of the stimulus argument.

An I-language theory takes Is the man who is here tall? to be a grammatical sentence on the linguist's own say-so without hunting for observations of native speakers using it in actual speech. But single-sentence evidence is difficult to use in acquisition research since native children cannot be asked directly whether they accept Is the man who is here tall? as their answer would not be meaningful. Children are by and large not capable of attesting unambiguously that a particular sentence is or isn't generated by their grammar. Other than single-sentence evidence or the pure poverty of the stimulus argument, what else can count as evidence of language acquisition in a UG framework? One possibility is to use experimental techniques and statistical procedures from the psycholinguistic tradition. The research of the past decade has employed a wealth of techniques ranging from the elicited imitation tasks used by Lust and her associates (1989) to the comprehension tasks employed by Matthei (1981) and others; indeed the specific case of structure dependency seen in Is the man who is here tall? was investigated by Crain and Nakayama (1983) through a question production task.

The validity of such forms of evidence is not the concern here; an account of some of their merits and demerits is seen in Bennett-Kastor (1988). Instead the discussion will be restricted to one type of evidence that has been used within UG theory, namely the use of sentences observed in actual children's speech, which can be called "observational data". If a child is heard to say Slug coming, what status does this sentence have as evidence for UG theory? The main argument here is that there is an inherent paradox in using observational data to support a UG model that needs to be aired, even if it cannot be resolved. Observational data belongs in essence to E-language; the typical E-language study of acquisition looks at statistically prominent features found in a substantial collection of children's speech, say Brown (1973) or Wells (1985). The major problem is how to argue from E-language descriptive data of children's actual speech to their I-language knowledge, a problem first perhaps highlighted in Chomsky (1965) as "a general tendency... to assume that the determination of competence can be derived from description of a corpus by some sort of sufficiently developed data-processing technique". While it is interesting and instructive to use observational data to investigate the UG claims, the chain of qualifications and inferences between such data and language knowledge is long and tortuous.

3. OBSERVATIONAL DATA, PERFORMANCE AND DEVELOPMENT

There are two related dimensions to this within the UG theory - performance and development. Any use of performance data by linguists faces the problem of distinguishing grammatical competence from the effects of production and comprehension processes, short term memory, or other non-competence areas of the mind involved in actual speech production. Single-sentence evidence is immune to all of these factors. In this sense children's speech presents exactly the same problems for the I-language analyst as the speech of adults. GB oriented linguists base their syntactic analyses on single example sentences rather than on chunks of performance; they too have problems with deriving the knowledge of the native speaker from samples of raw performance.

But children's language also ties in with their development on other fronts; the actual sentences they produce reflect their developing channel capacity, that is to say a mixture of cognitive, social, and physical development, from which the effects of language acquisition need to be filtered out. The distortions that performance processes cause in actual speech are doubly difficult to compensate for in language acquisition research because they may be systematically, or nonsystematically, different from those of adults - short term memory may be smaller in capacity or organised in a different way, cognitive schemas may be different, and so on; insofar as these are involved in language performance they affect children differently from adults. "Much of the investigation of early language development is concerned with matters that may not properly belong to the language faculty ... but to other faculties of the mind that interact in an intimate fashion with the language faculty in language use" (Chomsky, 1981b). Cook (1988) distinguishes "acquisition" - the logical problem of how the mind goes from S₀ (zero state) to S_s (steady state) - from "development" - the history of the intervening stages, S₁ S₂, and so on. To argue from observation of children's development to the theory of acquisition means carefully balancing air these possibilities. Linguists are frequently struck by the child's presumed difficulties in dealing with primary linguistic data; their own difficulties in deriving a representation of grammatical knowledge from samples of children's performance are not dissimilar, or indeed worse since children's sentences are more deficient than the fully grammatical sentences spoken by caretakers (Newport, 1976). So the child saying Slug coming may be suffering from particular production difficulties shared by adults or from specific deficits in areas that have not yet developed, say the articulatory loop in working memory (Baddeley, 1986). The apparent syntax of the sentence may be different from the child's competence for all sorts of reasons.

Observational data thus raise two problems related to performance; one is the distortion resulting from the systematic or accidental features of psychological processes; the other is the compounding effect of the development of the child's other faculties. For observational data to be used in a UG context, eventually these distortions need to be accommodated within a developmental framework that includes adequate accounts of the other faculties involved in the child's language performance, which, needless to say, does not yet exist. Furthermore, observational data of children's speech are still only evidence for production rather than comprehension, the two processes being arguably distinct in young children (Cairns, 1984).

4. REPRESENTATIVENESS OF OBSERVATIONAL DATA

Let us now turn to some methodological issues with observational data. Taking the step from single-sentence evidence to E-language performance evidence incurs several obligations. The first is to take a reasonable sample of data. Any linguistics book or article in the I-language tradition assumes that it is meaningful to discuss people's knowledge of John is easy to please or That he won the prize amazed even John or John caused the book to fall to the floor, it is beside the point whether such sentences have ever been uttered, provided that they reflect the knowledge of a native speaker. An actual child's sentence, however, is not concocted by a linguist to illustrate a particular syntactic point and so removed from processing constraints, discourse connections, and so on; nor can it be checked against the speaker's intuitions about his grammatical knowledge. Because of the deficient nature of children's language, an example or two can probably be found of almost any syntactic possibility one cares to name, say Verb Subject order as in Comes a mummy reported in Cook (in progress) or apparent structure-dependency violations as in What does sheep make a noise? (Cook, 1988). Such isolated examples are not sufficient for E-language analysis, nor are arbitrary lists or collections of sentences; data need to be typical of the child and of children in general if they are to be of value, because of the many other causes that may be involved. As E-language data, a single sentence such as Comes a mummy might be a performance slip or a garbled attempt at a nursery rhyme or a genuine reflection of competence. A sample of sentences is required to rule out the accidental or non-accidental but non-linguistic effects of performance. Again for E-language data a reasonable sample of children need to be included; it prejudges the universality issue if arguments are based on observational data from a few children or from one child. So transcripts of actual children's speech must include a reasonable sample of children and a reasonable sample of speech for each stage of each child. They need to be based on a consistent analysis and on the same sample of children, to eliminate possible regional, social, and individual differences, as discussed in Bennett-Kastor (1988); a good example of such research is seen in Stromswold's study of some 16000 examples of sentences with wh-words from 12 children (Stromswold, 1988). Information on the relevant makeup of the children should be explicitly stated, once the initial step into observational data has been taken. All of these requirements are beside the point for single-sentence evidence and for the poverty of the stimulus argument. They are however perfectly plausible demands on E-language evidence. A requirement for combining the I-language approach to knowledge with the use of E-language data is the quantification of aspects of children's sentences; frequency of occurrence needs to be counted, calculations need to be made, and statistical reliability becomes an issue. As Bloom et al (1976, p.34) put it, "if structural features occur often enough and are shared by a large enough number of different multi-word utterances, then it is possible to ascribe the recurrence of such regular features to the productivity of an underlying rule system ...". The key words when handling E-language data are "often enough" and "large enough". A single sentence could be highly revealing of the child's grammar; but, because of the possibility that any single spoken sentence is an isolated freak due to memory problems, rote repetition, discourse constraints, or any of a host of other factors, the UG analyst using E-language data has a duty to put them in the context of a broader picture of children's speech. This is not to say that frequency of occurrence is crucial; it is clearly irrelevant to single-sentence I-language evidence. But E-language evidence needs to be safeguarded by showing that the data reflect some general property of the child's knowledge rather than being one-off instances.

5. OBSERVATIONAL DATA AND ADULT PERFORMANCE

A major point also concerns what data from children should be compared with - adult competence or adult performance? The significant paper by Radford (1986), for instance directly compares two sets of sentences, one consisting of actual child performance such as That one go round, the other of bracketed adult versions such as Let [that one go round], as if they were the same type of data (pp. 10-11). It is difficult to offer children's E-language data as evidence for their knowledge of language without comparing them with E-language data from adults. A comparison of observational data from children with single-sentence evidence from adult competence begs many questions. Once it is conceded that adult perfor­mance needs to be used, a range of phenomena must be taken into account that GB syntax has mostly excluded. Let us take the pro-drop parameter as an illustration. The main criterion for a pro-drop language is the absence of certain subjects in declarative sentences. In her important work with pro-drop Hyams (1986) found that children from three different language backgrounds have null-subject sentences; she regarded this as confirmation of an initial pro-drop setting, later rephrased as [+uniform] morphology (Hyams, 1987). However, adult E-language data for English reveal that subjects are often omitted in actual speech and writing, usually at the beginning of the sentence. Taking a random selection of sources, Can't buy me love and Flew in from Miami Beach come from well-known song lyrics; the opening pages of the novel The Onion Eaters (Donleavy, 1971) contain Wasn't a second before you came in, Must be ninety now, and Hasn't been known to speak to a soul since anyone can remember, a column writer in The Weekend Guardian with the pseudonym "Dulcie Domum" typically uses sentences such as Drive to health food shop for takeaway, In fact might be too exciting, and Replies that it's in my desk drawer (Domum, 1989) - indeed in this article some 34 out of 68 sentences have at least one null-subject; an anecdote in Preston (1989) concerns a prescriptively oriented teacher denying that she uses gonna "Ridiculous" she said; "Never did; never will". Adult speakers of English appear to use null-subject sentences, even if they only utilise them in certain registers and situations. So, if the performance dimension of variation between styles of language is taken into account, null-subjects may be expected to appear in children's performance and children may also be expected to have encountered them in some forms of adult speech.

But also, given the many ways in which children are different from adults, an argument based on observational data has to explore the alternative developmental explanations that might cause something to be lacking from their speech. One explanation might indeed be a more frequent use by children of some performance process that the initial elements or elements in the sentence can be omitted - a clipping of the start of the sentence - which creates the illusion of pro-drop among other effects. A counterargument is that the null-subject is not always initial and hence not a product of utterance-initial clipping; however in a sample of children's language discussed in Cook (in progress) only 3 out of 59 null-subject examples had non-initial null-subject. Another explanation might be the "recency" effect whereby children pay attention chiefly to the ends of sentences (Cook, 1973), thus being more likely in SVO languages to omit subjects than objects and hence giving the illusion of null subject sentences. Hyams (1986) presents the counterargument that, at the same time as children produce subjectless sentences, they also produce ones with subjects, so that the lack of overt subjects is not a memory limitation; while this may well be true, the use of null-subject sentences by English-speaking adults is equally not a product of memory limitations. A further explanation might be found in the type of subject that is missing. Children may leave out some first person subjects because they feel they are not needed, and this might be a cognitive universal; Sinclair and Bronckart (1971) suggested that at a certain period children see themselves as the implicit subject of the sentence; Halliday (1985) sees first person subjects as the most prototypical form. According to Hyams (1986, p.69), however, "the referent of the null-subject is not restricted to the child himself. Yet in the same sample of sentences some 39 out of 59 null subjects were apparently first person; a high proportion, though not all, of children's null-subject sentences could be attributed to this source.

To decide between these conflicting explanations would entail crosslinguistic evidence based on large samples of non-first person null-subject sentences. The general point is that sheer observation of forms in children's language is susceptible to several explanations other than language know­ledge per se. Other evidence is required to decide between the explanation based on adult E-language data, the explanations based on performance and cognitive development, and the linguistic UG explanation. The cor­rectness of the pro-drop explanation cannot be uniquely shown from E-language observational data without in some way examining the other explanations within a larger framework of children's development and of language use by adults. If E-language performance by children is compared directly with adult E-language data rather than with adult I-language knowledge, an apparent peculiarity of the child's language may be shown to be a fact of performance shared by adults but not taken into account in the standard descriptions of grammatical competence within GB theory. Adult users of a language tend to be regarded as grammatically perfect without seeing that they are subject to the same factors of performance as children: some adults, on some occasions, do produce null-subject sentences in English. In one sense this claims simply that alternative explanations can be found for apparently linguistic phenomena; since the alternative proposals are not precisely formulated, what is wrong with accepting the linguist's account? But in the early stages of child speech this must be faut mieux; until there is an overall theoretical framework that encompasses the diverse aspects of children's development, it will not be possible to extract language acquisition from language development by means of observational data alone.

6. EVIDENCE OF ABSENCE

The other problem is what counts as observational data. Sherlock Holmes once drew attention to the behaviour of the household dog during a burglary; Watson pointed out that there wasn't any behaviour since the dog had not even barked; Holmes showed the very absence of barking implicated one of the people in the household since otherwise the dog would have barked at a stranger. Absence of something that might be expected to occur may in itself be relevant evidence. Both Hyams (1986, 1987) and Radford (1986,1988) make extensive use of evidence of children's non-performance. The major absence discussed by Hyams (1986) is the subject of the sentence but she also talks about the absence of auxiliaries and of inflections. Similarly there are twelve summary statements in Radford (1986) to show that children's grammars lack Inflection and Complement Phrases; eight claim that children "lack" or "may lack" Complementisers, infinitival to, Modal Auxiliaries, Tense, Agreement, nominative subjects, overt subjects, and VP; two statements suggest children "have no" inverted Auxiliaries and proposed wh-phrases; one states that "Child independent clauses may be nonfinite" (i.e. they may lack a finite verb); only one is phrased positively - "Child clauses have particle negation" - which partly implies that their sentences lack auxiliaries. All the arguments except one come down to the absence of elements from children's sentences. Doubtless there are problems of phrasing here; some facts could be equally expressed positively or negatively. But, whatever the phrasing, the point applies to the type of evidence that is used - evidence of positive occurrence or evidence of absence.

Arguing from absence is intrinsically problematic. If the possibilities of occurrence are circumscribed so that a precise prediction can be made as to what will be missing, then it seems perfectly acceptable. Providing a list of the objects that are on my desk at the moment is easy; compiling a list of what is not on my desk is difficult and eventually means listing all the absent objects in the universe, unless there are precise expectations of what should be on desks. Evidence of absence is hard to interpret when alternative explanations exist. After all the dog might have been drugged, or chasing a rat, or just fast asleep. When the main evidence for acquisition is what children can't do or the evidence for their knowledge of language is what they don't produce, interpretation becomes difficult if the predictions are not tightly constrained. In a sense the object of Universal Grammar is to specify the constraints on what might be expected in human language, to say what cannot occur. The difficulty in applying evidence of absence to observational data is that at the early stages of children's language many things are absent and the child is deficient in many areas: many explanations in cognitive and performance terms could be therefore advanced for things that are absent. Features that are observed to occur can be used directly as evidence; features that are not found are ambiguous if there are many of them and if they are susceptible to explanations other than in terms of grammatical competence.

The child has sometimes been referred to as a "mini-linguist"; let us reverse the metaphor for a moment by talking about the linguist as a "maxi-child". The problem that the child faces is deriving a grammar only from positive evidence; negative evidence in the shape of sentences that don't occur plays a less prominent role. It has been argued, for example by Saleemi (1988), that it should only figure as ancillary support for ideas already vouched for by positive evidence. The linguist attempts to derive the grammar of children from observational data that constitutes evidence for what occurs. The strongest requirement is that the maxi-child should keep to positive evidence alone; in this case what doesn't occur in children's speech should only support ideas already vouched for by observational data or by other types of evidence, such as comprehension tests. This is not to deny that absence of a precisely defined property is not a valid form of observational data; the dog after all didn't bark. Evidence of absence is however a tainted source in the early stages of children's acquisition since virtually anything in adult grammar could be argued to be missing; such indirect evidence needs support from other evidence of one kind or another.

One counter-argument is that evidence of absence maintains the standard generative use of what people don't say as a form of evidence; the use of starred sentences is taken for granted as a normal and legitimate form of argumentation and is shorthand for saying these sentences are not found. The constraints of UG on the mind such as structure-dependency are typically shown by demonstrating that people who have never heard an example of a structure-dependent sentence nevertheless reject it on sight. But this still constitutes single-sentence I-language evidence: is this sentence generated by the grammar of the speaker? It can be backed up with experimentally gained grammaticality judgements or comprehension tasks for many sentences or many people, though this raises the difficulties of experimental method, as seen in Carroll et al (1981). Shifting to observational data, there are indeed particular sentences that people do not use; one might point to the total lack of structure-independent sentences in some vast corpus. But is this the same as observing that there are aspects of the sentence that children leave out? The prediction is of a different kind; evidence of absence cannot be justified simply by the single-sentence appeal to what people do not say.

7. CORRELATIONS WITHIN OBSERVATIONAL DATA

Finally, as has already been seen, analyses depending on observational data frequently employ intuitive concepts of simultaneity and correlation; because X occurs at the same time as Y it must be the same phenomenon or indeed it must cause Y. Hyams (1987) for instance argues that "the child who allows null-subjects must also be analysing his language as morphologically uniform" based on evidence such as "the acquisition of the present and past tense morphemes coincides with the end of the null-subject stage in English". One problem is the matter of defining simultaneity and, conversely, sequence: what counts as the same time? what counts as a sequence? what does "coincide" actually mean? the same day, the same month, the same stage, or what? Statements that forms coexist or that they occur in a sequence mean little without an explicit framework for measuring time and sequence: developmental research needs a clock. Furthermore, if data from more than one child are being used, it is necessary to define coexistence in terms of chronological age, mental age, MLU, LARSP, grammatical stages or whatever developmental schedule one prefers: developmental clocks need to be set to the same standard time if comparisons are to be made.

Secondly, there is the question of how different aspects of behaviour correlate within a stage. Ingram (1989) finds four meanings for "stage" when considered in terms of a single behaviour and four more when considered in terms of two behaviours; in his terms, much of the UG related research goes beyond the simple "succession" stage to the "co-occurrence" stage where two behaviours occur during the same time span or the "principle" stage in which a single principle accounts for diverse forms of behaviour. In terms of observational data, given that many forms occur or don't occur at the same stage, how can it be shown which correlate with each other and which don't? All the forms present at the same stage correlate in the sense that they coexist and are part of the same grammar; what are the grounds for believing some are more closely related than others? There may be entirely independent reasons why two things happen at the same time; a paradigm statistical example showing correlation is not causation is the clocks in town all striking twelve simultaneously. Correlation is more problematic when it relies on absence rather than presence of forms. The Hyams (1986) analysis depends on a link between null-subject sentences and lack of expletive subjects; the Hyams (1987) analysis depends on a link between null-subject sentences and lack of inflections. UG can predict grouping of precise features missing from children's language and their absence can be correlated closely; but, if their absence coincides with large numbers of other absent features, the validity of such a correlation becomes hard to test; why should any two pairs of missing bits of the sentence be more related than any two other missing bits? Early children's language is difficult for observational data because it is so deficient: arguing from absence provides too unconfined a set of possibilities for correlation.

8. GENERAL REQUIREMENTS FOR OBSERVATIONAL DATA IN UG RESEARCH

Let us sum up some of the requirements for observational data in a series of statements:

- development is not acquisition. Any data from observation of children must be related to the other processes and faculties involved in speech production, i.e. to performance. Sentences from actual children's speech do not have a unique explanation in terms of linguistic competence, as single-sentence evidence may have; alternative explanations from non-linguistic sources have always to be taken into account. At the moment, because of the unknown aspects of the developmental process, the linguistic explanations have to be considered defaults to be kept or modified when a broader framework is available.

-   E-language data must be representative. Because of the distortions of actual performance, isolated sentences or small numbers of sentences cannot be trusted. Statements can never be based on isolated sentences, simply because it is uncertain how accidental these may be. Ways of balancing E-language corpora must be employed to ensure representativeness across children in general and across the speech of one child in particular; tests of statistical significance can be employed. Without this the data may illustrate a point but cannot be trusted for firmer conclusions.

-   like must be compared with like. The sentences produced by children cannot be compared directly to the I-language single sentences used in linguistic theories but should be compared with the equivalent adult productions. That is to say, performance sentences should not be compared with idealised example sentences, again a familiar point rephrasing Chomsky (1965, 36), "... one can find out about competence only by studying performance, but this study must be carried out in clever and devious ways, if any serious result is to be obtained". Children's performance should be compared with adults' performance rather than with adults' competence as shown in single-sentence evidence.

-   evidence of presence is preferable to evidence from absence. In an E-language account the first responsibility is to what actually occurs. Evidence from absence is intrinsically open to many interpretations; it loses some of its strength in early children's speech because of the multiplicity of plausible nonlinguistic reasons for children's deficiencies.

-   correlations should chiefly correlate positive data. Correlations between non-occurring data should be treated with caution in early children's observational data, again because of the general deficiencies of the child's speech.

The main conclusion to this paper is that the use of observational data within the UG theory of language acquisition must always be qualified; such data should be treated as showing the interaction of complex performance processes that are themselves developing. The work with observational data by Hyams, Radford, and others has provided a tremendous revitalisation of the UG theory in recent years; greater discussion should take place on the methodological status of observational data within a basically I-language theory. Observational data are one of the many sources of evidence that are available outside the bounds of the poverty of the stimulus argument. This paper has concentrated on the pitfalls and possibilities of this form of data. Other techniques of investigation have their own pros and cons - grammaticality judgements, elicited imitation, act-out comprehension techniques, and so on - some of which are described in Bennett-Kastor (1988); many of them could complement observational data; for example the pro-drop explanation might be preferred over its alternative because of evidence from experiments with comprehension or elicited imitation of VS or null subject sentences. We should not be disappointed if one source of evidence is not in itself sufficient. As Fodor (1981) pointed out, a scientific theory should not confine itself to a certain set of facts but "any facts about the use of language, and about how it is learnt... could in principle be relevant to the choice between competing theories". Together these alternative forms of evidence complement the basic appeal to the poverty of the stimulus used by the UG theory; in the Feyerabend view of science multiple approaches to the same area should be explored simultaneously (Feyerabend, 1975). Observational data is one useful tool in our kit, provided it is used with appropriate caution and supplemented with other tools when necessary.

REFERENCES

Baddeley, A. D. 1986. Working Memory. Oxford: Clarendon Press.

Bennett-Kastor, T. 1988. Analyzing Children's Language. Oxford: Blackwell.

Bloom, L., P. Lightbown and L. Hood. 1975. Structure and Variation in Child Language. Monographs of the Society for Research in Child Development Serial No 160, Vol 40, No 2.

Brown, R. 1973. A First Language: The Early Stages. London: Allen and Unwin. Cairns, H. S. 1984. Current issues in research in language comprehension. In R. Naremore (ed.) Recent Advances in Language Sciences. College Hill Press.

Carroll, J. M., T. G. Bever, and C. R. Pollack. 1981. The non-uniqueness of linguistic intuitions. Language 57. 368-383.

Chomsky, N. 1965. Formal discussion: the development of grammar in child language. In U. Bellugi and R. Brown (eds.) The Acquisition of Language. Purdue University, Indiana.

Chomsky, N. 1980. On cognitive structures and their development. In M. Piattelli-Palmarini. 1980 (ed.) Language and Learning: the Debate between Jean Piaget and Noam Chomsky. London: Routledge Kegan Paul.

Chomsky, N. 198la. Lectures on Government and Binding. Dordrecht: Foris.

Chomsky, N. 1981b. Principles and parameters in syntactic theory. In N. Hornstein and D. Lightfoot (eds.) Explanations in Linguistics. London: Longman.

Chomsky, N. 1986a. Knowledge of Language: Its Nature, Origin and Use. New York: Praeger

Chomsky, N. 1986b. Barriers. Cambridge, Massachusetts: MIT Press.

Chomsky, N. 1988. Language and Problems of Knowledge: The Managua Lectures. Cambridge, Massachusetts: MIT Press.

Cook, V. J. 1973. The comparison of language development in native children and foreign adults. IRAL XI/1. 13-28.

Cook, V. J. 1988. Chomsky's Universal Grammar: An Introduction. Oxford: Blackwell. [Latest edition 2007)

Cook, V. J., in progress. Universal Grammar and the child's acquisition of word order in phrases. [Never appeared]

Crain, S. C. and I. Nakayama. 1983. Structure dependence in grammar formation. Language 63. 522-543.

Domum, D. 1989. Plumbing the bidet depths. The Weekend Guardian, llth April, p.11.

Donleavy, J. P. 1971. The Onion Eaters. London: Eyre and Spottiswode.

Dulay, H. C. and M. K. Burt. 1973. Should we teach children syntax? Language Learning 23. 245-258.

Feyerabend, P. 1975. Against Method. London: Verso.

Fodor, J. A. 1981. Some notes on what linguistics is about. In N. Block (ed.) Readings in the Philosophy of Psychology. 197-207.

Halliday, M. A. K. 1985. An Introduction to Functional Grammar. London: Edward Arnold.

Hyams, N. 1986. Language Acquisition and the Theory of Parameters. Dordrecht: Reidel. Hyams, N. 1987. The setting of the null subject parameter: a reanalysis. Paper presented to the Boston University Conference on Child Language Development.

Ingram, D. 1989. First Language Acquisition. Cambridge: Cambridge University Press.

Lust, B., J. Eisele and N. Goss (in prep.). 'The development of pronouns and null arguments in child language', Cornell University.

Matthei, E. 1981. Children's interpretation of sentences containing reciprocals. In Tavakolian (ed.) 58-101.

Morgan, J. L. 1986. From Simple Input to Complex Grammar. Cambridge, Massachusetts: MIT Press.

Newport, E. L. 1976. Motherese: the speech of mothers to young children. In N. Castellan, D. Pisoni, and G. Potts (eds.) Cognitive Theory, vol 2. Hillsdale: Erlbaum.

Preston, D. R. 1989. Sociolinguistics and Second Language Acquisition. Oxford: Blackwell.

Radford, A. 1986. Small children's small clauses. Bangor Research Papers in Linguistics 1. 1-38.

Radford, A. 1988. Small children's small clauses. Transactions of the Philological Society 86. 1-43.

Saleemi, A. 1988. Learnability and Parameter Fixation. Doctoral Dissertation, University of Essex.

Sinclair, H. and J. Bronckart. 1971. SVO a linguistic universal? Journal of Experimental Child Psychology 14. 329-348.

Stromswold, K. 1988. Linguistic representations of children's wh-questions. Papers and Reports in Child Language 27.

Wells, C. G. 1985. Language Development in the Preschool Years. Cambridge: Cambridge University Press.