.

The poverty of the stimulus argument and structure-dependency in L2 users of English

.

Vivian Cook 
Spelling data
Writing Home
SLA Home

IRAL 2003, 41, 201–221

 

 

Abstract

Testing the principle of structure-dependency in second language acquisition means showing that L2 learners know structure-dependency in questions regardless of whether their L1s have syntactic movement. A grammaticality judgment test examined relative clauses, questions with relative clauses and questions with structure-dependency violations with 140 mixed L2 learners of English and 35 native speakers. All L2 learner groups overwhelmingly rejected the ungrammatical   sentences with structure-dependency violations, only 9 individuals scoring less than 5/6 correct. While groups with L1s with movement (Finnish, Polish, Dutch) and those without (Japanese, Chinese, Arabic) differed significantly, these were variations within a high level of success. Structure-dependency is therefore active in all L2 learners, with some residual effect from the L1: L2 users know something which they have not acquired from outside their own minds.

 

1. Introduction

 

One of the compelling arguments for the innateness of Universal Grammar (UG) is ‘Plato’s problem’, alias the poverty-of-the-stimulus argument: “How do we come to have such rich and specific knowledge, or such intricate systems of belief and understanding, when the evidence available to us is so meagre?” (Chomsky 1987: 33). People know more about language than they could have learnt from the samples of language they have encountered: where else could the knowledge have originated than inside their own minds? This resembles a modern version of the theological “argument by design” (Paley 1802): the world is so complex that it could not have come into being of its own accord and so must have a designer.

 

1.1. The nature of the poverty-of-the-stimulus argument

To use the poverty-of-the-stimulus argument to show the innateness of a particular aspect of syntax means going through the following stages (Cook 1991):

A. demonstrating that a native speaker knows this aspect of syntax;

B. showing that this aspect of syntax was not learnable from the language evidence typically available to all children;

C. arguing that this aspect of syntax is not acquired from outside the mind, say by correction or explanation by the child’s parents;

D. concluding that this aspect of syntax is therefore built-in to the child’s mind.

Interpreted in the terms used by Pullum and Scholz (2002: 9),

    Stage A addresses the “acquirendum” – establishing what the speaker knows;

    Stage B the “lacunae” – what sentences are missing from the language input;

    Stage C “inaccessibility” – evidence that the lacuna sentences are not actually available to the learner – and “positivity” – the unavailability of indirect negative evidence;

    Stage D “indispensability” – the argument that the acquirendum could not be learnt without access to the lacunae. In other words “if you know X, and X is underdetermined by learning experience, then the knowledge of X must be innate” (Legate and Yang 2002: 153).

 

A similar argument applies to second language (L2) users: if they possess L2 knowledge they could have acquired neither from the language input they have encountered nor from the first language (L1) they already know, where else could it have come from other than the Universal Grammar in their minds (Cook 1991)? The variations from L1 acquisition occur in steps B and C: L2 users may meet different types of input from L1 children and they already, by definition, have an L1 in their minds. The argument can be reformulated as:

 

A'. demonstrating that an L2 user knows some aspect of L2 syntax;

B'. showing that this aspect of L2 syntax was not learnable from the language evidence typically available to L2 learners;

C'. arguing that this aspect of L2 syntax is neither acquired from outsidethe mind, i.e., from teacher’s explanations or corrections, nor transferred from the L1 already present in the mind;

D'. concluding that this aspect of L2 syntax is therefore built-in to the L2 user’s mind.

 

The arguments about the relevance of Universal Grammar to L2 users have taken many twists and turns as generative models and syntactic descriptions have changed. Yet the poverty-of-the-stimulus argument has remained a constant in generative language acquisition theories: if L2 users demonstrably know things they have not encountered in their first or second languages or from other input, these must come from within their own minds. It does not matter whether the syntactic knowledge that is acquired is fully understood by linguists. So long as the knowledge exists and could not come from external input from others in one form or another, then it must be an integral part of the mind, regardless of the fact that it happens to be manifested in a second language rather than a first. The poverty-of-the-stimulus argument has been critically discussed by Pullum and Scholz (2002), who tease out a number of internal sub-arguments, arguing more against the evidence that has been presented as proof than against the argument itself: “linguists have not achieved what they are widely thought to have achieved” (Pullum and Scholz 2002). This paper is then an attempt to put forward actual experimental evidence for an innate principle manifesting itself in L2 users’ grammars.

 

From Chomsky (1971) to Crain and Lillo-Martin (1999) the archetypal example of such innate knowledge is the principle of structure-dependency: “all known formal operations in the grammar of English, or of any other language, are structure-dependent” (Chomsky 1971: 30). Movement is a traditional concept in grammar. The question in (1) is related in some way to the statement in

(2):

(1) Is John going?

(2) John is going.

The relationship can be conceptualized as moving the element is from some prior position to the one in which it actually appears in the question. Many recent theories of syntax have assumed that the actual form of the sentence we encounter differs from an underlying form by having elements in different places, what Chomsky (2000: 23) calls “the pervasive fact that phrases are interpreted as if they were in some different position in the structure where such items are actually sounded”. Movement relates this surface sentence to the underlying structure in which the elements are in the positions necessitated by the grammar. Understanding a sentence is using a transposition code to find the plaintext in the coded message.

 

The principle of structure-dependency requires that the element to be moved must have a particular structural role in the sentence, not simply be in a particular place in its linear order. Thus the rule for English question movement must specify which element in the structure is moved, not which word in the sequence or which type of word. It is the fact that is is an auxiliary within the structure of the sentence that means it can move from (3) to (4), not that is is the second word.

(3) John is going.

(4) Is John going?

Furthermore only the copula be in the main sentence can be moved, not the copula in the subordinate clause, so that (5) becomes (6), not (7):

(5) Sam is the cat that is black.

(6) Is Sam the cat that is black?

(7) *Is Sam is the cat that black?

The element that is moved to form a question – an auxiliary or a copula – must then occur in a particular structural role rather than in a given linear position: “the rules of language do not consider simple linear order but are structuredependent . . .” (Chomsky 1988: 45). The only counter-example cited to this is Serbo-Croatian in which clitics move to be the second element, if necessary within a constituent (Comrie 1990).

 

The particular sentence-type with relative clauses and copula verbs has been the Leitmotif in Chomsky’s discussion of structure-dependency, occurring “eight or nine times” according to Pullum and Scholz (2002). Chomsky (1980), for example, contrasts (8) with (9):

(8) Is the man who is here tall?

(9) *Is the man who here is tall?

Chomsky (1988: 42) uses a Spanish equivalent (10):

    (10) Está el hombre, que está contento, en la casa?

            Is    the man,       who is     happy,     at  home?

           ‘Is the man who is happy at home?’

 

Step A of the poverty-of-the-stimulus argument involves demonstrating that native speakers know structure-dependency, usually taken for granted because native speakers invariably reject structure-dependency violations in this archetypal question form and never seem to make this kind of error in speech.

 

Step B means establishing that structure-dependency is unavailable to children in their usual environment from external input because they do not encounter sentences which violate structure-dependency. This step is in fact challenged by Pullum and Scholz (2002) for structure-dependency on the grounds that children encounter many sentences that illustrate the movement of the main-clause auxiliary, such as (11):

(11) Where’s the other doll that was in here?

However, as Lasnik and Uriagareka (2002) point out, this is evidence for auxiliary fronting, not structure-dependency.

 

Step C then argues that, on the one hand, parents never explain structure-dependency to their children, on the other, that parents cannot correct the children’s deviant utterances simply because children do not produce sentences that violate structure-dependency, and so there is nothing to be corrected. Crain and Nakayama (1987) indeed tested the production of yes/no questions by 30 English-speaking children aged 3:2 to 5:11 and found no mistakes at all that could be ascribed to structure-independent movement.

 

Step D concludes that structure-dependency is not acquired from evidence supplied by the world outside the learner’s mind and must therefore be inherent in the children’s minds.

 

The L2 steps in the argument are similar. Step A' means showing that L2 users know structure-dependency; Step B' that it was not available to them in the L2 samples they have heard; Step C' that it is not acquired from their environment or from their first language; Step D' that the only source left is therefore their own minds. If L2 users know structure-dependency, this shows that at least one principle that forms part of L2 knowledge is innate. Step C' has then more qualifications in L2 acquisition than Step C in L1 acquisition as some L2 users might conceivably have had the principle explained or taught to them by a language teacher. Though these are logical possibilities, they are unattested and about as plausible as suggesting parents teach structure-dependency explicitly to L1 children. Or L2 learners might encounter structure-dependency in a grammar book. Apart from introductions to generative linguistics, grammar books never mention structure-dependency, certainly none of those designed for students of English as a foreign language. A more plausible source of structure-dependency in Step C' might be the first language that is already in the learners’ minds: L2 users could transfer the principle from one language to another, and it would be no concern of L2 research how it was acquired in the first place.

 

The usefulness of structure-dependency to test this issue is that some languages do not have syntactic movement in questions. English and Dutch for instance form questions with movement and hence show structure-dependency in questions. Arabic forms questions by inserting question particles rather than moving elements; it does not need structure-dependency in questions. Dutch L2 users of English could transfer structure-dependency to English questions directly from Dutch questions; Arabic learners users could not. Structure-dependency for questions is not then incorporated in the competence of children acquiring languages like Arabic as first languages. So speakers of these are effectively tabula rasa for structure-dependency in L2 questions: if they show knowledge of structure-dependency, it could not come from their L1, thus enabling Step C' to be fulfilled. This is not of course the same as saying that structure-dependency is missing from Arabic or indeed from any language, since it is involved in many other aspects of the grammar. As principles are now held to apply across a range of constructions, it is indeed impossible to isolate the effects of any principle to a single test construction. Hence the poverty-ofthe-stimulus argument has to some extent been weakened by the inability to show in absolute terms that a principle is absent from a grammar without in effect describing the whole grammar.

 

Most second language acquisition research into the principles and parameters version of Universal Grammar has investigated whether parameters are reset from L1 to L2, whether the parameters are loosely linked to principles,as in the case of the pro-drop parameter, or are involved in the operation of the principle itself. Little has been said about whether the principles themselves actually exist in the L2 user’s grammar. Tsimpli and Roussou (1991) take principles to be the unvarying element in the second language; learners reset or wrongly set L2 parameters because of the ways in which their first languages work, but do not vary the principles themselves.

 

Actual L2 research into structure-dependency has been sparse. The exception is research by Naoi (1989), which involved first teaching English relative clauses to 11 teenage Japanese schoolgirls in the ninth grade and then asking them to change relative clause sentences into questions; ten of the girls succeeded without breaking structure-dependency, although not necessarily producing the exact questions expected. Despite structure-dependency not being required for Japanese questions, these L2 users clearly knew it in English. Naoi (1989: 65) claims therefore “there is virtually no difference between L1 and L2 as far as structure-dependence is concerned”. One issue with this research is that, according to Naoi (1989: 71), in the Japanese educational system “the question formation rule is one of the items learned at early stages while the relative clause is supposed to be studied in the third grade of junior high school, when the learners are 14–15 years old”. The research did not test the girls’ first attempts at question formation by movement (it is hard to do a year of English without encountering questions!) but their first attempts at making questions out of relative clauses.

 

The present research attempts to provide more extensive evidence for this celebrated test-case from a wider range of L2 learners. It is one outcome from a project to widen the scope of evidence in the second language acquisition discussions of Universal Grammar by testing a range of syntactic points across L2 users of English with different L1s. The specific aim here is to test the extent to which structure-dependency is known by L2 users of English regardless of their first language. The first goal is to establish that all the L2 users know structure-dependency in questions: this supports Step A_ of the argument, i.e., that it is actually acquired, but does not rule out its source being the first language already present in the user’s mind (even if that in turn is based on an innate principle) (Step C_). The second goal is to establish that L2 users with L1s without question-movement also show knowledge of structure-dependency in questions, hence ruling out straightforward transfer from one language to another. These two goals test the ubiquity of structure-dependency in L2 acquisition regardless of first language.

 

Universal Grammar is, however, a theory of grammars, not of languages; that is to say, it aims to account for the grammar in the mind of an individual, not the social construct of a language shared by a community of speakers: “The grammar in a person’s mind/brain is real [. . .] The language (whatever that may be) is not” (Chomsky 1982: 5). Hence, as Davies (1996) reminded us, it is important to see users of language not just as groups but as individuals. The ultimate question is whether there are individuals who do not know structure-dependency, not whether the average person in a group knows structure-dependency.

 

The research here is not intended to be specific to one particular framework of syntactic description. Culicover (1991) and Freidin (1991) have argued that structure-dependency derives from other general principles rather than being a principle in its own right.Within the mid-nineties Minimalist Program (Chomsky 1995), movement formed part of tree-formation and was driven by morphology in a different mode than in earlier models. But these developments are irrelevant to the poverty-of-the-stimulus argument, which applies to the phenomena called structure-dependency unless it can be shown either that native speakers do not know structure-dependency or that children acquire it from their environment. To say that a bird flies does not mean one has to know the precise structure of its wings. Nor is the behaviour of monolingual native speakers relevant; the poverty-of-the-stimulus argument would be equally valid for L2 users if the responses of native speakers and L2 users differed, so long as the L2 knowledge did not come from correction or explanation by others.

 

2. Research design

 

2.1. Research questions

The questions to be investigated, then, are whether:

   1. all L2 users know structure-dependency,

   2. L2 users with first languages that do not have question movement have more difficulty with structure-dependency than those with first languages that do.

 

The testable hypotheses are that all L2 users of English will show near-perfect knowledge of structure-dependency; and that there will be differences between the groups of L2 users with and without question-movement in their first languages. Confirming the first hypothesis would show that the principle of structure-dependency is present in all L2 users; confirming the second would suggest some effects of transfer from the L1.

 

2.2. Method

The research instrument used consisted of a grammaticality judgment test. This method poses many methodological problems (Birdsong 1989; Paolillo 2000). Cook (1993) drew attention to the instability of grammaticality judgments in L2 research because metalinguistic awareness is in itself changed by learning another language and to the dangers of confusing the linguist’s single sentence evidence based on intuition with grammaticality judgment evidence based on experiments. Nevertheless, grammaticality judgments are the conventional method in the Universal Grammar area of L2 research and hence grammaticality judgments results are comparable to other research carried out in this paradigm.

 

In the format used here the subjects were given a three-way choice of OK, Not OK, and Not sure, avoiding words such as grammatical or acceptable. The structure-dependency sentences were mixed with sentences testing various parameters of subjacency, adverb position and pro-drop, which will not be discussed here; effectively they will count as distracters.

 

2.3. Materials

The sentences testing structure-dependency (given in full in the Appendix) came in three types:

 

A. Relative clauses: these tested whether the subjects were able to handle ordinary relative clauses with copula is, such as (12):

(12) Joe is the dog that is black.

B. Questions with relative clauses: these tested whether the subjects could handle the same type of relative clause sentence with grammatical question movement of the copula is. In other words they were the correct questions with relative clauses that did not break structure-dependency, as in (13):

(13) Is Joe the dog that is black?

C. Structure-dependency violations in which the wrong copula has been moved: these tested structure-dependency violations by moving the is from the relative clause to the initial position. They are modelled on the classic Chomskyan examples and are equivalent to the B type questions except for movement involving a structure-dependency violation, such as in (14):

(14) Is Joe is the dog that black?

There were six examples of each sentence type, permuting a limited set of nouns and adjectives. These test sentences were mixed with other sentences in the same randomised order for all subjects. There were 96 sentences in all; the results of 18 are discussed here.

 

2.4. Subjects

The same test was distributed over a period of time to L2 students of English with six L1s and to native speakers of English. All the L2 subjects were university students of English.1 Since it was not possible to test the English proficiency of all the students, there may be variations between university level English in different countries; hence to some extent each group has to be treated independently. The groups consisted of 35 native speakers of English working or studying at an English university, 27 Japanese and 23 Finnish students visiting an English university for short courses, 26 Dutch students in the Netherlands, 22 Arabic-speaking students in Morocco, 22 Polish-speaking students in Poland and 20 Chinese-speaking students in Hong Kong: they were all therefore educated young adults.

 

2.5. Languages

It is, needless to say, hard to establish the details of question movement within similar grammatical models across a range of seven languages. In terms of each language used:

– Japanese yes/no questions are formed by inserting the particle ka at the end of the sentence (Kuno 1973).

– Chinese yes/no questions are formed by adding the particles ma or ba at the end of the sentence (Po-Ching and Rimington 1997).

– Arabicyes/no questions are formed by adding a particle to the questioned element without movement (Wright 1874).

– Polish yes/no questions require movement when the question word czy is omitted but it is otherwise to some extent optional (Fisiak, Lipi_ska-Grzegorek, and Zabrocki 1978).

– Finnish yes/no questions involve movement and the addition of the particles -ko or (Karlsson 1983).

– Dutch yes/no questions have movement (Fehringer 1999).

– English yes/no questions have movement and do-support.

It is of course possible to take issue with the analysis for any individual language. These languages were selected so as to be as wide a range as possible, so that at least the claims would not be based on the peculiarities or uncertainties about a single language. They will then be treated as falling into two groups: languages with movement (Polish, Finnish, Dutch, English) and languages without movement (Japanese, Chinese, Arabic).

 

3. Results

 

The scoring was strict in that only positive evidence of knowledge counted: OK or Not OK were treated as correct as appropriate to the sentence type; Not sure and No response were counted as incorrect in addition to the overtly incorrectanswers. The correct answer for types A (relative clauses) and B (questions with relative clauses) was therefore OK, and for type C Not OK. The results will be presented first for the groups with different L1s, then for individuals.

 

3.1. Scoring by first languages

The results presented in Table 1 and Figure 1 are scored in terms of the numbers of correct responses for each sentence type.

 

Using a one-sample t test, all L1 groups scored significantly above chance on the type A relative clauses except for the Polish speakers; the scores for the English, Finnish, Dutch and Japanese were significant on the type B questions but not for the Arabic, Chinese and Polish speakers; on the type C structuredependency violations the scores for all groups were significant at p. < 0.001.

 

All groups strongly reject violations of structure-dependency, and accept the other two sentence types less strongly. To make the data easier to visualize, they are displayed in Figure 1 in percentage scores.

 

Comparing the three types of sentence, type A relative clauses are relatively difficult for all groups, ranging from 58.3% correct for Polish students to 81.8% for the Finnish speakers with the English natives scoring 95.7%. Type B questions are slightly more difficult than type A for all groups including the native speakers, ranging from 50% for the Polish students to 81.8% for the Finnish students with 91.9% for the English speakers; only the Dutch have a slight advantage of 80.1% for questions over 78.8% for relative clauses. Type

 

 

C structure-dependency violations are the easiest for all groups, ranging from 86.7% correct for the Chinese to 100% for the Dutch, compared to the native speakers’ 99.5%; most groups have a scattering of one or two mistakes, the only groups scoring less then 95% being the Chinese with 86.7% and the Arabic speakers with 87.1%. The scores for type C structure-dependency violations then stand out from the rest.

 

Comparing the groups statistically, for type A relative clauses there is a difference between at least one pair of groups (Kruskal-Wallis chi-squared=38.66, df. 6, p <0.001), attributable to the English native speakers (Multiple Comparisons, p. < 0.05); for type B questions, there is again a difference between at least one pair (Kruskal-Wallis, chi-squared = 48.30, df. 6, p < 0.01), attributable to two groupings: English versus Arabic, Chinese, Poles, Japanese, Dutch and Finnish versus Arabic speakers and Poles (Multiple Comparisons, p. < 0.05); for type C structure-dependency violations there is a difference between at least one pair of groups (Kruskal-Wallis chi-squared = 54.65, df. 6,

p. < 0.001).

 

Let us now compare the two groups of L1s with and without question movement, i.e., Dutch, Polish and Finnish versus Arabic, Chinese and Japanese.Using a t test, the differences are not significant for type A relative clauses and type B questions, but reach significance for type C structure-dependency violations (p. < 0.001); using a Mann-Whitney test, there is no significant difference for type A, some significance for type B (p. < 0.028) and a high level of significance for type C (p. < 0.001).

 

3.2. Scoring by individuals

We now turn to the scores expressed in terms of the number of subjects scoring at particular levels across all groups. The raw figures showing how many people scored at each level from 0–6 for the three sentence types are given in Table 2 for the L2 users.

 

As shown in Figure 2, both type A relative clauses and type B questions have a gradual rise from left to right with a much lower proportion of individuals obtaining 6 out of 6; type C structure-dependency violations, however, peak sharply at 6, with some users obtaining 5, and very few getting lower scores. There is a significant difference between type C and both types A and B (Pairwise comparisons test, with Bonferroni adjustment for multiple comparisons, p. < 0.001) but the differences between types A and B are not significant at thep. < 0.01 level.

 

Suppose we adopt the arbitrary criterion that knowing the structure means scoring 5 or 6 sentences correctly out of 6 (83.3% or better). Treating all the L2 groups together, type A relative clauses were mastered to criterion by 76 out of 140 subjects, type B Questions by 65 out of 140, and type C structure-dependency violations by 131 out of 140. These are displayed as percentages in Figure 3.

 

Clearly there is a large difference between sentence types A and B with 54.3% and 46.4% respectively, and the type C structure-dependency violations with 93.6%.

 

 

 

Using the same criterion of 5 responses or better out of 6, Table 3 breaks the L2 users down into groups and adds the native speakers of English. Again nearly every group has a high proportion of people scoring 5 or better for type C structure-dependency violations, far fewer for types A and B. These data are presented in Figure 4 in terms of percentages of individuals in each

group.

 

The individuals who meet the criterion of scoring 5 or more out of 6 for type A relative clauses range from 27.3% of the Arabic students to 70.4% of the Japanese and 91.4% of the native speakers. Percentages for type B questions range from 22.7% of the Arabic students to 65.2% of the Finnish students, compared to 94.2% of native speakers. In contrast the range for type C structure-dependency violations is from 80% of Chinese students to 100% of Polish, Finnish and Dutch students, with 100% of natives: the only groups with less than 90% are the Chinese with 80% and the Arab speakers with 80.9%. Again, when considered in terms of individuals, the type C structuredependency violation sentences stand out from the others.

 

4. Discussion and conclusions

 

As we saw earlier, though all the students were studying English at university, their English proficiency may have varied both between the groups and between individuals; differences in scores may relate to other uncontrolled differences in their backgrounds, even the improbable language teaching of structure-dependency. Yet the results for the type C structure-dependency sentences are highly similar for everyone: the overall scores for each group exceed 85 %, most are above 95 %; all the L2 groups have 80% or more members who meet the 5/6 criterion, 3 having 100 %. Put another way, across the whole test only 9 out of 140 subjects failed to meet the 5/6 criterion, i.e., 6.4 %. Reducing the criterion to 4/6 cuts it down to 5 subjects (3.6%). Clearly, whatever the differences between groups may have been, they all score well on structure-dependency.

 

Many of the subjects performed badly at accepting what were believed to be straightforward sentences without structure-dependency violations: students from most groups had difficulties with both the relative clauses and the questions. Yet, despite not knowing these, they could reject a structure-dependency violation almost unerringly. The ability to tell that the structure-dependency violation sentences were not grammatical does not apparently depend on the ability to handle relative clauses and questions; types A and B were indeed correlated but neither type A nor type B correlated to an appreciable extent with type C (Pearson correlation: A with B .732, A with C .167, B with C .270; p. < .01, two-tailed). The ungrammaticality of these sentences was recognizable to the subjects even if they were uncertain about relative clauses and question-movement themselves. Some proportion of these results could perhaps be explained in terms of the alleged asymmetry of grammatical and ungrammatical judgments (Birdsong 1994) since all the structure-dependency violations involved rejection of ungrammatical sentences, all the relative clauses and questions acceptance of grammatical sentences.

 

Knowledge of incorrect movement does not then necessarily build on a knowledge of relative clauses or of correct movement, unlike the assumptions of cumulative build-up made in, say, Johnson and Newport (1991). White (1989: 65) claims “there is no point testing for a universal principle if subjects have not mastered the kinds of structures in which that principle would operate”. This is not supported by the present results, which show far superior knowledge of structure-dependency questions over the questions and relative clauses that they seem to presuppose. In a sense, they show that knowledge of principles applies to sentences even when people have only partial knowledge of the elements that make them up. The growth of language knowledge is not cumulative by piling up constructions but may depend on some overall general principle, just as Marcus et al. (1999) showed that seven-month-old infants knew a principle of syntax before they knew any words of English.

 

A more general reservation concerns the problem of testing a ‘principle’ of Universal Grammar. Principles by definition apply across the board in the grammar: they are not construction specific. Hence it cannot be claimed that, say, Japanese as a whole lacks structure-dependency; it simply does not require it in questions. It may be that L2 learners can apply a broad principle to other constructions in a new language and can extrapolate from the use of structure-dependency in one part of the grammar to other parts. This would imply that there is effectively no difference between languages so far as structure-dependency is concerned and would effectively push the poverty-of-the-stimulus argument back on to first language acquisition; hence it would explain the differences in success rate between different L1s as comparative ease in extrapolating across constructions (Cook 1988). The argument that a principle could never be isolated and tested separately goes outside the brief here and would affect essentially all the areas that have been put forward as evidence for the poverty-of-the-stimulus argument in linguistics: the theory would not be capable of being disconfirmed from experimental evidence from one grammatical construction.

 

Since Universal Grammar theory concerns individual minds, not a collective body of language, the results were stated in terms of individuals as well as groups. Only nine individuals out of a hundred and forty failed to meet the 5/6 criterion for structure-dependency sentences, diverse as their other results may be – six Arabic speakers and three Japanese. Given the performance noise built-in to grammaticality judgment tests, this seems as low a percentage of unsuccessful individuals as one could reasonably hope for. Nevertheless there was a significant difference between the two groups with L1s with and without movement for structure-dependency but not for the other sentence types. These are, however, differences between high levels of success, not between success and failure.

 

It is hard to separate out the movement difference from other differences between the two groups of languages. The languages of the non-movement group also use different orthographic systems from English, whether characters or vowel-less alphabets, which are known to hinder L2 development in writing (Haynes and Carr 1990). There may also be vocabulary or cultural problems in the sentences specific to one group of users – Hong Kong students remarked that you cannot say Mary is the cat that is black because cats are not called Mary, perhaps supporting the view that Chinese speakers judge sentences more by word meanings than word order (Miao 1981). It may also be that the level of ‘university English’ was slightly lower in these groups. In other words the residue of 6.4% of subjects who did not seem to know structure-dependency might merit further investigation. But it hardly undermines the main finding that 93.6% of the L2 students tested knew structure-dependency using a fairly strict criterion, compared to the 54.3% who knew relative clauses and the 46.4% who knew questions.

 

To come back to the hypotheses, almost all the L2 users demonstrated a knowledge of structure-dependency using the standard grammaticality judgment method employed in this field. L2 users with L1s without syntactic movement know structure-dependency to a high level, even if they score slightly less than users with L1s with question movement. It would be hard to find research using a grammaticality judgments paradigm that gives a more clear-cut result. There may be faults with the grammaticality judgments method, in particular asymmetry of response (Birdsong 1994) and there are indeed other problems with grammaticality judgments (Paolillo 2000); it could have been carried out in various other ways, say with reaction timing (Cook 1994). Yet rejecting the method itself as unsound would undermine most second language acquisition research in the Universal Grammar paradigm.

 

On the other hand, perhaps, although structure-dependency remains “a basic tenet” in much contemporary work (Crain and Lillo-Martin 1999: 179), these sentences may not “really” be structure-dependency at all but some other syntactic phenomenon to be described in some totally different way. This would only affect the poverty-of-the-stimulus argument if the knowledge could be shown to have been acquired from some external evidence, as say Pullum and Scholz (2002) argue for first language acquisition. In terms of the basic argument, if L2 users are aware that these sentences are ungrammatical and this knowledge has not been acquired from input, what else could it be but built-in to their own minds? Rejecting this means either denying that the phenomena covered by structure-dependency exist, or proving that their source lies outside the mind, or claiming that the grammaticality judgments methodology is invalid, or rejecting the poverty-of-the-stimulus argument itself, all of which would be fatal not just to this experiment but to most second language acquisition research in the Universal Grammar perspective. L2 users clearly know a principle of Universal Grammar in their second language which they have not acquired from other people or from aspects of the world outside their own minds.

 

Perhaps, however, we are approaching the argument from the wrong perspective. Linguistics currently concerns itself with monolinguals and single grammars, not with multilinguals with more than one grammar in the samemind, and so finds the second language acquisition poverty-of-the-stimulus argument of marginal relevance. Yet arguably the normal human mind possesses more than one grammar whether statistically (probably there are more L2 users in the world than monolinguals) or logically (language acquisition models should explain what any child can do, not just those exposed to a single language) (Cook 2002). Roeper (1999: 3) argues that bilingualism occurs “whenever two properties exist in a language that are not statable within a single grammar”, in particular when children going from one ‘stage’ of development to another have two grammars with different properties and when adults with ‘optional’ rules are effectively choosing between two grammars, both cases where the speaker can choose between two possible Numerations whether speaking two dialects, two ‘stages’, two genres or two languages. This is then one interpretation of Chomsky’s occasional gnomic remarks on bilingualism, for example “ ‘multilingualism’ is a vague intuitive notion; every person is multiply multilingual in a more technical sense” (Chomsky 2000: 44).

 

From a learnability perspective Satterfield (1999) has argued that the standard account of language acquisition has to accommodate children who know two grammars with two sets of parameter-settings at the same time; acquisition theories cannot be restricted to children who have been brought up in an environment where they only hear one language: “a theory purporting to account for universal language learnability cannot be considered adequate if it excludes the non-monolingual speakers of this world” (Satterfield 1999: 137). The onus seems more on those who base linguistics on the human mind that knows a single grammar to defend their position than on researchers who start from the normal human condition, statistically and developmentally, to show that the language of L2 users is in some sense ‘natural’.

 

Cook (2002) takes the argument further by suggesting that, since the normal potential of the human mind is to know more than one language, linguistics should not base itself on minds with only one grammar or one set of parameter settings. This line of theoretical development suggests that the normal poverty-of-the-stimulus argument is in fact steps A'–D', i.e., the L2 scenario. The nub is whether any learner has had evidence of structure-dependency from either L2 or L1; claims for innateness cannot rest solely on the environment impoverished case of monolinguals. Grammars with and without structure-dependency for movement relate to each other within the same mind. The relationship itself is not at stake here; it is discussed further in Cook et al. (2003).

 

But the argument for the innateness of structure-dependency is stronger when it is based on evidence that 131 out of 140 L2 users of English with a variety of first languages are highly accurate at recognizing the deviance of English questions that violate structure-dependency, a construction which they have never previously encountered.

 

As we progressed through the above argument, it became clear that any research into the innate possession of a principle has to be qualified by uncertainties over a number of issues:

 

– Can a single principle or a single construction be isolated from the others and tested in terms of the steps of the poverty-of-the-stimulus argument, whether structure-dependency, or any other? Since all principles might potentially be involved in all constructions, testing a single principle adequately might at minimum involve testing the whole range of the grammar. It may be that the classic argument needs redefining in terms of the whole state of the grammar, not of a specific aspect. In which case, the possibility of putting the poverty-of-the-stimulus argument to actual test gets even more remote.

 

– Can the research method of grammaticality judgments bear the weight that is put on it? If the method is suspect and the measure changes for L2 users because of their superior metalinguistic awareness, grammaticality judgments may need supplementing in various ways. But it is hard toknow how the bulk of research in this field could manage without this basic research tool.

 

– Can the poverty-of-the-stimulus argument accommodate views of ‘universal bilingualism’ that see all speakers as having more than one grammar? The implication is that basing competence on monolinguals may be misleading; the potential multi-competence of human beings contains more than one grammar, whether separate or integrated in various ways (Cook 2002). Again this has repercussions not just for this experiment but for all research into the acquisition of first and second languages.

 

Like an earlier contribution (Cook 1969), this paper has been in a sense a ‘what if’ paper: if the poverty-of-the-stimulus argument is correct, if a single principle can be studied in isolation, if grammaticality judgments are a valid research technique, if separate grammars for L1 and L2 can be described rather than an integrated grammar for both, what are the consequences? This nevertheless renders it no more hypothetical than the vast majority of papers in the Universal Grammar paradigm of research into second language acquisition, which implicitly make such assumptions. Whatever reservations and doubts there may be about interpretation, the experiment clearly demonstrates something at the core of the poverty-of-the-stimulus argument, namely that L2 users know information about language that they are extremely unlikely to have learnt from experience.

 

Appendix

 

Test sentences

 

Type A. Relative clauses

Sam is the cat that is black.

Peter is the dog that is brown.

Sarah is the woman who is English.

Bill is the student who is French.

Joe was the man who was late.

Mary was the teacher who was early.

 

Type B. Questions with relative clauses

Is Joe the dog that is black?

Is Mary the cat that is brown?

Is Bill the man who is English?

Is Sarah the woman who is early?

Was Sam the teacher who was late?

Was Peter the student who was French?

 

Type C. Structure-dependency violations in which the wrong copula has been moved

Is Sam is the cat that brown?

Is Peter is the dog that black?

Is Sarah is the woman who early?

Is Joe is the student who late?

Was Bill was the man who French? Was Sarah was the teacher who English?

 

Note

I am grateful to the students and staff of the Universities of Essex, Fez, Utrecht, Krakow and the Chinese University of Hong Kong, for their co-operation with this experiment, to Phil Scholfield for his invaluable statistical help and comments, and to an anonymous reviewer for various helpful suggestions.

 

References

Birdsong, David (1989). Metalinguistic Performance and Interlanguage Competence. New York:

Springer.

— (1994). Asymmetrical knowledge of ungrammaticality in SLA theory. Studies in Second Language

Acquisition 16: 463–473.

Chomsky, Noam (1971). Problems of Knowledge and Freedom. New York: Pantheon.

— (1980). Rules and Representations. Oxford: Blackwell.

— (1982). Some Concepts and Consequences of the Theory of Government and Binding. Boston,

     MA: MIT Press.

— (1987). Transformational grammar: Past, present, and future. In Studies in English Language

     and Literature, 33–80. Kyoto University.

— (1988). Language and Problems of Knowledge: The Managua Lectures. Cambridge, MA:

MIT Press.

— (1995). The Minimalist Program. Boston, MA: MIT Press.

— (2000). The Architecture of Language. New Delhi: Oxford University Press.

Comrie, Bernard (1990). Second language acquisition and language universals research. Studies in

     Second Language Acquisition 12 (2): 209–218.

Cook, Vivian (1969). The analogy between first and second language learning. International Review

     of Applied Linguistics in Language Teaching 7 (3): 207–216.

— (1988). Language learners’ extrapolation of word order in phrases of micro-artificial languages.

     Language Learning 38 (4): 497–529.

— (1991). The poverty-of-the-stimulus argument and multi-competence. Second Language Research

     7 (2): 103–117.

— (1993). Linguistics and Second Language Acquisition. Basingstoke: Macmillan.

— (1994). Timed grammaticality judgments of the head parameter in L2 learning. In The Dynamics

     of Language Processes, G. Bartelt (ed.), 15–31. Tubingen: Gunter Narr.

— (2002). Background to the L2 user. In Portraits of the L2 User, Vivian Cook (ed.), 1–28.

     Clevedon: Multilingual Matters.

Cook, Vivian, Elisabet Iarossi, Nektarios Stellakis, and Yuki Tokumaru (2003). Effects of the second

     language on the syntactic processing of the first language. In Effects of the Second Language

     on the First, Vivian Cook (ed.), 214–233. Clevedon: Multilingual Matters.

Crain, Stephen and Diane Lillo-Martin (1999). An Introduction to Linguistic Theory and Language

     Acquisition. Oxford: Blackwell.

Crain, Stephen and Mineharu Nakayama (1987). Structure dependence in grammar formation.

     Language 63 (3): 522–543.

Culicover, Peter W. (1991). Innate knowledge and linguistic principles. Behavioral and Brain Sciences

     14 (4): 615–616.

Davies,William D. (1996). Morphological uniformity and the null subject parameter in adult SLA.

     Studies in Second Language Acquisition 18: 475–493.

Fehringer, Carol (1999). AReference Grammar of Dutch. Cambridge: Cambridge University Press.

Fisiak, Jacek, Maria Lipi_ska-Grzegorek, and Tadeusz Zabrocki (1978). An Introductory Polish-

     English Contrastive Grammar. Warsaw: Pa_stwowe Wydawnictwo Naukowe.

Freidin, Robert (1991). Linguistic theory and language acquisition; a note on structure-dependence.

     Behavioral and Brain Sciences 14 (4): 618–619.

Haynes, Margot and Thomas H. Carr (1990). Writing system background and second language

     reading: A component skills analysis of English reading by native-speaking readers of Chinese.

     In Reading and its development: component skills approaches, T. H. Carr and B. A.

Levy (eds.), 375–421. San Diego: Academic Press.

Johnson, Jacqueline. S. and Elissa Newport (1991). Critical period effects on universal properties

     of language: The status of subjacency in a second language. Cognition 39: 215–68.

Karlsson, Fred (1983). Finnish: An Essential Grammar. London and New York: Routledge.

Kuno, Susumu (1973). The Structure of the Japanese Language. Cambridge, MA: MIT Press.

Lasnik, Howard and Juan Uriagareka (2002). On the poverty of the challenge. Linguistic Review

     19: 147–150.

Legate, Julie A. and Charles D. Yang (2002). Empirical reassessment of stimulus poverty arguments.

     The Linguistic Review 19: 151–162.

Marcus, Gary F., S. Vijayan, S. Bandi Rao, and P. M Vishton (1999). Rule learning by sevenmonth-

      old infants. Science 283: 77–80.

Miao, Xiao-Chun (1981). Word order and semantic strategies in Chinese sentence comprehension.

     International Journal of Psycholinguistics 23: 109–22.

Naoi, Kazuhiro (1989). Structure-dependence in second language acquisition. MITA Working Papers

     (Keio University) 2: 65–77.

Paley, William (1802). Natural Theology. Cited in: Stephen J. Gould (1993). Eight Little Piggies.

     London: Jonathan Cape.

Paolillo, John (2000). Asymmetries in Universal Grammar: The role of method and statistics. Studies

     in Second Language Acquisition 22 (2): 209–228.

Po-Ching, Yip and Don Rimington (1997). Chinese: An Essential Grammar. London and New

     York: Routledge.

Pullum, Geoffrey K. and Barbara C. Scholz (2002). Empirical assessment of stimulus poverty

    arguments. The Linguistic Review 19(1): 9–50.

Roeper, Thomas (1999). Universal bilingualism. Bilingualism: Language and Cognition 2: 169–

     186.

Satterfield, Teresa (1999). Bilingual Selection of Syntactic Knowledge: Extending the Principles

      and Parameters Approach. Dordrecht: Kluwer.

Tsimpli, Ianthi and Maria Roussou (1991). Parameter-resetting in L2. UCL Working Papers in

      Linguistics 3: 149–189.

White, Lydia (1989) Universal Grammar and Second Language Acquisition. Amsterdam: John

     Benjamins.

Wright, William (1874). A Grammar of the Arabic Language. Cambridge: Cambridge University

     Press.