Hawkins-2001-The Theoretical Significance of Universal Grammar in Sla

The theoretical significance ofUniversal Grammar in secondlanguage acquisitionRoger Hawkins University of Essex

The evidence that native language acquisition is possible only becausechildren are born with an innately- determined language faculty – UniversalGrammar – is considerable. The evidence that the same innate ability isinvolved in second language acquisition (SLA) by older learners issuperficially less clear. There are differences both in the context ofacquisition and the nature of development. One recent approach suggeststhat only ‘poverty of the stimulus’ phenomena – where neither the firstlanguage (L1) nor the second language (L2) are possible sources for L2representations – can provide incontrovertible evidence for UniversalGrammar (UG) in SLA. I argue that while ‘poverty of the stimulus’phenomena are important landmarks in theory development in SLAresearch, they are not the most compelling reason for assuming theinvolvement of UG. More compelling are attempts to explain L2–L1differences. The latter are likely to lead to real progress, not only inunderstanding the nature of SLA but also the structure and organization ofthe language faculty itself.

I Introduction

The context in which second language acquisition (SLA) occursdiffers from first language acquisition (FLA) along a number ofdimensions:

1) In SLA another language is already present.2) Other components of mind have already matured, whereas

arguably FLA and the development of other cognitivecapacities go hand in hand.

3) Input is usually encountered differently, and may involvewritten as well as spoken language.

The contextual differences inevitably give rise to differences in thecourse of development of L2 grammars. The issue to be addressedin this article is why, despite the differences between second and

© Arnold 2001 0267-6583(01)SR189OA

Address for correspondence: Department of Language and Linguistics, University of Essex,Wivenhoe Park, Colchester, Essex CO4 3SQ, UK; email [email protected]

Second Language Research 17,4 (2001); pp. 345–367

at BEIJING FOREIGN STUDIES UNIV on July 28, 2013slr.sagepub.comDownloaded from

http://slr.sagepub.com/

346 The theoretical significance of UG in SLA

first language acquisition, much of the work in the ‘mentalist’tradition initiated in studies like those of Corder (1967), Selinker(1972) and Dulay et al. (1982) would assume that UniversalGrammar (UG) is inextricably involved, even in SLA which takesplace beyond adolescence.

One reason, which has stimulated considerable recent work, is‘poverty of the stimulus’ (POS) phenomena. These are cases wherethere is evidence for L2 grammatical representations which arederivable neither from properties of the speaker’s first language(L1) nor deducible from input. Some theorists have suggested thatonly POS phenomena can provide incontrovertible evidence for thepresence of UG in the grammars of L2 speakers (Dekydtspotter etal., 1997; Dekydtspotter and Sprouse, 2000; Schwartz and Sprouse,2000). I argue that while POS phenomena are important forestablishing the UG-constrained nature of L2 mentalrepresentations, they are not the most compelling reason for itsinvolvement. The reason is that establishing that the‘epistemological space’ (to borrow a term from a reviewer) withinwhich L2 representations are constructed is defined by UGdoes little to advance our understanding either of SLA itself,or of UG more generally. A more compelling reason, I suggest,comes from the hypotheses that result from attempts toexplain L2–L1 differences from a UG perspective. (I take theexplanation of L2–L1 differences to have been a central goal of theearliest generative work on SLA.) Such hypotheses represent realprogress in understanding the nature of SLA and may even shedlight on the structure and organization of the innate languagefaculty.

The article is organized as follows. In Section II there is a briefhistory of how the field arrived at a position where POSphenomena are seen as the only source of incontrovertible evidencefor Universal Grammar in SLA. Section III discusses the limitationsof POS studies, and argues that ‘difference-oriented’ approachesoffer the potential for advancing understanding of both SLA andUG.

II Universal Grammar and the significance of L2–L1 differences

One definition of UG runs along the following lines: UG is thesystem of principles and computational procedures which defineand place limits on the form that grammars for human languagescan take. A ‘grammar’ for a particular language is the set ofprocedures (traditionally called ‘rules’) which generate all and onlythe grammatical sentences of that language, with appropriate



Roger Hawkins 347

specifications for their sound structure and meaning.1 Byhypothesis, UG is independent of the specific grammars thatindividuals (unconsciously) construct on contact with samples oflanguage (the ‘primary linguistic data’). Also by hypothesis, theprinciples and computational procedures of UG are specific to thelanguage faculty and are genetically determined. These geneticallydetermined principles and procedures constrain the form thatparticular grammars can take so tightly that acquisition by L1learners proceeds uniformly, rapidly and with full success (undernormal conditions).2 Where parameter values need to be fixed inthe course of development, these are resolvable from easilydetectable properties of the input.

In early studies which seriously investigated whether thisgenetically-determined language faculty is involved in (post-childhood) SLA, observable developmental differences between L2and L1 learners played a central role. Such differences include theinfluence exerted by the L1 at the onset of acquisition and insubsequent development, and apparent fossilization of L2behaviour in some areas in some learners prior to the achievementof full target-like behaviour (among other differences; ninedifferences between the two types of language learners are listedby Bley-Vroman, 1989: 43–49).

The differences led some researchers at various times to theconclusion that UG is not involved at all in the construction of L2mental grammars by older learners, so that SLA is ‘fundamentallydifferent’ from FLA (Felix, 1985; Clahsen and Muysken, 1986; Bley-Vroman, 1989). A recent representative of this line of thinking isNewmeyer (1998: 74) who, on the basis of the differences, suggeststhat ‘the null hypothesis, surely, is that adult-acquired L2s are notsystems of grammatical competence’. Interestingly, he uses thisposition as an argument to strengthen the claim that nativegrammars are autonomous representational systems, separate from

1 It is important to stress that ‘grammatical’ is not the same thing as ‘sensible’ or‘interpretable’. As a reviewer points out, this has been emphasized in generative work sinceits inception. Grammatically well-formed sentences may make little sense (e.g., Bachelors aremarried women) and perfectly interpretable sentences may be ungrammatical (e.g., Thereseemed someone to be in the room).2 A reviewer suggests that rapidity in FLA is not an argument per se that development isUG constrained; speed has more to do with ‘fast neural growth in the first year of life,processing windows, etc.’ However, given the range of possible hypotheses about languagethat a child could entertain, only a theory proposing some version of an innate UG wouldseem to predict the speed at which development in fact occurs (i.e., apparent acquisition ofall the major structural properties of a language between 31/2 and 5 years). This feat is evenmore remarkable if not all of the neural connections required for language are in place atbirth, and if children have less space in working memory than adults, as some have proposed(e.g., Crain and Lillo-Martin, 1999: 27).




the cognitive mechanisms responsible for language use. If UG issubject to a critical period – so that speakers cannot construct UG-constrained representations beyond a certain age – then nativegrammars must be distinct from the mechanisms of languageuse, because mechanisms of language use continue to beavailable throughout life. Older L2 learners can use theirknowledge of the L2, but that knowledge is derived from othercognitive abilities and, hence, is fundamentally different from thatof native speakers.

For those who maintained, unlike Newmeyer, that the grammarsof adult L2 learners are, indeed, ‘systems of grammaticalcompetence’ derived from the interaction between UG and primarydata, the differences between SLA and FLA acquisition were amajor explanatory challenge. This challenge appears to have beenvery much to the fore in early generative work on SLA. Forexample, one of the problems addressed in a study of thedevelopment of knowledge of preposition stranding in the L2English of Dutch speakers reported by van Buren and SharwoodSmith (1985: 19) was the extent to which ‘the learner’s knowledgeof UG contribute[s] to the acquisition process in SLA’, where theissue was how the properties of the L1 and the constraints of UGinteract to produce L2-specific representations. In the same (first)issue of Second Language Research, White discusses the possibilitythat, as the result of the transfer of L1 parameter settings, theprinciples of UG ‘do not work in identical fashion in L1 and L2’(1985: 1). The refutation by du Plessis et al. (1987) of Clahsen andMuysken’s (1986) ‘fundamental difference’ account of the SLA ofGerman word order was centrally concerned with how adult L2 andchild L1 developmental differences could be captured in UG-derived L2 grammars. Eubank’s (1994) theory of ‘valueless features’is a UG-based attempt to explain why L2 initial-state grammarsmight differ from L1 initial-state grammars, as is Vainikka andYoung-Scholten’s ‘minimal trees’ theory (1994; 1996; 1998).Although many of these accounts explicitly defend the claim thatUG is fully available to (post-childhood) L2 learners, I will refer tothem as ‘difference-oriented’, because they attempt to explain whythere might be observable differences between child L1 learnersand adult L2 learners on the basis of how UG, the L1, primary dataand perhaps other factors interact.

The simple dichotomy of the ‘access to UG’ question addressedin work of the 1980s and early 1990s – ‘is it available to adult L2learners or not?’ – which forced researchers to give some accountof L2–L1 differences, has been superseded in more recent work bythe idea that if it can be shown that L2 representations fall within



Roger Hawkins 349

the ‘epistemological space’ defined by UG, then whether or notL2 and L1 speakers have the same representations for aparticular target language is irrelevant. What is important iswhether those representations are constrained by the principlesand properties allowed by UG (Dekydtspotter et al., 1998). That is,work should aim to determine whether or not L2 mentalrepresentations meet the criterion of being UG-derived (Epstein etal., 1996; Dekydtspotter and Sprouse, 2000; Schwartz and Sprouse,2000).

By implication, L2–L1 differences are no longer central. Thiswork has inevitably focused on POS phenomena – where L2grammatical representations are underdetermined both by the L1and by the L2 input – as the ideal case within which to test such ahypothesis. To illustrate, consider the well-known study byMartohardjono (1993) of locality constraints on wh-phrasemovement in L2 grammars (also reported in Epstein et al., 1996).The locality constraints imposed by UG have as a consequencethat sentences like the following are more or less ungrammaticalfor native speakers of English (examples from Epstein et al.,1996):

1) a. * Which mayor did Mary read [the book that praised ___]?(Movement out of a relative clause)

b. * Which soup did the man leave the table [after the waiterspilled ___]?(Movement out of an adjunct)

2) a. ? Which patient did Max explain [how the poison killed ___]?(Movement out of a wh-island)

b. ? Which car did John spread [the rumour that the neighbourstole ___]?(Movement out of a noun complement)

In early studies of the accessibility of these locality constraints toL2 speakers of English whose L1s do not have overt wh-movementin questions (and, hence, in whose L1s the issue of how far wh-constituents can move overtly is not raised), the focus was onwhether or not speakers displayed target-like intuitions about theseungrammatical cases. In other words, the criterion for determiningwhether L2 grammars are UG-derived was an absolute one: if UGconstrains the operation of wh-phrase movement in L2 grammars,then speakers should respond in the same way as native speakers.Typically it was found that older L2 speakers did not respond inthe same way as native speakers unless their L1s had movementlike English. This led to claims that adult L2 speakers at best have




partial access to this area of UG (Bley-Vroman et al., 1988;Schachter, 1990; Johnson and Newport, 1991).

Martohardjono (1993) argued that the ‘absolute’ criterion is toostrict. Assuming the model of UG proposed in Chomsky (1986),examples like (1) and (2) display different types of localityconstraints. The violations illustrated in (1) (extraction from non-argument positions) lead to ‘strong’ ungrammaticality, while theviolations in (2) (extraction from argument positions) lead to ‘weak’ungrammaticality. If L2 speakers’ intuitions about the relativestrength of ungrammaticality are investigated, it turns out that thosewho are sufficiently advanced to have acquired grammatical wh-phrase movement make significant distinctions between the twotypes. In a study of L1 speakers of Chinese, Indonesian and Italianwho were deemed to be advanced learners of English,Martohardjono found that on a grammaticality-judgement task allof the groups were making significant distinctions between thestrong and weak violations, even though only the Italian speakerswere performing in absolute terms like the native English-speakingcontrols (see Table 1).

The inference that Epstein et al. (1996: 688) draw from theseresults is that ‘although L2 learners may lag behind native speakerswith regard to accuracy rates, their judgements of wh-structuresmay still derive from their knowledge of UG principles’. Theexplanation for the differences between the groups lies not inwhether their grammatical representations are different, but infactors inherent in the experimental task. As Martohardjono (1998:155) more recently puts it ‘we would expect variousextragrammatical factors to intervene in this, as in any other typeof task, with the result of depressing L2 learners’ accuracy rates vis-à-vis native speaker rates’.3

Table 1 Relative rejection rates of ‘strong’ and ‘weak’ constraints on movementin English by speakers of different L1s

Language group Strong violations (%) Weak violations (%)

English controls 99 78Italian 91 62Indonesian 88 46Chinese 75 44

Source: based on Martohardjono, 1993.

3 Usha Lakshmanan (personal communication) points out that it is possible withexperimental sentences like Which patient did Max explain how the poison killed? that L2speakers may be accepting them as cases of short wh-movement (e.g., where which patienthas come from the indirect-object position of explain: Max explained (to) which patient howthe poison killed). However, two pieces of evidence suggest that L2 informants are not in



Roger Hawkins 351

A similar, but nevertheless different, type of POS argument thatrepresentation in L2 grammars falls within the epistemologicalspace defined by UG can be found in Schwartz and Sprouse (2000).Their study addresses the question of how SLA research shouldrespond to revisions in linguistic theory that affect specific SLAproposals based on earlier versions of the theory. (This is a questionthat had already been posed by van Buren and Sharwood Smith(1985: 21) as: ‘When linguistics coughs, should second languageacquisition studies catch pneumonia?’) Schwartz and Sprouse arguethat cases where L2 speakers construct an interlanguagerepresentation which is found neither in the L1 nor the L2 – but isfound in one of the world’s languages – are instances of POS andwill remain instances of POS whatever the linguistic theory (sinceany theory will need to account for the fact that this property occursin a natural language). One example comes from their own(Schwartz and Sprouse, 1994) study of the acquisition of Germanword order by an L1 Turkish speaker, Cevdet. Turkish is a verb-final language (both in main and embedded clauses) while Germanis a verb-second language (in main clauses). For a period of aboutone year Cevdet’s grammar for German allowed verb-second withpronominal subjects but not with non-pronominal subjects(examples from Schwartz and Sprouse, 2000: 172):

3) a. in der Türkei der Lehrer kann den Schüler schlagenin the Turkey the teacher can the pupil beat‘In Turkey, the teacher can hit the pupil.’

b. dann trinken wir bis neun Uhrthen drink we until nine o’clock‘then we drink until nine o’clock.’

The property of subject–verb inversion with pronominal subjectsalone – whatever the appropriate linguistic explanation for it – isfound neither in Turkish nor in German. Crucially, however, it isfound in other languages; it is found in French, for example, wherein questions it is possible to invert pronominal subjects and verbs,but not non-pronominal subjects:

fact doing this. First, Johnson and Newport (1991), in a study discussed in Section III of thepresent article, conducted an interpretation task to test precisely this question and found noevidence that L2 speakers treated experimental sentences as cases of short wh-movement.Secondly, the fact that all of Martohardjono’s groups differ in their judgements of violationsof strong and weak constraints suggests that they are not treating experimental sentences ascases of short movement. If they did, they would not differ in their judgements on the twotypes.




4) a. Le professeur bat-il ses élèves?The teacher beats he his pupils‘Does the teacher hit his pupils?’(*Bat le professeur ses élèves?)

b. Bat-il ses élèves?Beats he his pupils?‘Does he hit his pupils?’

This kind of POS is different from the POS observed in localityconstraints on wh-phrase movement. In the case of localityconstraints there is no positive evidence in the input bearing on thematter whatsoever. By contrast, in the case of subject–verbinversion Cevdet constructs a grammar by selecting a subset of thepositive evidence he is presented with. German offers positiveexamples of verb-second both with pronominal and non-pronominal subjects. Cevdet has ‘misrepresented’ these data byselecting only the cases of pronominal subjects for possibleinversion, but he has done this in a way that is found in other extantnatural languages. On this basis, Schwartz and Sprouse (2000: 175)conclude ‘that regardless of the precise formulation of syntactictheory, Cevdet’s Interlanguage behaviour can be accounted for onlyby assuming that his Interlanguage hypotheses are guided by UG’.

Let me refer to studies like these as ‘POS-oriented’ accounts ofhow UG constrains grammatical representations in SLA, todistinguish them from the ‘difference-oriented’ accounts of accessto UG described earlier. The shift in the research agenda from‘difference-oriented’ to ‘POS-oriented’ approaches isunderstandable. First, despite an assumption in the early‘difference-oriented’ studies that L2–L1 differences only affectinitial-state and transitional-state grammars and are temporary, itwas nevertheless observed that there are properties that sometimesgive rise to persistent L2–L1 differences. This is unexpected if L2grammars are UG-derived. Such was the finding, for example, inthe early studies of locality constraints (Bley-Vroman et al., 1988;Schachter, 1990; Johnson and Newport, 1991). Secondly, thecontinuing appeal of the fundamental difference hypothesis amonggenerative linguists not working on SLA (like Newmeyer, forinstance, who is probably among the group likely to be the best-informed about SLA research) obviously called for more ‘vigorous’defence of UG involvement in post-childhood SLA. Thirdly, shiftsin linguistic theory that undermine specific ‘difference-oriented’proposals for L2 phenomena are troublesome, as Schwartz andSprouse (2000) point out.

However, while the concentration of research effort on POSphenomena appears to provide a strategy for countering claims that



Roger Hawkins 353

SLA is ‘fundamentally different’ from FLA, and for avoidingdifficult issues of how acquisition research handles changes intheories of UG, it is problematic in other ways. It obviously doesnot, of itself, offer an account of how SLA differs specifically fromFLA. In that respect it does not go beyond offering glimpses of UGin SLA. Once it is accepted that UG is involved, a POS approachoffers nothing else. Moreover, it probably cannot contribute to ourunderstanding of UG, since the rationale of the approach is to findphenomena that already exist in mature native grammars. For thesereasons, while ‘POS-oriented’ studies are invaluable for providingevidence that post-childhood L2 grammars are UG-derived – hencerepresent an important stage in theory development in SLAresearch – they may not offer the most compelling reason why SLAresearch should assume that L2 grammars are the product of theinnate language faculty.4 In Section III, I argue that ‘difference-oriented’ research, which was central in early generativeapproaches, offers a more compelling reason for UG-basedresearch.

III UG, SLA and explanation

‘POS-oriented’ studies of locality like Martohardjono’s describedabove (and related studies: Uziel, 1993; Li, 1998; White and Juffs,1998) are interested in the relative sensitivity of L2 learners tostrong and weak locality violations. All the cited studies find suchrelative sensitivity, clearly suggesting that L2 learners’ mentalrepresentations are UG-derived, and all also find differences inlevels of accuracy between nonnative and native informants. Thequestion is: Beyond showing that UG can be ‘glimpsed’ in the kindsof representations that L2 learners construct, does this sort of POSstudy advance our understanding of (1) UG; (2) SLA? The answerappears to be ‘no’ in both cases. The operation of the locality

4 A reviewer questions whether the strong UG hypothesis can be assumed on the basis ofexisting evidence, and suggests that a function of ‘POS-oriented’ studies is to provide suchevidence in a ‘more direct and potentially convincing fashion’ than was available in‘difference-oriented’ studies of the 1980s and early 1990s. This implies that there is somepoint at which sheer weight of evidence will tip the balance in favour of the argument thatL2 representations are UG-derived. My own view currently is that it is unlikely that such a‘persuasion point’ will ever be reached, given the complexity of SLA data and the numberof variables involved, and that assuming the UG hypothesis is more an act of faith thanevidence-supported persuasion. Undoubtedly, studies that demonstrate that there are POSphenomena in SLA strengthen the credo of such an act of faith. The point that is made inSection III is that if – for the purposes of theory construction and in the pursuit of researchquestions – the decision is taken that L2 representations are UG-derived, at that point theinvestigation of POS phenomena per se offers no more insight than the comparativeinvestigation of mature native grammars.



constraints is by hypothesis the same in both native and nonnativegrammars. If we wish to deepen understanding of how principles ofUG interact with language-specific properties in such cases wewould do better to look at native grammars, where results fromperformance tests are subject to fewer variables and are muchclearer. This kind of ‘POS-oriented’ study also makes L2–L1differences of peripheral interest: L2 learners ‘lag behind’ nativesbecause they have not had the same amount of time to establish atarget-like lexicon or target-like processing routines, or they maybe suffering from a ‘production problem’ of unspecified nature(Epstein et al., 1996: 688 and 706). So it tells us nothing about SLA;only investigation of how input affects development or howperformance interacts with competence would tell us about SLA,and there has been an absence of detailed proposals about thesematters in the literature.

In the case of Schwartz and Sprouse’s proposal a similar problemarises. Although they describe a different kind of POSphenomenon, where L2 speakers ‘misrepresent’ positive evidencein a way which produces a grammar consistent neither with the L1nor the L2, but with a third naturally-occurring language, theiraccount does not advance our understanding of UG. Crucially forthem, the interlanguage representation constructed must be‘mirrored in at least one natural language grammar’ (2000: 175).Given this requirement, research might as well focus on thecomparative syntax of native grammars, since no property willemerge in POS-oriented studies of SLA that cannot be found innaturally occurring native grammars. Again such properties areclearer in native grammars than in L2 grammars.5

In terms of furthering explanation in the field of SLA itself,Schwartz and Sprouse offer a potential account of transfer whichcould at least take forward understanding of initial-state L2grammars. They suggest that, with respect to a given linguisticfeature, research should compare the developmental paths of L2speakers of typologically different L1s. If there is divergentdevelopment, then this constitutes evidence for transfer; if there isuniform development, then this would constitute evidence against


5 A reviewer points out that Schwartz and Sprouse make no claim that SLA research in itscurrent state can contribute to the development of the theory of UG. The argument is thattheoretical linguists generally have not been convinced that L2 grammars are UG-constrained, and until they are (by POS-oriented studies of SLA) it is unrealistic to expectSLA research to make any contribution to the development of UG theory. This suggests thatat some future point POS-oriented studies may contribute to the development of UG theory.The argument in the text, however, is that POS-oriented studies, by their nature, have nogreater potential for offering insight into UG than the comparative investigation of maturenative grammars, a domain in which the data are much clearer than they are in SLA.



transfer. (For earlier proposals along similar lines, see Zobl, 1980;1982.) While true, it does not help us understand which features arelikely to be affected (because, as we know from the many studiesof development undertaken since the 1970s, not all features aresubject to L1 influence) nor why just these features and not othersshould be affected.

In contrast to studies that seek out evidence for the fundamentalidentity of L2 and native grammars, consider the potential impacton our understanding both of SLA and UG of trying to explain thedifferences found between SLA and FLA. Several views of theextent of UG involvement are compatible with ‘difference-oriented’approaches. If a ‘full access to UG’ position is adopted, then thetask is to begin constructing a theory of performance (if differencesresult from production problems) or a theory of the interaction ofinput and acquisition (if differences are the effect of L2 speakers‘lagging behind’ natives). This raises direct questions about thenature of the ‘interface’ between (components of) UG and othermodules of mind. Full access accounts must do this because if it isaccepted that UG is available in the same form in SLA and FLAthe differences between them could only arise at the interfacebetween UG-derived grammars and other mental systems. (For arecent and innovatory attempt to construct a theory of theinteraction between input and acquisition see Carroll, 1999; 2000.)In this case results will not only produce testable proposals aboutwhy and how SLA is different from FLA, but also how UG relatesto other components of mind. SLA would then provide a differentkind of ‘window’ on UG; that is, a view that is different from thestudy of native grammars, from FLA and from studies of disorderedlanguage. SLA research would make a contribution to thedevelopment of the theory of UG, rather than seeking out clonesof constructs found elsewhere.

Alternatively, suppose that ‘difference-oriented’ studies adopt theview that UG itself ceases to operate in precisely the same way asit does in childhood. The interactions between UG and othermodules of mind remain constant (so that SLA research will notnecessarily contribute anything to the understanding of interfacerelations). It then becomes of interest to ask whether the changescan be linked to identifiable components of UG. If this turned outto be the case, it would not only produce testable proposals abouthow SLA differs from FLA, but also about what might count as adiscrete module in UG. Since this is a line of research that I believeoffers considerable promise, I will illustrate in slightly more detailwith two examples. (Whether UG is assumed to be fully availableor not, the point remains the same that it is ‘difference-oriented’

Roger Hawkins 355



studies which are likely to be more fruitful in deepeningunderstanding than ‘POS-oriented’ studies.)

Consider again the case of locality constraints. In ‘POS-oriented’accounts differences between nonnative and native grammars arerecognized, but are attributed to performance or production factors.However, it is clear that the differences are L1-related.Martohardjono’s results in Table 1 suggest that, although alllearners make a distinction between weak and strong violations oflocality constraints in English, the L1 speakers of Italian are closerto target norms than speakers of Indonesian or Chinese. A‘difference-oriented’ approach that assumes that UG has changedin some way must ask: What has changed to give rise to theseeffects? It cannot simply be a problem of resetting whateverparameter is involved in wh-phrase fronting in questions, becausethe learners in the study accepted grammatical questions involvingwh-phrase fronting which were not violations of locality. Hence, insome sense they have reset a parameter. This forces us to look atother possibilities. In a related study of locality violations in the L2English of Chinese speakers, Johnson and Newport (1991)incorporated alongside a grammaticality-judgement task aninterpretation test. The results of this test have gone largelyunnoticed in the subsequent literature. Seven native speakers ofChinese and 6 native speakers of English were presented withwritten sentences like (5a) and asked questions like (5b):

5) a. The policeman knew where the thief hid the jewels.b. What did the policeman know where the thief hid?

Observe that the question here weakly violates locality because anargument has been moved out of a wh-island. The test involvedboth weak and strong violations, and its purpose was to ensure thatthe Chinese speakers were not accepting violations in thegrammaticality-judgement task because they were assigning suchsentences different interpretations from native speakers.Informants’ responses were scored as ‘correct’ (jewels would bethe correct answer in this case), as ‘other interpretation’ (if theygave some other response) or as ‘don’t know’. Results are givenin Table 2.

While the results show that Chinese speakers are not assigning‘other’ interpretations to sentences which violate localityconstraints to any significant degree (hence the difference betweentheir judgements of grammaticality and those of native speakerscannot be attributed to the sentences ‘meaning somethingdifferent’) they also show, surprisingly, that the Chinese speakers




are much better at interpreting strong violations than nativespeakers are. Strong violations are the result of movingnonarguments too far, whereas weak violations result from movingarguments too far. Roberts (1997) has suggested that localityviolations involving arguments are awkward but interpretable,whereas violations involving nonarguments are difficult even tointerpret. The results of the native controls in Table 2 are consistentwith this claim. But the results of the Chinese speakers are not,suggesting that relative clauses do not have the same adjunct-likestatus for them as they do for native speakers.

A similar kind of result from an interpretation task is reportedin Li (1998). Li compared the performance of 180 L1 Chinese-speaking university students of English in China with 25 nativeEnglish-speaking controls. The test required informants to readshort stories of 4 or 5 sentences in length. They were then askedquestions in which a wh-phrase could have been extracted frommore than one position. For example, one story was about a boywho had forgotten his grandmother’s birthday and who asked hismother if she would telephone his grandmother with him. Followingthe story informants were asked: Who did he ask ___ [to call ___]?where there are two possible responses: his mother (correspondingto the object gap after ask) and his grandmother (corresponding tothe object gap after call). Some of these ambiguous questionsinvolved ungrammatical wh-phrase extraction of arguments andnonarguments; for example:

6) a. Who did the boy ask a [how to help ?b ]?b. When did John know a [how to fix his bike *b ]?

Roger Hawkins 357

Table 2 Mean responses of subjects interpreting wh-phrases on a localityviolation comprehension task

Extraction from L1 Correct Other Don’t knowinterpretretation

N complement Chinese 3.57 0.14 0.28(weak violation) English 3.83 0.00 0.16

wh-island Chinese 3.28 0.28 0.43(weak violation) English 3.00 0.00 1.00

Relative clause Chinese 2.57 0.14 1.28(strong violation) English 1.16 0.00 2.83

Note: The maximum potential correct score for the interpretation of eachconstruction type is 4.Source: adapted from Johnson and Newport, 1991: 242.



Since responses that informants can give to the questions arepotentially ambiguous on the basis of the stories they have heard,examples like (6) involve a conflict between what is interpretivelypossible (both gaps) and what is syntactically possible (only the firstgap gives rise to clear grammaticality). Li scored the results in termsof informants’:

• acceptance of just the interpretation associated with the ‘a’ gaps;• acceptance of both interpretations ‘a’ and ‘b’, and• acceptance only of interpretation ‘b’.

The results are given in Table 3. (There were two tokens involvingpossible weak violations and two tokens involving possible strongviolations, hence the total potential number of responses for eachinterpretation is 360 for the Chinese informants and 50 for thenative controls.)

While both sets of informants are about equal in interpreting the‘b’ position as a possible target for the interpretation of the wh-phrase when a weak violation is involved, the Chinese speakers arebetter at interpreting this as a potential target when a strongviolation is involved. Again, this suggests that the constructions inquestion may not have the same structural status for the nonnativespeakers as for the natives: the Chinese speakers may not beinterpreting them as extractions of nonarguments.

At least two possibilities come to mind for explaining thedifference. The first is that for Chinese speakers overt movement isnot involved. Instead, wh-phrases are merged in a topic-likeposition either with TP or CP, and are co-interpreted with a nullpronominal (rather than moving and binding a variable). This is anidea proposed in White (1992: 457–58):


Table 3 Informant responses to the interpretation of argument and adjunct wh-phrases (percentage; proportions in brackets)

Sentence type: Who did the boy ask a [how to help ?b ]?

L1 Only a Both a and b Only bChinese (n = 180) 51.9 (187/360) 43.9 (158/360) 1.7 (6/360)English (n = 25) 56.0 (28/50) 34.0 (17/50) 8.0 (4/50)

Sentence type: When did John know a [how to fix his bike *b ]?

L1 Only a Both a and b Only bChinese (n = 180) 59.4 (214/360) 30.0 (108/360) 5.8 (21/360)English (n = 25) 78.0 (39/50) 14.0 (7/50) 2.0 (1/50)

Source: based on Li, 1998.



Instead of presuming that L2 learners have the same representation as nativespeakers of English (i.e. wh . . . t) . . ., we should consider other possiblerepresentations for such sentences consistent with their linguistic behaviour. . . I propose that, when they appear to accept subjacency violations, theyanalyse the English sentences as containing a base-generated wh-phrase, witha base-generated null resumptive pro in the gap.

Since Chinese does not have overt wh-phrase movement inquestions, and it is standardly assumed that overt movement to thespecifier position of functional categories is driven byuninterpretable features made available optionally by UG(compare Chomsky, 1998), the absence of moved wh-phrases inChinese–English interlanguage could be attributed to the absenceof feature-driven overt movement in Chinese. It might then besuggested that UG changes beyond childhood in such a way thataccess to optional uninterpretable features is lost unless they havebeen encoded in the L1. On this interpretation, SLA research wouldprovide potential evidence to support the claim that there areoptional uninterpretable features in UG.

A second possibility is that the Chinese speakers’ Englishgrammars involve a different kind of wh-phrase movement fromsyntactic feature-driven movement, namely scrambling. One viewof scrambling is that it is optional movement involving‘reconstruction’ of the unmoved structure at LF (Saito and Fukui,1998); it is, hence, not subject to syntactic locality constraints. Thereason why Chinese speakers might adopt a ‘scrambling’representation for fronted wh-phrases in English would be thesame: unavailability of the optional uninterpretable feature whichdrives syntactic wh-phrase movement.

In both cases the fact that Chinese speakers do make distinctionsbetween the acceptability of weak and strong locality violationswould need to be the result of something other than feature-drivensyntactic movement. There is, however, evidence that differencesbetween weak and strong violations exist at the level of LF inChinese, in that subjacency violations (weak violations) appear notto hold while ‘Empty Category Principle’ violations (strongviolations) do (see Huang, 1982). If LF is universally invariant,Chinese speakers would simply be carrying over interpretivepossibilities that are hard-wired into UG, and judging Englishsentences accordingly.

Observe that we have here testable proposals about SLA itself:that speakers of languages without overt wh-movement will notconstruct representations involving feature-driven movement, butwill draw on other resources made available by UG for modelling

Roger Hawkins 359



fronted wh-phrases (‘merge’ and pronominal binding, orscrambling). This might lead us to investigate other cases of feature-driven movement – verb raising, verb second, noun raising – witha fresh eye. Where it appears, superficially, that L2 speakers whoseL1s do not have feature-driven movement have acquired thatproperty in the L2, is this really the case or are they using someother UG-derived operation? We also have possible evidence tosupport the idea that optional uninterpretable features form anatural subclass within UG, as already noted.

The second example concerns two apparently unrelatedphenomena, but which under one construal turn out to be relatedto the locality constraint case:

• an example of ‘selective fossilization’ in the L2 English of aChinese speaker described by Lardiere (1998a; 1998b; 2000);6

• inconsistency in marking gender concord in the L2 French ofEnglish speakers (Hawkins, 1998).

Lardiere’s informant arrived in the USA at the age of 22 when herfirst consistent exposure to English as an L2 began. She wasrecorded in conversation after 10 years of residence, and twice more81/2 years later. The second 81/2-year period was one of near-totalimmersion in English. Lardiere reports the performance of theinformant on three properties, none of which havemorphophonological reflexes in Chinese: Case-marking ofpronouns (e.g., he–him–his, I–me–my, etc.), the marking of simplepast tense (e.g., ran, walked) and the marking of the 3rd personsingular -s. The results for the use of nominative Case-markedpronouns, simple past-tense marking of all verbs and 3rd personsingular -s marking of thematic main verbs are given in Table 4.

Hawkins examined L2 French data from native speakers of


6 Although the informant – Patty – is referred to in the text as a native speaker of Chinese,it should be noted that her linguistic background is varied (Lardiere, 1998a: 12). Her parentswere native speakers of two Chinese languages: Hokkien and Mandarin. Patty herself wasborn in Indonesia and learned Indonesian at school, and later Cantonese in mainland China.Patty herself considers Hokkien and Mandarin to be her native languages.

Table 4 Correct use of the morphophonological reflexes of three properties inEnglish by a Chinese speaker

Sample Nominative Simple past 3rd person singular -s

1 49/49 (100%) 24/69 (34.8%) 2/42 (4.8%)2 378/378 (100%) 191/548 (34.9%) 0/4 (0%)3 76/76 (100%) 46/136 (33.8%) 1/22 (4.5%)

Source: based on Lardiere, 1998a; 1998b.



English collected from a task where informants were asked to givean oral description of a short animated film. There were two groups,each consisting of 10 informants, one in the UK the other inCanada. The groups were comparable in that they were alluniversity students and advanced speakers with similar lengths ofexposure (although the Canadians had undergone a late immersionprogramme, while the UK informants had had tutored French plusa minimum of 6 months residence in France). The study focused onthe consistency of informants in using target-like masculine- orfeminine-marked articles: le (masculine) vs. la (feminine) for thedefinite article, and un (masculine) vs. une (feminine) for theindefinite. Results suggested that the errors made by informantswere not equally distributed. Typically in the speech of anindividual, one article from each pair was likely to be used in amore target-like way than the other, which was overgeneralized. Forexample, for a hypothetical target French set of un triangle (‘atriangle’), un cercle (‘a circle’), une pyramide (‘a pyramid’), une balle(‘a ball’), an individual might produce un triangle, un cercle, *unpyramide, une balle, where there is no incorrect use of une, but unis overgeneralized. However, the member of the pair which wasmost target-like varied across individuals.

The overall distribution of gender-marked articles is shown inTable 5. The number of errors made indicates that one member ofthe masculine–feminine pair is more ‘target-like’ than the other.Bruhn de Garavito and White (2000) found similar results in theL2 Spanish of (much less advanced) L1 French speakers.

Lardiere has argued, in the case of her Chinese informant, thatthe abstract syntactic representation for English tense andagreement is present, but that she has a problem ‘mapping’ thesyntactic representation onto the correct morphophonologicalreflexes. Evidence for this comes from her perfect use of Case-marked pronouns. Pronouns in Chinese are not overtly marked forCase (and indeed are often null, since Chinese is a null argumentlanguage). She has, therefore, established appropriate L2morphophonological reflexes in this case, even though L1 and L2

Roger Hawkins 361

Table 5 Number of errors made in producing the more and less target-likemembers of masculine–feminine articles by two groups of L2 speakers

Definite DPs Indefinite DPs

Group ‘more TL’ ‘less TL’ ‘more TL’ ‘less TL’

UK 0/88 (0%) 16/133 (12%) 3/76 (4%) 26/135 (19%)Canada 4/104 (4%) 19/108 (18%) 4/59 (7%) 38/96 (40%)

Source: based on Hawkins, 1998.



differ. On standard assumptions, nominative Case-marking islicensed (checked) between a DP and a finite tense category in aspecifier-head configuration. Since finite T is required for Case-marking, and the informant’s Case-marking is perfect, it is notimplausible to maintain that T is also abstractly specified for pasttense and agreement. The reason why the informant is much lesssuccessful in marking the morphophonological reflexes of the latteris that in English they are affixal, whereas Case-marked pronounsare not, and it is affixes that cause this speaker difficulty.

Bruhn de Garavito and White offer a similar explanation for thegender-concord case. They assume that the L2 speakers of Frenchand Spanish reported above have assigned gender features tonouns, and abstractly this gender feature ‘percolates’ to categorieswithin the DP, like the articles. The problem for the L2 speakersarises in mapping the fully-specified DP onto the appropriatemorphophonological reflexes of the articles.

Both of these accounts are attempts to explain L2–L1 differencesfrom a UG perspective. The underlying assumption is that themorphological component of UG has changed in some waybetween childhood and adulthood (or at least the ability of themorphological component to map syntactic representations ontomorphophonological forms). This then offers testable predictionsfor SLA research: We might expect to see older L2 learners havingdifficulty mapping syntactic representations to affixal reflexes inother domains. If correct, this line of enquiry is also consistent witha view that sees morphological structure as a discrete module ofUG, distinct from syntactic representation.

Hawkins (1998) has proposed a different interpretation of thegender-concord facts, which could be extended to the tense andagreement case. Suppose that ‘percolation’ of gender features fromthe noun to other constituents within the same DP is viewed as aninstance of feature checking involving an interpretable genderfeature assigned to N, and uninterpretable gender features onconstituents like articles which must be checked by N. If theuninterpretable gender features are made available as options byUG – in the sense that languages like English have not selectedthem – then English speakers learning French as an L2 may haveproblems accessing them. Because of this, learners have to modelthe gender-concord facts in a different way, perhaps usingassociative memory processes in the lexicon. It might be claimedthat the speakers in question establish a default article (eithermasculine or feminine) and then learn a list of exceptions. Such asystem would give rise to one member of each pair of gender-marked articles being overgeneralized, and the other largely target-




like, the kind of pattern observed. An extension of this idea to tenseand agreement would require that the morphophonological reflexesin English result from a checking relation between the category Tand the inflected verb. The absence of tense and agreementmorphemes in Chinese would be the effect of an absence ofuninterpretable features. Chinese speakers learning English as anL2 would then need to model tense and agreement markingdifferently; again, perhaps the uninflected verb would be a defaultform, and tense- and agreement-marked forms would be listed asexceptions. Selection of appropriate past-tense-marked and 3rd-person-singular-marked verb forms would then depend on whetheran L2 learner had listed these forms in the lexicon and on the levelof activation of the listed forms during lexical insertion intointended past tense / 3rd person singular contexts. Notice thatthis account is of the same order as the account of L2–L1differences in sensitivity to locality violations described above: It isthe absence of optional uninterpretable features in the L1 grammarthat forces L2 speakers to draw on other resources made availableby UG.

Whether either of these proposals – the ‘mapping problem’account or the ‘absence of uninterpretable features’ account – turnout to be correct or not, they are the kinds of hypothesis that canlead to progress in developing a theory of SLA. If learners are ableto construct target-like syntactic representations, but havepersistent problems mapping representations to morphophonology,then it suggests that we need to focus more research effort on thenature of L2 morphology. If L2 learners have difficulty withoptional uninterpretable syntactic features, then we need to find outmore about how other resources made available by UG aredeployed by L2 speakers. Similarly, in the case of UG; if eitherapproach is correct, this will provide evidence that the relevantcomponent of UG (morphology or uninterpretable features) havechanged since childhood. If both accounts are wrong, it may be thatUG has not changed at all, and that L2 research should be lookingat interface relations between the language faculty and othercomponents of mind more carefully. Whatever the outcome,‘difference-oriented’ studies will advance our understanding bothof SLA and of UG, whereas ‘POS-oriented’ studies can onlyconfirm that L2 grammars are UG-derived.

IV Conclusion

SLA research which seeks out ‘poverty of the stimulus’ (POS)phenomena in L2 grammars provides clear evidence that L2

Roger Hawkins 363



grammars are UG-derived. This is a finding of major importance incountering claims that SLA is ‘fundamentally different’ from nativelanguage acquisition. Nevertheless, POS phenomena in L2grammars are no more than ‘glimpses’ of properties that can befound in clearer form in native grammars. I have argued that a morecompelling reason for assuming that L2 grammars are UG-derivedemerges in studies that aim to explain L2–L1 differences. Suchstudies must account for the differences either in terms of changesin the way that UG interacts with other components of mind or interms of changes that occur in components of UG. Such work hasthe potential not only to contribute to our understanding of thedevelopment and nature of L2 grammars, but also to offer a viewof UG that complements insights gained from work on nativegrammars, native language acquisition and disordered language.

Acknowledgements

I would like to thank Usha Lakshmanan and an anonymousreviewer for perceptive comments on an earlier version of thisarticle and for useful advice on bringing out the arguments moreclearly. They are not to be held responsible for what remains.

V References

Bley-Vroman, R. 1989: What is the logical problem of foreign languagelearning? In Gass, S. and Schachter, J., editors, Linguistic perspectiveson second language acquisition. Cambridge: Cambridge UniversityPress, 41–68.

Bley-Vroman, R., Felix, S. and Ioup, G. 1988: The accessibility of universalgrammar in adult language learning. Second Language Research 4,1–32.

Bruhn de Garavito, J. and White, L. 2000: L2 acquisition of Spanish DPs:the status of grammatical features. Unpublished manuscript, McGillUniversity, Montreal.

Carroll, S. 1999: Putting ‘input’ in its proper place. Second LanguageResearch 15, 337–88.

–––– 2000: Input and evidence: the raw material of second languageacquisition. Amsterdam: John Benjamins.

Chomsky, N. 1986: Barriers. Cambridge, MA: MIT Press.–––– 1998: Minimalist inquiries: the framework. MIT Working Papers in

Linguistics 15, 1–56.Clahsen, H. and Muysken, P. 1986: The availability of universal grammar

to adult and child learners: a study of the acquisition of German wordorder. Second Language Research 2, 93–119.

Corder, S.P. 1967: The significance of learners’ errors. International Reviewof Applied Linguistics 5, 161–70.




Crain, S. and Lillo-Martin, D. 1999: An introduction to linguistic theory andlanguage acquisition. Malden, MA: Blackwell.

Dekydtspotter, L. and Sprouse, R. 2000: Syntax-semantics composition inEnglish-French interlanguage: a reply to Newmeyer on grammaticaldesign and second language epistemology. Unpublished manuscript,Indiana University, Bloomington, IN.

Dekydtspotter, L., Sprouse, R. and Anderson, B. 1997: The interpretiveinterface in L2 acquisition: the process-result distinction in English-French interlanguage grammars. Language Acquisition 6, 297–332.

–––– 1998: Interlanguage A-bar dependencies: binding construals, nullprepositions and universal grammar. Second Language Research 14,341–58.

Dulay, H., Burt, M. and Krashen, S. 1982: Language two. Oxford: OxfordUniversity Press.

Du Plessis, J., Solin, D., Travis, L. and White, L. 1987: UG or not UG, thatis the question: a reply to Clahsen and Muysken. Second LanguageResearch 3, 56–75.

Epstein, S., Flynn, S. and Martohardjono, G. 1996: Second languageacquisition: theoretical and experimental issues in contemporaryresearch. Brain and Behavioural Sciences 19, 677–714.

Eubank, L. 1994: Optionality and the initial state in L2 development. InHoekstra, T. and Schwartz, B., editors, Language acquisition studies ingenerative grammar. Amsterdam: John Benjamins, 369–88.

Felix, S. 1985: More evidence on competing cognitive systems. SecondLanguage Research 1, 47–72.

Hawkins, R. 1998: Explaining the difficulty of French gender attributionfor speakers of English. Paper presented at the 8th annualEUROSLA conference, Paris.

Huang, C-T. 1982: Logical relations in Chinese and the theory of grammar.Unpublished doctoral dissertation, MIT, Cambridge, MA.

Johnson, J. and Newport, E. 1991: Critical period effects on universalproperties of language: the status of subjacency in the acquisition ofa second language. Cognition 39, 215–58.

Lardiere, D. 1998a: Case and tense in the ‘fossilized’ steady state. SecondLanguage Research 14, 1–26.

–––– 1998b: Dissociating syntax from morphology in a divergent L2 end-state grammar. Second Language Research 14, 359–75.

–––– 2000: Mapping features to forms in second language acquisition. InArchibald, J., editor, 102–29.

Li, X. 1998: Adult L2 accessibility to UG: an issue revisited. In Flynn, S.,Martohardjono, G. and O’Neil, W., editors, 89–110

Martohardjono, G. 1993: Wh-movement in the acquisition of a secondlanguage: a cross-linguistic study of 3 languages with and withoutovert movement. Unpublished doctoral dissertation, CornellUniversity, Ithaca, NY.

–––– 1998: Measuring competence in L2 acquisition: commentary onpart II. In Flynn, S., Martohardjono, G. and O’Neil, W., editors,151–57.

Roger Hawkins 365



Newmeyer, F. 1998: Language form and language function. Cambridge,MA: MIT Press.

Roberts, I. 1997: Comparative syntax. London: Arnold.Saito, M. and Fukui, N. 1998: Order in phrase structure and movement.

Linguistic Inquiry 29, 439–74.Schachter, J. 1990: On the issue of completeness in second language

acquisition. Second Language Research 6, 93–124.Schwartz, B. and Sprouse, R. 1994: Word order and nominative Case in

nonnative language acquisition: a longitudinal study of (L1 Turkish)German interlanguage. In Hoekstra, T. and Schwartz, B., editors,Language acquisition studies in generative grammar: papers in honorof Kenneth Wexler from the 1991 GLOW workshops. Amsterdam:John Benjamins, 317–68.

Schwartz, B. and Sprouse, R. 2000: When syntactic theories evolve:consequences for L2 acquisition research. In Archibald, J., editor,Second language acquisition and linguistic theory. Malden, MA:Blackwell.

Selinker, L. 1972: Interlanguage. International Review of AppliedLinguistics 10, 209–31.

Uziel, S. 1993: Resetting Universal Grammar parameters: evidence fromsecond language acquisition of subjacency and the empty categoryprinciple. Second Language Research 9, 49–83

Vainikka, A. and Young-Scholten, M. 1994: Direct access to X’-theory:evidence from Korean and Turkish adults learning German. InHoekstra, T. and Schwartz, B., editors, Language acquisition studies ingenerative grammar: papers in honor of Kenneth Wexler from the 1991GLOW workshops. Amsterdam: John Benjamins, 265–316.

–––– 1996: Gradual development of L2 phrase structure. Second LanguageResearch 12, 7–39.

–––– 1998: the initial state in the L2 acquisition of phrase structure. InFlynn, S., Martohardjono, G. and O’Neil, W., editors, The generativestudy of second language acquisition. Mahwah, NJ: LawrenceErlbaum, 17–34.

van Buren, P. and Sharwood Smith, M. 1985: The acquisition of prepositionstranding by second language learners and parametric variation.Second Language Research 1, 18–46.

White, L. 1985: The acquisition of parameterized grammars: sub-jacency in second language acquisition. Second Language Research 1,1–17.

White, L. 1992: Subjacency violations and empty categories in secondlanguage acquisition. In Goodluck, H. and Rochemont, M., editors,Island constraints: theory, acquisition and processing. Dordrecht:Kluwer, 445–64.

White, L. and Juffs, A. 1998: Constraints on wh-movement in two differentcontexts of nonnative language acquisition: competence andprocessing. In Flynn, S., Martohardjono, G. and O’Neil, W., editors,The generative study of second language acquisition. Mahwah, NJ:Lawrence Erlbaum, 111–29.




Zobl, H. 1980: Developmental and transfer errors: their common basesand (possibly) differential effects on subsequent learning. TESOLQuarterly 14, 469–79.

–––– 1982: A direction for contrastive analysis: the comparative study ofdevelopmental sequences. TESOL Quarterly 16, 169–83.

Roger Hawkins 367



Hawkins-2001-The Theoretical Significance of Universal Grammar in Sla

Documents

language l2

language research

department of language

spoken language

language acquisition

native language acquisition

slafirst language acquisition

nature of sla