To what extent do you think innate linguistic knowledge is involved in second language learning?

The nativist perspective on second language acquisition

Roger Hawkins Department of Language and Linguistics, University of Essex, Wivenhoe Park, Colchester,

Essex CO4 3SQ, United Kingdom

Received 5 January 2007; received in revised form 20 February 2007; accepted 25 February 2007

Available online 2 July 2007

Abstract

Two central questions in SLA research are (i) what are the properties of an L2 grammar that give rise to

observed performance, and (ii) how does an L2 learner arrive at that grammar? Three arguments are offered

to support the view that answers to these questions cannot be achieved without assuming that L2 learners

have innate linguistic knowledge that determines the form their grammars take: constraints on grammars

that are specifically linguistic, the non-randomness of the features that learners identify in constructing

grammars, and selective persistent divergence from native speakers on properties for which there is positive

evidence in the input.

# 2007 Elsevier B.V. All rights reserved.

Keywords: Second language acquisition; Universal Grammar; Overt Pronoun Constraint; Article meaning; Phonemic

contrast; Verbal inflection

1. Introduction

Below is a transcript of a typical fragment of speech from a second language (L2) speaker of

English who was asked to describe a picture as if it were a scene outside his window:

(1) This morning at 7 o’clock when I look out the window and I saw the woman buying

cabbage on market stand next to my house. And that stand also sell potato and things

like that and there are peoples.

This speaker produces a number of forms that a native speaker would probably not have

produced: look with an apparently intended past tense meaning, but lacking the past tense

inflection –ed; market stand lacking an article; sell lacking a 3rd person agreement –s; potato

www.elsevier.com/locate/lingua

Lingua 118 (2008) 465–477

E-mail address: [email protected].

0024-3841/$ – see front matter # 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.lingua.2007.02.011

lacking a plural –s; peoples inflected for the plural where it is not required. Most of the other

properties of this speaker’s speech are target-like.

Two of the questions that any theory of SLA must answer about observations of this kind are:

what is the nature of the speaker’s mental grammar that would give rise to this performance?

How did he arrive at this grammar? An emergentist approach to answering these questions proposes

that L2 learners have mental capacities which, while not themselves ‘linguistic’, can extract

grammatical representations from experience. Factors that are particularly relevant to such

extraction in emergentist L2 learning are frequency of forms, cue competition, salience,

interference and blocking (for Ellis, 2002, 2006), and conservatism, analogical generalisation,

competition between forms and monitoring of indirect negative evidence (for MacWhinney, 2004).

As O’Grady (this issue: ‘The emergentist program’) points out, the mental capacities that

emergentists assume may well be innate and specific to humans (given that human language is

species specific), but the main claim is that they are not specifically linguistic in nature.

A nativist approach to the two questions, in contrast, proposes that they cannot be fully answered

without appeal to specific linguistic knowledge that does not appear to be available from experience

and which does not look like the effect of general properties of mind. This knowledge must

therefore be innate: Universal Grammar (UG). In the case of L2 acquisition there are three types of

evidence (at least) that point to the need to assume innate, specifically linguistic knowledge to

explain the properties of L2 speakers’ mental grammars and how learners arrive at those grammars.

Each is described in the remainder of this article. The observations made here are not new. They

have already been made by other researchers who have arrived at the conclusion that there are

aspects of second language speakers’ knowledge and their acquisition of that knowledge that can

only be explained by postulating that UG constrains L2 mental grammars.

2. L2 grammars are constrained by linguistic principles that cannot be derived

from non-linguistic sources

A number of studies in recent years have identified UG-specific constraints on L2 grammatical

knowledge that appear not to be deducible from L2 input, from a speaker’s L1, from instruction

or conscious learning, or from non-linguistic properties of mind. One example of this is a study of

the ‘Overt Pronoun Constraint’ by Pérez-Leroux and Glass (1997, 1999). The Overt Pronoun

Constraint (OPC) was originally proposed by Montalbetti (1984) as a putative general principle

associated with binding in human language. It can be informally stated as in (2):

(2) Where there is an alternation in a language between an overt/null pronoun, only the

null pronoun can take a quantified expression as antecedent (e.g. someone, nobody,

which student).

This can be illustrated by the following example from Spanish, a null subject language (examples

from Pérez-Leroux and Glass, 1997). Given the context in (3), (4a) and (4b) are possible

continuations. In (4a) a referential matrix subject is a possible antecedent for the pronominal

embedded subject; in (4b) a quantified matrix subject is a possible antecedent:

(3) Pedroj y Juani están participando en un concurso.

Peter and John are participating in a contest.

(4) a. Juani cree que éli/j/Øi/j ganará el premio.

John thinks that he/Ø will win the prize.

R. Hawkins / Lingua 118 (2008) 465–477466

b. Nadiei cree que él*i/j/Øi/j ganará el premio.

Nobody thinks that he/Ø will win the prize.

Crucially, the null embedded subject (Ø) can refer both to nadie and Juan, but the overt subject él

‘he’ can only refer to Juan. This follows from the Overt Pronoun Constraint. The restriction does not

apply where there is no alternation, and only él can appear, as in complement of a preposition

position:

(5) Todo el mundoi dice que el presidente habla de éli/*Ø

‘Everyone says that the president speaks about him’

It is not clear that English-speaking L2 learners of Spanish could infer this principle from

experience. They will encounter examples where, in subject position, Ø can refer to both

referential and quantified antecedents. They will also encounter examples where overt pronouns

like él can refer to quantified antecedents when the pronoun is in non-subject position, and

to referential antecedents when it is in any position. They would need to infer that él cannot refer

to a quantified antecedent just in the case where él is in subject position. This would appear to

require negative evidence, which is arguably not reliably available even to L2 learners.

Furthermore, English provides no clue to this restriction because overt pronoun subjects of

finite clauses freely refer to quantified antecedents:

(6) Nobodyi thinks that hei will win the prize

Moreover, there is little evidence that this phenomenon is taught in classrooms, or that Spanish

language teachers are generally even aware of it. Devices like tallying frequencies of forms in the

input or ‘conservative learning’ will not help here. Given examples in input like (4a) and (4b),

a learner would have to realise that the contrast between NPs that are referential and those that are

quantified is relevant for pronominal binding, just in the case where null and overt pronouns

alternate, before they can monitor the frequencies of forms or be conservative in restricting the

type of pronoun bound by a quantified NP. But this means they have pre-existing knowledge

equivalent to the OPC. It is not clear that domain general learning or processing principles would

fare any better.

Pérez-Leroux and Glass (1999) tested 3 groups of English speakers at 3 proficiency levels in

Spanish (elementary, intermediate and advanced) and 20 NS controls. The task they were given

involved a set of contexts in English such as:

In the O.J. Simpson trial, it is clear that the press has a negative bias against the defendant

in their reporting. Some journalist said that he was a wife beater.

This was followed by a sentence to translate into Spanish, where the first few words were already

given, for example:

But no journalist said that he is guilty.

Pero ningún periodista . . .

The property of interest here is how subjects translate the pronoun subject of the embedded clause

(. . . that he is guilty): as an overt or null form. The stories were designed so that half of them

R. Hawkins / Lingua 118 (2008) 465–477 467

favoured a referential antecedent for this pronoun (as in the example given) or the quantified

matrix subject of the clause to be translated. The prediction was that if informants’ grammars are

constrained by the OPC, they should be significantly less likely to produce an overt pronoun in

their translations where the favoured antecedent is the quantified matrix subject than when it is a

referential DP in the preceding story. Because of space limitations, results from just the advanced

proficiency group and the NS controls are considered here. The proportions of overt and null

pronouns supplied are presented in Table 1.

Each group produced more null pronouns than overt pronouns when the antecedent favoured

by the context was the quantified matrix subject, and more overt pronouns when the favoured

antecedent was a referential DP in the preceding context than when it was the quantified matrix

subject of the translated sentence. An analysis of variance involving all groups in the study

showed the difference in choice of pronoun in these two conditions to be significant, but there was

no significant interaction between group membership and choice of pronoun: in other words,

both L2 speakers and native controls were responding in similar ways. It appears that L2 speakers

of Spanish (whose L1 is English) are sensitive to the Overt Pronoun Constraint where they have

little evidence for it from their prior or current linguistic experience. Therefore their grammars

are constrained by UG in this domain.

Three objections might be raised to this conclusion, given the detail of the responses shown in

Table 1. The first is that the NS controls are allowing overt pronouns to refer to quantified

antecedents (14% in Table 1), which is perhaps unexpected if the OPC is really a UG-derived

constraint. In fact, responses from informants in performance tests, including responses from

native speakers, are rarely categorical (exceptionally, the L2 group in Table 1 are categorical in

this case in not supplying overt él with a quantified antecedent). Extraneous task variables such as

inattention, misinterpretation of the context and other factors may intervene. The important point

is whether subjects are making the relevant contrast, and not whether their judgements are

categorical. (See White, 2003:29–30 for discussion of this important point).

The second objection is that the L2 informants appear to be producing more null subjects than

the NS controls, particularly where the antecedent is a referential DP, and this might suggest that the

OPC is not what determines their responses. A comparison by Pérez-Leroux and Glass of the

‘proportional use’ of null subjects by each group (operationalized as the difference between use of

null subjects with referential antecedents and quantified antecedents, divided by total number of

null subjects used) found no significant difference between the natives and non-natives. Despite

appearances, their responses are not significantly different from those of the native controls,

statistically.

The third objection is that the responses could be given an alternative explanation on the basis

of informants choosing null pronouns when the antecedent is relatively ‘close’ to the target

pronoun, and choosing overt pronouns as the antecedent becomes more distant in the discourse.

This is a potentially serious challenge to the claim that the OPC is what determines choices: in the

R. Hawkins / Lingua 118 (2008) 465–477468

Table 1

Proportion of overt/null pronouns produced (Pérez-Leroux and Glass, 1999)

Antecedent Pronoun L2 Spanish (n = 18) Native Spanish (n = 20)

Quantified él 0% 14%

Ø 93% 85%

Referential él 32% 68%

Ø 58% 31%

task used by Pérez-Leroux and Glass, all the quantified antecedents are relatively close to the

embedded subject pronoun (being in matrix subject position in the same sentence), while all the

referential antecedents are somewhere in the preceding story. This possibility cannot be ruled out

on the basis of this study alone. However, the design of a second study, by Kanno (1997), offers a

control for this potential problem. Kanno investigated English speakers of intermediate

proficiency in L2 Japanese. Japanese also allows an alternation between null and overt subject

pronouns, and like Spanish appears to obey the Overt Pronoun Constraint. The task used to elicit

responses involved biclausal sentences where the subject of the embedded clause was either null,

or an overt pronoun. Informants had to decide who the null or overt subject of this embedded

clause referred to: the matrix subject or some other antecedent not mentioned in the test sentence.

Underlying the methodology was the assumption (taken from Grimshaw and Rosen, 1990:200–

201) that ‘intrasentential antecedents are strongly preferred when they are grammatically

permissible’ (Kanno, 1997:270). The task then offered informants sentences where the matrix

subject was either a referential or a quantified DP, for example:

(7) a. Dareka-ga [kare-ga/Ø Kaataa-san-o sitteiru to] ittemimasita yo

Someone-Nom [he-Nom/Ø Carter-Mr-Acc know that] was-saying

‘Someone was saying that he/Ø knows Mr Carter’

b. Tanaka-san-wa [kare-ga/Ø kaisya de itiban da to] itte-iru

Tanaka-Mr-Top [he-Nom/Ø company in best is that] saying-is

‘Mr Tanaka says that he/Ø is the best in the company’

If informants show a significant difference between allowing null and overt pronouns to take the

matrix clause subject as an antecedent when it is a quantified expression, in conflict with their

assumed preference for choosing an intrasentential antecedent, this would suggest that they are

constrained by the OPC. Results are presented in Table 2, in a similar format to those of

Pérez-Leroux and Glass.

The results are similar to those that Pérez-Leroux and Glass obtained with L2 speakers of

Spanish, suggesting that their findings do reflect the fact that L2 learners’ grammars are constrained

by the OPC, and were not the effect of the method used. Since different target L2s, different groups

of L2 learners at different proficiency levels and different methodologies were used in the two

studies, the results suggest that the phenomenon is robust, as White (2003:28) has observed.

A final objection that might be made is that the OPC is a rather stipulative and construction-

specific constraint that does not fit well with recent attempts within the spirit of the minimalist

program to identify more general principles of UG. This is certainly the case. The OPC would seem

to be something of an analytic convenience, describing a phenomenon that surely follows from a

more general principle determining the distribution of non-referring expressions. What is important

here is not so much what that principle might turn out to be, as the fact that the asymmetrical

R. Hawkins / Lingua 118 (2008) 465–477 469

Table 2

Proportion of overt/null pronouns selected (based on Kanno, 1997:272–273)

Antecedent Pronoun L2 Japanese (n = 28) Native Japanese (n = 20)

Quantified kare 13% 2%

Ø 79% 83%

Referential kare 42% 47%

Ø 82% 100%

treatment by both native and non-native speakers of possible antecedents, where overt and null

pronouns alternate, only receives an explanation if it is assumed that a principle of UG is involved. (I

thank Cris Lozano (personal communication) for reminding me of the importance of this point.)

There is a growing list of studies of other phenomena that appear to be under-determined by

experience in similar ways to the OPC, to which L2 speakers are sensitive and which seem to

require appeal to innate, specifically linguistic knowledge for explanation (see Dekydtspotter and

Sprouse, 2003, for a list of studies of such phenomena). Such cases are an explanatory challenge

for emergentist theories. They would need to show that general principles of learning or

processing that lack linguistic content are sufficient to allow learners to deduce the grammatical

principles in question.

3. L2 grammars can be different from the target L2 and from the L1, but are not infinitely variable

Input from an L2 may lead learners to establish grammars that are different from their L1, but

are also not the same as the grammars of native speakers of the target language. This can be a

temporary phase of development or a characteristic of a steady-state grammar (where further

development is unlikely). Where this happens, the non-target properties identified do not appear

to be random aspects of experience. Rather they are just those options hypothesised to be offered

by UG (because they are required to explain a property realised in some natural language). An

example of this emerges in a study by Ionin et al. (2004) of the acquisition of English articles by

speakers of L1s that lack articles (Russian and Korean). If a language learner were starting from

scratch to establish the meaning and distribution of the, a and Ø (no article) without any

pre-existing linguistic knowledge, the potential factors that could be relevant are very large, as

Ionin et al. (this issue) observe. Consider the in (8):

(8) Tom is waiting to get an autograph from the new champion.

Without some constraint on the hypothesis space, potential hypotheses a learner could entertain are:

use the when the noun is in clause-final position, when the noun is modified by an adjective, when

the noun is singular, when the noun is concrete, when the noun is human (or animate), when the

referent of the noun exists (or doesn’t exist, as in the case where the match deciding the new

champion has not yet been held), when the noun is mentioned for the first time in a discourse, when

the noun is in a PP, when it is in a non-argument position, when the noun ends in a nasal sound, and

so on.

It appears, however, that L2 learners identify just two properties as potentially relevant for

article choice in a language that has articles: definiteness and specificity. English articles encode

definiteness, but not specificity. This means that although the definite article can never be used

with indefinite meaning or the indefinite article with definite meaning (e.g. She ate the apple can

never have the interpretation ‘She ate an apple’, just as She ate an apple can never have the

interpretation ‘She ate the apple’), they can be used freely to indicate specific or non-specific

referents, as illustrated in (9) and (10). The two possible continuations for each sentence show

that the English articles are ambiguous between a specific and a non-specific reading:

(9) Tom is waiting to get an autograph from the new champion.

a. She is coming out to sign autographs at any moment. (specific)

b. He’ll have to wait for the match to finish to find out who that is. (non-specific)

R. Hawkins / Lingua 118 (2008) 465–477470

(10) John wants to marry a singer.

a. Her name is Kelly, and she lives down the street. (specific)

b. He hasn’t met one yet, though. (non-specific)

In some languages, apparently, articles encode the opposite: specificity rather than definiteness,

for example Samoan (Ionin, 2006).

Ionin et al. (2004) investigated knowledge of English articles by intermediate to advanced

proficiency L1 speakers of Russian and Korean using a forced-choice elicitation task. Subjects

were asked to read short scenarios in which therewas an NP lacking an article. The task was to select

one of the, a or Ø for the gap. Each scenario favoured one of four interpretations for that NP:

(i) definite specific (e.g. ‘She is meeting the principal of my brother’s elementary school. He is a

very nice man’) (ii) definite non-specific (e.g. ‘He is talking to the principal of his school, whoever

that is’) (iii) indefinite specific (e.g. ‘She is having dinner with a girl from class. Her name is Angie’

(iv) indefinite non-specific (e.g. ‘She is having dinner with a girl from class. But I don’t know who’).

Results show that the Russian and Korean speakers differ from native speakers in their use of

English articles: the is selected in both ‘definite’ and ‘indefinite specific’ contexts, and a is

selected in both ‘indefinite’ and ‘definite non-specific’ contexts. Proportions of responses are

illustrated in Table 3.

Notice that the Russian speakers are using a with a definite, non-specific NP in 33% of cases

and the with an indefinite, specific NP in 36% of cases. The Korean speakers are using a with a

definite, non-specific NP in 14% of cases and the with an indefinite, specific NP in 22% of cases.

However, this non-target use is not random. Speakers are considerably less likely to use a to refer

to definite, specific NPs or the to refer to indefinite, non-specific NPs.

Ionin et al. (2004) argue that ‘definiteness’ and ‘specificity’ are two innately-specified options

offered by UG for the grammatical representation of articles, and that some languages select just

one or the other (definiteness for English, specificity for Samoan). When language learners

encounter languages that have articles, the claim is that UG will lead them to assume that these

properties, and not whether the noun is in clause-final position, is modified by an adjective and so

on, are relevant for the representation of articles. For mature native speakers of English, only

definiteness is relevant. However, for L2 speakers whose grammars are still developing,

it appears that fluctuation between the two options is a possibility, and this gives rise to the

performance illustrated in Table 3. Ionin et al.’s subjects have grammars for English, therefore,

which are licensed neither by the L1 (which lacks articles) nor by the L2 (in which articles only

encode definiteness). At the same time their grammars do not allow random use of articles, nor do

they seem to entertain a range of possibilities that experience might allow. Usage follows from

the possibilities allowed by their innate linguistic endowment.

R. Hawkins / Lingua 118 (2008) 465–477 471

Table 3

Article choice (based on Ionin et al., 2004:30, Tables 12 and 13)

+def (target the) �def (target a)

L1 Russian (n = 26) the a the a

+spec 79% 8% 36% 54%

�spec 57% 33% 7% 84%

L1 Korean (n = 39)

+spec 88% 4% 22% 77%

�spec 80% 14% 4% 93%

The challenge for an emergentist account of these facts is to explain why just the property

‘specificity’ should have been identified along with ‘definiteness’ as potentially relevant to article

interpretation, and not other properties that, if knowledge is acquired only on the basis of learning

or processing mechanisms that are not specifically linguistic, would seem to be a priori relevant,

such as the type of noun, its position in the sentence, the type of referent, and other conceivable

alternatives. Again, there is a growing list of studies showing that L2 speakers arrive at grammars

different both from the L1 and the target L2 that are consistent with innately determined

linguistic options (see White, 2003:chapter 2.3 for discussion).

Note that the argument here supports a specifically linguistic nativist account of article

acquisition, even though fluctuation is almost certainly only a temporary stage of development

for L2 learners. With longer exposure to English, positive evidence that articles encode

definiteness will pre-empt the specificity feature (see Ionin et al., this issue, section 5.2 for

discussion).

4. L2 speakers do not fully establish properties for which they have positive evidence

from experience

4.1. Phonology

In the domain of phonology, work by Brown (2000) has shown that novel phonemic contrasts

may be differentially difficult for the same group of L2 speakers in ways that are difficult to

explain on the basis of experience alone. There are certain phonemic contrasts that are present in

English that are not present in Japanese, Mandarin Chinese and Korean: /s-u/, /l-r/ and /f-v/. All three languages have the phonemes /s/ (and Japanese and Mandarin have segments that

correspond to English /f/), but lack /u, v/. Furthermore, on Brown’s analysis, Mandarin Chinese and Korean have the phoneme /l/, but lack /r/, whereas Japanese has a flapped phoneme that is

traditionally equated with English /r/, but lacks the phoneme /l/ (although one of the allophonic

realizations of is a [l]-like sound).

In an auditory discrimination task, 15 Mandarin, 15 Japanese and 11 Korean speakers, all of

high intermediate to advanced proficiency in L2 English, together with 10 native English-

speaking controls, were asked to listen to pairs of invented one-syllable words, like those in (11),

and to indicate which of the pairs was different:

(11) a. (i) ra-la (ii) ra-ra

b. (i) fa-fa (ii) va-fa

All groups, non-native and native, correctly distinguished /f-v/ in over 90% of items. However,

the three non-native groups’ accuracy dropped below 80% with /s-u/, and the Japanese and Korean speakers’ accuracy dropped below 70% on the /l-r/ contrast. The native speaker controls

were over 90% accurate in all cases, and the Mandarin speakers were over 90% correct on the /l-r/

contrast. All groups differed significantly from natives on the /s-u/ contrast, but only the Japanese and Korean speakers did on the /l-r/ contrast.

Because none of the contrasts exists in the L2 speakers’ L1s, what is surprising is that the /f-v/

contrast has been acquired to native-like levels while the /s-u/ contrast has not, and the /l-r/ contrast has not for the Japanese and Korean speakers. What could explain this selective

influence of the L1? Brown argues that it is an effect of innate linguistic knowledge (i.e. UG). At

birth the language faculty provides a set of distinctive phonological features like [voice],

R. Hawkins / Lingua 118 (2008) 465–477472

[aspiration], [continuant], [coronal], etc., that allow the child potentially to represent any

phonemic contrast found in human language. When the child is exposed to a particular language

or languages, only a subset of this universal inventory is required to establish a phonological

representation. Unused features cease to be available to the learner at some point in development.

When a learner comes to acquire an L2 later, only the range of features selected in the L1(s) can

be used for representing the phonology of the L2. New phonemes are acquirable providing that

their feature composition involves features present in the L1, even if the phonemes are not present

in the L1. Brown claims that the feature inventory for Mandarin includes the feature [coronal],

which is required to distinguish Mandarin alveolar /s/ from retroflex /ş/. This feature is also

crucial for distinguishing /l/ and /r/ in English (on Brown’s analysis). So when Mandarin speakers

encounter English, they are able to set up a contrast between /l/ and /r/ involving the feature

[coronal], even though /r/ is not a phoneme in Mandarin. By contrast, Japanese and Korean do not

have the feature [coronal]. The /l-r/ contrast therefore remains persistently problematic for them.

As for the /f-v/ contrast, it involves the features [continuant], [labial] and [voice], and although

only /f/ is present (in Japanese and Mandarin), the relevant features are present. Hence /f-v/ poses

no difficulty to these L2 learners. Finally, none of Japanese, Mandarin or Korean has the feature

[distributed] which, according to Brown, distinguishes /u/ from /f/ and /s/ in English, so again /u/ remains persistently problematic for all three groups of speakers.

Since the phonemes /u, v, l, r/ are likely to be frequently represented in the input that L2 learners receive, it is unclear why all the L2 learners studied would be less successful at

discriminating /u/ than /v/, or why Mandarin speakers would be more successful at discriminating /l, r/ than the Japanese and Korean speakers. Some internal knowledge state

must be predisposing them to discriminate some acoustic properties of their L2 linguistic

experience, but not others. An analysis that assumes that there are abstract distinctive

phonological features provided by UG, together with the claim that there is a critical period for

access to these distinctive features, offers an explicit account of the observed L2 behaviour. Such

observations again pose a challenge to accounts of SLA that assume that L2 speakers do not have

specifically linguistic innate knowledge derived from UG.

4.2. Morphosyntax

A phenomenon similar to that described in section 4, but relating to morphosyntax, also

suggests that an emergentist approach may be insufficient. The case is persistent non-target-like

behaviour despite the ready availability of positive evidence. This can be illustrated from a study

by Lardiere (1998a,b, 2007) of a bilingual speaker of Mandarin and Hokkien who was first

immersed in English at the age of 22. Samples of the spontaneous speech of this informant were

collected after 10 years of immersion (sample 1) and 18.5 years of immersion (samples 2 and 3).

Immersion was considerable, including study at an American University and marriage to a

monolingual speaker of English. Lardiere examined the distribution of four properties (illustrated

in (12)) for which there is clear and relatively frequent positive evidence in English input. Three

of these properties differ between English and Chinese:

(12) Phenomena studied Present in Chinese?

Past tense verbal inflection (was, went, walked) No

Pronoun Case (he vs him, she vs her) No

S-V agreement (is, am, walks) No

No main verb raising (*She likes not the book) Yes

R. Hawkins / Lingua 118 (2008) 465–477 473

The extent to which these forms were realised in speech by the informant in obligatory contexts

for native speakers is shown in Table 4.

Although the informant is target-like on the no-verb-raising and pronoun Case properties, she

is not target-like on past tense marking and subject-verb agreement. Since verbs in Chinese are

inflected neither for tense nor agreement, it is not clear why this speaker should have persistent

difficulty in producing these properties in English, given that there is considerable positive

evidence for them in input.

Different explanations have been offered for this. Lardiere, for example, hypothesises that the

speaker has target abstract English syntactic representations. The evidence for this is:

� 100% correct performance on pronoun Case and no verb raising. � Productive and correct marking of past tense and 3rd person –s, when they are used (i.e. there

are no over-generalisation errors). It is just that the forms are not used in all obligatory contexts.

� Past tense/subject-verb agreement is marked on the auxiliary verbs be, have, do, and the copula be in 93% of obligatory contexts.

� In spontaneous writing (e-mails) past tense marking in obligatory contexts rises to 78%.

One of the reasons why this speaker’s performance differs from that of a native, on Lardiere’s

view, concerns the production of word-final consonant clusters, the result of persistent influence

from Chinese, which has a basic (C)V(Nasal) syllable structure, but no syllable- or word-final

clusters. Observe that regular past tense inflection and 3p –s regularly give rise to word-final

clusters: walk (-k), walks (-ks), walked (-kt), phone (-n), phones (-nz), phoned (-nd). Separating

past-tense-marked main verbs into irregulars (went) and regulars (walk-ed) reveals that the

irregulars are inflected in 46% of cases, while the regulars in only 6%.

If this is correct, then it might be claimed that the phenomenon is as compatible with an

emergentist account as it is with a nativist one. The behaviour is simply the result of a

phonological restriction imposed by Chinese that constrains inductive learning from

English input. There are reasons to think, however, that if Chinese phonology is involved,

under-determined principles of UG are also implicated. One prediction of a simple ban on word-

final consonant clusters is that mono-morphemic words like fact /-kt/, lense /-nz/ should be

affected as much as inflected forms like walked /-kt/ and phones /-nz/. This is not always found.

Goad et al. (2003) looked specifically at the suppliance of subject-verb agreement –s by 12 L1

speakers of Mandarin at intermediate and low-advanced levels of English proficiency. None had

problems in producing final consonant clusters in mono-morphemes, but the group divided into

two in terms of their suppliance of the –s suffix. Six supplied –s in only 10% of obligatory

contexts, while six supplied –s in 49% of obligatory contexts.

R. Hawkins / Lingua 118 (2008) 465–477474

Table 4

Tokens supplied in obligatory contexts (based on Lardiere, 1998a,b)

Sample Past Pronoun Case 3p –s No verb raising

1 24/69 49/49 2/42 69/69

(35%) (100%) (5%) (100%)

2 191/548 378/378 0/4 122/123

(35%) (100%) (0%) (99%)

3 46/136 76/76 1/22 42/42

(34%) (100%) (5%) (100%)

Like Lardiere, Goad et al. argue that all 12 have acquired the abstract syntactic representation

for English S-V agreement, but none have acquired English prosodic structure for pronouncing

specifically English verb + inflection, although they can accommodate mono-morphemic

word-final clusters. They are using Chinese prosodic structure differently in conjunction with

their syntactic knowledge. Goad et al. argue that two UG-derived constraints on prosodic words

are relevant to this case:

(i) Syllable rhymes (vowel + coda) are maximally binary. This means that a word like

wide, despite its superficially monosyllabic appearance, consists of two syllables:

waj.dØ, because if it were one syllable the rhyme would involve three segments: *-ajd.

(ii) Adjacent syllables with empty vowels are prohibited within the same prosodic word.

This makes a derived word like *wideth impossible: *waj.dØ.uØ. Syllable shortening is required: wit.uØ.

Since verb inflection as in arrive-s produces adjacent syllables with empty vowels - raj.vØ.zØ -

the inflection cannot be in the same prosodic word as the verb, because this would violate the

constraint against adjacent syllables with empty vowels within the same prosodic word. It must

therefore be in a separate prosodic word:

(13) [PWd [PWd 3raj.vØ] zØ]

The claim now is that Mandarin just does not allow adjunction to a prosodic word. Therefore,

although Mandarin does not have final consonant clusters, speakers could accommodate final

clusters in English mono-morphemes into a single prosodic word structure transferred from

Mandarin:

(14) a. [PWd fak.tØ]

b. [PWd len.zØ]

Goad et al.’s subjects appear to do this. However, where inflections are involved, the resulting

phonological structure might be accommodated within a single prosodic word, or it might

require adjunction to a prosodic word (to avoid adjacent empty syllables) – the option that

Mandarin does not allow. Examples are given in (15). Note that while builds in isolation or

followed by a word beginning with a consonant must involve adjunction to the prosodic word,

when followed by a word beginning with a vowel, the /-z/ can form the onset of the following

syllable.

(15) a. fill-s [PWd fıl.zØ]

b. build-s [PWd [PWd bıl.dØ]zØ] but not *[PWd bıl.dØ.zØ]

c. build-s on [PWd bıl.dØ] [PWd zon]

d. race-s [PWd rej] [PWd s3z]

Given the ‘no adjunction to a prosodic word’ restriction, six speakers in the sample opted mostly

for uninflected verbs in English while the other six accommodated inflection within a single

prosodic word, where possible. The distribution of supplied forms is summarised in Table 5,

where ‘no accommodation’ means the six speakers who mostly opted for uninflected verbs, and

‘accommodation’ means the six speakers who supplied inflections in 49% of cases.

R. Hawkins / Lingua 118 (2008) 465–477 475

Only a combination of the constraint imposed by transfer of the ‘no adjunction to prosodic

words’ property from Chinese and the UG-derived knowledge of the binary nature of rhymes and

the restriction on adjacent empty syllables would appear to explain this kind of performance

where learners have clear positive evidence from input for target inflected forms. Both binary

rhymes and the ban on adjacent empty syllables appear to be properties of the language faculty,

independent of experience, and unlikely to be derived from domain-general principles of learning

and processing.

5. Conclusion

I have suggested that a nativist theory of SLA is supported by four kinds of evidence:

(1) Learners have grammatical knowledge that does not appear to derive from experience,

for example knowledge of the type of antecedent that overt and null pronouns can be

bound by (Kanno, 1997; Pérez-Leroux and Glass, 1997, 1999).

(2) Learners can deduce grammars that are different both from the target L2 and their L1,

but that are not infinitely variable, for example the representation of English articles

by Russian and Korean speakers where interpretation fluctuates between ‘definiteness’

and ‘specificity’ (Ionin et al., 2004).

(3) Different properties of the L2 may be differentially difficult for L2 speakers where

experience appears to offer similar positive evidence, for example the relative difficulty

of English /u/ for Japanese, Mandarin and Korean speakers compared with /v/, and the persistent difficulty of perceiving the /l/-/r/ contrast for Japanese and Korean speakers,

but not for Mandarin speakers. The relevant segments are absent from the L1s, but

are frequently present in the L2 (Brown, 2000). This appears to follow from a critical

period that affects specific phonological features offered initially by UG that then

‘disappear’ if not selected by the L1 grammar.

(4) Learners may fail to converge on the target grammar where there is positive evidence

in input for the relevant target property, for example Chinese speakers who persistently

inflect English verbs for tense and agreement optionally (Lardiere, 1998a,b; Goad et al.,

2003). Such persistent optionality appears to be the result of UG-derived constraints on

L2 representations that are themselves the effect of transfer of an L1 property (in this

case the restriction that inflections cannot adjoin to prosodic words).

The two questions that the article began with were: what is the nature of an L2 speaker’s

mental grammar that would give rise to particular kinds of performance, and how do L2 speakers

R. Hawkins / Lingua 118 (2008) 465–477476

Table 5

Suppliance of –s by the no-accommodation and the accommodation groups

No accommodation (%) Accommodation (%)

fill-s 7 68

build-s on 7 75

race-s 0 27a

build-s 0 9

a Low performance here is unexpected, but may be because Mandarin disallows syllable-final consonants in unstressed

syllables (Goad et al., 2003:17).

arrive at such grammars? While it is undoubtedly the case that much L2 knowledge that speakers

acquire can be arrived at inductively from experience, or deduced by capacities of mind that are

not specifically linguistic, the four cases described above appear to require appeal to innate,

linguistic knowledge. One of the challenges for nativist work on SLA is to establish exactly how

much L2 knowledge is UG-derived, and how much should be attributed to other sources.

Acknowledgements

I thank William O’Grady and two reviewers for comments on an earlier version of this article.

References

Brown, C., 2000. The interrelation between speech perception and phonological acquisition from infant to adult.

In: Archibald, J. (Ed.), Second Language Acquisition and Linguistic Theory. Blackwell, Oxford, pp. 4–63.

Dekydtspotter, L., Sprouse, R.A., 2003. L2 performance: interlanguage representations, computations and intuitions.

In: Liceras, J.M., Zobl, H., Goodluck, H. (Eds.), Proceedings of the Sixth Generative Approaches to Second Language

Acquisition Conference (GASLA 2002): L2 Links. Cascadilla Proceedings Project, Somerville, MA, pp. 45–54.

Ellis, N.C., 2002. Frequency effects in language processing: a review with implications for theories of implicit and explicit

language acquisition. Studies in Second Language Acquisition 24, 143–188.

Ellis, N.C., 2006. Selective attention and transfer phenomena in L2 acquisition: contingency, cue competition, salience,

interference, overshadowing, blocking, and perceptual learning. Applied Linguistics 27, 164–194.

Goad, H., White, L., Steele, J., 2003. Missing inflection in L2 acquisition: defective syntax or L1-constrained prosodic

representations? Canadian Journal of Linguistics 48, 243–263.

Grimshaw, J., Rosen, S.T., 1990. Knowledge and obedience: the developmental status of the binding theory. Linguistic

Inquiry 21, 187–222.

Ionin, T., 2006. This is definitely specific: specificity and definiteness in article systems. Natural Language Semantics 14,

175–234.

Ionin, T., Ko, H., Wexler, K., 2004. Article semantics in L2 acquisition: the role of specificity. Language Acquisition 12,

3–69.

Kanno, K., 1997. The acquisition of null and overt pronominals in Japanese by English speakers. Second Language

Research 13, 265–287.

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

Lardiere, D., 1998b. Dissociating syntax from morphology in a divergent L2 end-state grammar. Second Language

Research 14, 359–375.

Lardiere, D., 2007. Ultimate Attainment in Second Language Acquisition: A Case Study. Lawrence Erlbaum Associates,

Mahwah, NJ.

MacWhinney, B., 2004. A multiple process solution to the logical problem of language acquisition. Journal of Child

Language 31, 883–914.

Montalbetti, M., 1984. After Binding. MIT, unpublished doctoral dissertation.

Pérez-Leroux, A.T., Glass, W., 1997. OPC effects in the L2 acquisition of Spanish. In: Pérez-Leroux, A.T., Glass, W.

(Eds.), Contemporary Perspectives on the Acquisition of Spanish, vol. 1: Developing Grammars. Cascadilla Press,

Boston, MA, pp. 149–165.

Pérez-Leroux, A.T., Glass, W., 1999. Null anaphora in Spanish second language acquisition: probabilistic versus

generative approaches. Second Language Research 15, 220–249.

White, L., 2003. Second Language Acquisition and Universal Grammar. Cambridge University Press, Cambridge.

R. Hawkins / Lingua 118 (2008) 465–477 477