There is an emerging awareness among linguistic theorists and psycholinguists that the study of adult second language (L2) acquisition is rapidly developing into one of the most dynamic and promising areas of inquiry in cognitive science. Traditionally, these language scientists have paid little attention to the issues surrounding L2 acquisition; yet mastering an L2 (or L3, L4,..., Ln) constitutes both a major and a commonplace achievement for most of the world's population. Until recently, however, this process was not well understood. In large part this was because traditional approaches to the study of L2 learning were inextricably linked to language pedagogy alone (see review in Newmeyer 1983; Newmeyer & Weinberger 1987). However, recent theoretical and empirical advances have convinced a growing number of scholars that careful investigation of the L2 acquisition process is likely to be a very fruitful endeavor in understanding the cognitive processes specific to language learning or the biological endowment for language, namely, Universal Grammar (UG).
A central, implicit argument of this paper is that understanding how L2 acquisition occurs will provide a unique and fundamentally important perspective on the mental processes involved in language learning and use. In particular, it provides an important context for investigating the interaction of general cognitive and specifically linguistic processes. This perspective is different from and complementary to that provided by the study of first language (L1) acquisition in children. L2 learners, specifically adult learners, bring capacities to bear on the language learning process that are both similar to and different from the capacities of children. The differences can be attributed to the fact that limitations due to general developmental deficits are typically irrelevant in adult language acquisition. Therefore, the study of L2 acquisition provides a unique context in which to examine language development independent of maturational issues. It also permits anexamination of the interaction of various cognitive capacities that are not fully developed in children. Furthermore, by contrasting adult L2 acquisition and child L1 performance, researchers may be able to identify the role played by experience in general, as well as by the existence or absence of an already functioning L1, in the types of processes that language learners at different ages might rely on.
On the other hand, L2 acquisition research has pointed to striking similarities between child and adult language acquisition, a fact that can be explained if we view these two cognitive processes as fundamentally similar in nature. At this level, then, the study of how individuals acquire an L2 is essential for the development of theories of language. If linguistic theory is to provide an empirically accurate and comprehensive characterization of language capacity and learning, it must accommodate the facts of L2 acquisition. A theoretical or empirical account that is responsive to the facts of children's L1 learning alone is almost certain to be incomplete and may be fundamentally misguided in its conclusions.
In short, L2 acquisition is a critical issue for the study of language and the mind. It represents a central domain of inquiry that is essential for the development of an understanding of language and language learning and their relation to other cognitive processes. In this paper we will attempt to demonstrate how investigations in this field can provide, and in some cases have already provided, important insights into broader issues of language learning.
Thus, a central question is, to what extent, if any, is adult L2 acquisition constrained by the linguistic principles that determine L1 acquisition?1 In this paper we adopt the Principles and Parameters framework of Universal Grammar and examine three logical possibilities that have been articulated regarding the role of UG in second language acquisition. The first is the no-access hypothesis, which claims that no aspect of UG is available to the L2 learner. The second is the partial-access hypothesis, which claims that only L1-instantiated principles and L1-instantiated parameter-values of UG are available to the learner. The third, called the full-access hypothesis, asserts that UG in its entirety constrains L2 acquisition. We argue that there is no compelling evidence to support either of the first two hypotheses, but that evidence does exist that is consistent with the third. In addition, we attempt to clarify some of the theoretical issues that arise with respect to positing UG as an explanatory theory of second language acquisition; investigate certain crucial methodological questions involved in experimentally testing the role of UG in second language acquisition; and present experimental results of our own.
The paper is organized as follows. In Section 1 we briefly outline the theory of UG as the biological construct characterizing the human language faculty and sketch several ways in which the relationship between UG and the process of language acquisition can be conceptualized. In Section 2 we consider the no-access hypothesis as typified by Clahsen and Muysken (1986), Clahsen (1988), and Bley-Vroman (1989). This hypothesis claims "that child first language and adult second language acquisition are guided by distinct cognitive principles" (Clahsen 1988, p. 47). Under this view, child L1 acquisition is constrained by principles of UG, whereas L2 learners use only "general learning strategies"--UG-independent principles--to guide their construction of the L2 grammar. We show that such proposals are both conceptually and empirically problematic, to the extent that they fail to distinguish an L2er's knowledge of language (as represented in a grammar) from the manner in which such knowledge is purported to be acquired (e.g., by "general learning strategies"). Since the no-access hypothesis often appeals to Lenneberg's critical period hypothesis for language acquisition, we discuss it briefly as well.
In section 3 we consider the partial-access hypothesis as developed by Schachter (1989), which states that
UG in its entirety will not be available as a knowledge source for the adult acquisition of a second language. Only a language-specific instantiation of it will be. (Schachter 1989, p.13)Although this hypothesis may at first glance seem less extreme and therefore more attractive than the no-access hypothesis, we will argue that it is nonetheless untenable.2
Section 3.3 contains the central focus of our paper: a case study of a particular version of the partial-access hypothesis, namely, the Weak Continuity Hypothesis. This approach has been developed to account for the acquisition of functional categories for both L1 (e.g., Radford 1990) and L2 acquisition (Vainikka and Young-Scholten 1991; 1992). It holds that although L1 and L2 learners have access to certain principles of UG, their early grammars are incomplete in that they lack functional categories (syntactic categories that are involved in such grammatical phenomena as verb tense, agreement, and question formation, etc.)3
In addition, the Weak Continuity Hypothesis for L2 acquisition claims that L2 acquisition differs from L1 acquisition in the way in which functional categories are eventually incorporated in learners' grammars. Using Vainikka and Young-Scholten's work as representative of the Weak Continuity Hypothesis for L2 acquisition, we examine their data in some detail and argue that serious theoretical and methodological difficulties undermine the purported support for such a model of L2 acquisition.
In section 4 we propose an alternative analysis of certain data presented by Vainikka and Young-Scholten using the model of full access to UG, which posits that both lexical and functional categories are fully available to the L2 learner at all stages of acquisition. In section 5 we present experimental results from our own study that investigated the acquisition of functional categories by child and adult native speakers of Japanese learning English. These results, we argue, support the full-access hypothesis, but neither of its competitors. In Section 6 we present conclusions and discussion.
1. UG Theory and language acquisition
1.1 Brief outline of UG theory
In the theory of Universal Grammar (UG), principles and parameters are hypothesized to constitute the innate cognitive faculty that makes language acquisition humanly possible. An important tenet of this theory is that this faculty is autonomous; in other words, it is an independent cognitive module that may interact with, but does not derive from other cognitive faculties.4
Chomsky (1980, p. 38) has summarized the mechanisms of UG and their role in language acquisition as follows: "UG consists of a highly structured and restrictive system of principles with certain open parameters, to be fixed by experience. As these parameters are fixed, a grammar is determined, what we may call a 'core grammar'." In this theory, the role of principles (i.e., the crosslinguistically invariant (universal) properties of syntax common to all languages) is to facilitate acquisition by constraining learners' grammars, that is, by reducing the learner's hypothesis space from an infinite number of logical possibilities to the set of possible human languages. The role of parameters (which express the highly restricted respects in which languages can differ syntactically) is to account for cross-linguistic syntactic variation. That is, UG principles admit of a limited number of ways in which they can be instantiated, namely those allowed by the parameters specifying "possible variation". The linguistic evidence available to the child during acquisition allows him to determine which parameter setting characterizes his language. Parameter setting eventually leads to the construction of a core grammar, where all relevant UG principles are instantiated. This process is schematized in figure 1.
1.2. Relationship between UG and the course of acquisition
Whether UG is accessible in L2 acquisition depends largely on how one understands the relationship between UG and core grammars.5 In principle, at least two relationships are possible. The first is shown in figure 2. Here, as parameters are fixed during L1 acquisition, UG itself becomes the core grammar. Under this view, parameter setting changes the initial form of UG. Subsequent relations between UG and the grammar of the L2 are necessarily indirect, mediated by the core grammar of the L1 (see figure 3.) An other possibility, illustrated in figure 4, is that UG and core grammars are distinct but related constructs. Thus, one could conceive of UG as the cognitive module that constrains grammar construction during acquisition but itself remains constant during this process. Under this latter view, parameter setting does not entail changing the basic form of UG but instead consists of incorporating into each stage of the grammar the particular UG option that accords with the primary language data (the utterances to which the language-learning child is exposed).
An argument for the latter model comes from bilingual child language acquisition (see related discussion in Flynn and Martohardjono 1994). Here the learner is potentially faced with the task of constructing core grammars requiring different parameter settings for the same principle.
In this situation it is difficult to maintain the hypothesis that parameter setting fundamentally alters the options provided by UG, making nonselected parameter settings unavailable. For example, consider the situation of the child learning both English and Japanese, who must set a parameter for head direction in two different ways. As shown in (1), Japanese "heads" go at the end of phrases and English "heads" appear at the beginning.
(1) Example of parameterized principle (from Cook 1988)6
a. Principle of head direction: A language has its heads on the
same side in all its phrases. This parameter has two
settings: head-first and head-final.
b. English is head-first.
(i) liked the man
(ii) in the bank
c. Japanese is head-last.
nihon-ni (prepositional phrase)
Japan-in
'in Japan'
Watahshi-wa nihonjin desu. (verb phrase)
I-topic Japanese am
'I am Japanese.'
To exemplify, the head of a prepositional phrase is the preposition;
the head of a verb phrase is the verb. Maintaining that UG changes
after parameter setting would force us to conclude that for the
bilingual Japanese-English child, knowledge about the grammatical
principle of head direction derives from UG for only one of the
languages being acquired, presumably the one for which evidence has set
the parameter first. The parametric value determining head direction
in the other language could then not be learned. Rather the knowledge
acquired would have to be explicitly specified (i.e., what is it, if
not a parametric value?), and it would presumably be assumed that
whatever it is it is not determined by the language faculty but by some
other cognitive faculty/faculties. However, this directly conflicts
with the claim that the language faculty is autonomous from these other
faculties (whatever they may be), and there is, in our view--abundant
evidence for such autonomy.Moreover, the alternative is highly
inexplicit--if the bilingual has not acquired 2 different parametric
values for the same parameter--what then has he acquired?--and what
exact cognitive capacity facilitates acquisition of this knowledge? In
order to maintain an explicit empirical hypothesis, we seem to need a
model where the form of UG is not altered by parameter setting but
where uninstantiated settings instead remain available, at least for
child language acquisition. If we accept this model for L1 language
acquisition, it could in principle be true for L2 acquisition as
well--and indeed, we will argue that this is the case.
2. The no-access hypothesis
Three hypotheses have been formulated with respect to the role of UG in adult L2 acquisition. In essence, the first of thesePthe no-access hypothesisPclaims that child L1 acquisition and adult L2 acquisition are fundamentally different cognitive processes, the former deriving from the language faculty, the latter determined by nonlinguistic processes. Proponents of this position (and of the partial-access position discussed in Section 3) often appeal to Lenneberg's Critical Period Hypothesis (CPH) for language acquisition. In this section, we will briefly summarize the proposal for a neurologically based critical period for language acquisition, as well as some of the evidence challenging it, as this is central to understanding the arguments advanced in support of both the No-access and the Partial-access hypotheses.
2.1 Critical Period Hypothesis
To our knowledge, the first proponents of the CPH were Penfield and Roberts (1959), who assumed that language learning increases in difficulty with age and who linked this difficulty to decreasing cerebral plasticity. Lenneberg (1967) fleshed out a more detailed theory of a critical period for language acquisition. Based on his analysis of existing clinical literature on unilateral brain damage and hemispherectomies, Lenneberg concluded that there is progressive lateralization of the language function to the left hemisphere, a process which is complete by puberty. He linked this putative phenomenon to what he believed to be a dramatic decrease in the ability to acquire language. Lenneberg's hypothesis of complete lateralization by puberty was based on two observations: (1) that aphasia as a result of right hemisphere lesions is more likely to occur in children as opposed to adults and (2) after left hemispherectomies children, but not adults, are able to transfer the language function to the right hemisphere. Both observations suggest equipotentiality for language in children but not in adults.
This position has been challenged by much subsequent research. Since Lenneberg's initial formulation of the CPH, increasing evidence has suggested that the notion of progressive lateralization of the language function culminating at puberty is incorrect. Although the evidence supports a biological substrate for language, it points to the presence of such a substrate at birth. Moreover, the evidence suggests neither that there is an optimal age at which the language function operates, nor that this age precedes puberty.
In a well-known reanalysis of the data which formed the basis for Lenneberg's claim, Krashen concluded that lateralization is established around age 5 (1973, p. 65). Similarly, Whitaker et al. (1981) have shown that if lateralization is to be linked to brain maturation, as postulated by Lenneberg, then puberty is an unlikely time for lateralization to be completed, since the language area of the brain takes on most of its adult characteristics by age 5. As noted by Snow (1987),
relying on brain growth as the biological process that times the critical period would predict a critical period ending at 5. This is not an age at which any sharp discontinuities in language acquisition can be observed however. (p. 188)
The observation that the language function can be transferred to the right hemisphere after left hemispherectomy was also seriously questioned by Dennis, who found permanent linguistic deficits even in children who were thought to have recovered their language fully after left hemispherectomy (cf., e.g., Dennis 1979, 1980; Dennis & Kohn 1975).
As a result of the evidence that seriously challenges the CPH, and in particular the claim that there is increasing hemispheric specialization of the language function that culminates at puberty, some researchers maintain that there is no evidence against L1 and L2 being subserved by the same neural stratum (e.g. Kinsbourne 1981). Today, researchers investigating the link between behavior and brain maturation tend to speak of "sensitive" rather than "critical" periods (see Oyama 1987; Bornstein 1987). This shift in terminology reflects the growing sentiment in the field that whatever biological determinants there are of behaviors/capacities, they are unlikely to become totally unavailable after a certain age. Rather, evidence suggests that during certain periods of development, certain organisms tend to have heightened sensitivity to certain environmental stimuli, thereby enjoying an optimal situation for a given behavior/capacity to develop. It is possible that something not dissimilar holds true for language acquisition as well. Furthermore, although there are clearly differences between child and adult language learning, these differences need not reside in accessibility or inaccessibility of universal linguistic principles themselves, as proposed in the no-access hypothesis. Instead, they could well derive from differential ways of instantiating such principles according to the particular demands of a language. For example, Johnson and Newport (1991) report an age effect on the acquisition of certain aspects of grammar. In their study, speakers who acquired L2 English after the age of 7;00 show a steady decline in the acquisition of rules such as third-person singular s-affixation or progressive ing-affixation. These, however, are language-particular morphological aspects of English, related to, but still different from, the principles and parameters of UG themselves. It might thus be the case that the acquisition of language-particular aspects of grammar is age-sensitive.
It is important to note that the full-access hypothesis does not deny the existence of differences between L1 and L2 acquisition, nor is it incompatible with the existence of linguistic development through time. Within this framework, however, the source of these differences is not a lack of access to UG in L2 acquisition. Instead, it is hypothesized that they follow, for example, from the differences involved in assignment of parameter values in L1 acquisition versus assignment of additional parametric values in L2 acquisition or e.g. from differences in the way children and adults acquire the lexicon and integrate UG with grammar-external performance systems. If correct, the consequences of any one of these processes could lead to what appear to be "major" differences between L1 and L2 acquisition, but which are in fact not central to arguments concerning the role of the domain-specific faculty for language, namely, UG (see discussion in Flynn & Manuel 1991; Flynn and Martohardjono 1991).
2.2 L2 acquisition as a nonlinguistic phenomenon
Clahsen and Muysken (1986), Clahsen (1988), and Bley-Vroman (1989) offer the most radical formulations of the position that L2 acquisition is fundamentally different from L1 acquisition. The basic claim is that L2 acquisition is governed by cognitive faculties that are separate and distinct from the domain-specific language faculty, UG. Clahsen, for example, suggests that processes similar to Slobin's (1973) Operating Principles can account for L2 learning. Bley-Vroman suggests that L2 learning strategies derive from Piaget's Formal Operating Principles; these include the capacity for "distributional analysis, analogy, hypothesis formation and testing" (1989, p. 54).
We will argue that although these no-access models claim that they explain L2 acquisition and present themselves as empirically viable alternatives to a UG-based theory of L2 acquisition, they fail to address the central issues confronting any L2 acquisition theory: (1) what exactly does (can) the L2 learner know and (2) how does the L2 learner come to know what he knows? Neither model provides adequate descriptions (i.e., explicit specification of the formal properties) of the grammars L2 learners internalize, that is, the knowledge-states L2 learners are in. Moreover, although both models implicitly assume that the L2 learner eventually attains a natural-language grammar, neither model shows how such grammars can be acquired by employing the nonlinguistic principles they claim to be central to L2 acquisition. Both proposals suggest that general learning processes are central; however, insofar as such analytical skills are merely learning strategies (i.e., "tools")with which L2 learners construct grammars, they tell us nothing about the actual content of the end-state grammar (knowledge) represented in the learner's mind/brain. And this is of course precisely where the explanatory power of a UG-based theory lies.
The hypothesis that L2 learning is not constrained by the language faculty fails empirically given what is known about the L2 learner's linguistic knowledge as represented in a grammar. For example, the capacity to distinguish between speech and other noises is provided by the language faculty, in particular, universal phonetics (see e.g., review in Eimas (1975)). Positing total inaccessibility to the language faculty would fail to explain how adults even distinguish primary L2 data from nonlinguistic acoustic disturbances ("noise"). Similarly, total inaccessibility would seem to entail that learners do not approach the L2 learning task knowing (for example) that, like all human languages, the target language has lexical items ("words"), a construct specified in UG (Chomsky 1980).7 In order to account for the facts regarding L2 acquisition without appeal to UG, one could of course (explicitly or implicitly) attribute to L2 learners knowledge that is empirically equivalent to that provided by UG in this domain. However, if one accepts the empirically equivalent perspective, then one has in effect abandoned the no-access hypothesis. We will discuss this and other related issues in much more detail in section 2.4 when we consider Bley-Vroman's (1989) specific version of the no-access hypothesis, the Fundamental Difference Hypothesis.
2.3. L2 acquisition as an astructural problem solving process
In order to account for the core linguistic knowledge that L2 learners have, Bley-Vroman (1989) appeals to the ability to analogize from the L1 to the L2. Appealing to analogy as a central process in L2 acquisition is problematic because it fails to distinguish between what the L2 learner knows (content of grammar) and how he attains this grammar (process). Furthermore, it entails a fundamental difference between the nature of L1 and L2 acquisition and thereby fails to account for the significant similarities between them, neither of which can be explained by analogy.
First, appealing to analogy (however it is to be defined) as a central process by which grammar is 'built' leaves the result of the process--the content of the grammar (= that which is known)--wholly unspecified and therefore certainly unexplained. That is, what the L2 learner knows is not even addressed. Clearly, whether or not "analogy" plays a role in acquisition, it cannot be part of the content of the grammar; that is, it is not part of what the L2 learner knows. For example, to say "Person P's grammar of German has/contains analogy" is anomalous (cf. "Person P's grammar of German has this subject-verb agreement rule, this word order specification, this lexicon, etc."). In sum, appealing to analogy as a central process by which grammar is built, fails to specify, hence fails to explain knowledge of language i.e., the content of the grammar. These hypotheses thus exhibit the questionable property that they make claims about how an object, X, is constructed without even minimally specifying the properties X has.
Second, if analogy were central in L2 acquisition, knowledge of language (hence, (un)grammaticality) would remain unexplained. We know that analogy cannot explain certain types of knowledge attained in L1 acquisition. For example, native speakers of English do not accept as grammatical a string such as *Who did you see John and? "on analogy with" the sentence Who did you see John with? Clearly, as syntactic research has shown, such contrasts are not isolated but in fact, infinitely varied; knowing a language entails knowing precisely such facts regarding grammaticality. The entirely empirical question here is: What does one know when one knows English? It is incumbent upon the proponents of analogy theories to define this notion explicitly enough to subject such theories to empirical tests.
L2 research has shown that knowledge of L2 (un)grammaticality is similar to that found in L1 acquisition (White 1989; Munnich, Flynn & Martohardjono 1991, 1994). Furthermore, this is the case even when the L1 cannot be the knowledge source (Martohardjono 1991, 1992; Uziel 1991). These studies show that L2 learners do not assume surface string grammaticality properties to be the same or even similar in the L1 and the L2, indicating that L2 learners do not "analogize grammaticality" from the L1 to the L2.
Martohardjono (1991) investigated knowledge of constraints on movement of wh-words (what, who, why, where, etc.) in L2 English by two groups of learners whose L1s do not instantiate overt wh-movement, namely, speakers of Chinese and Indonesian. The constructions tested were wh-questions (questions involving movement of a wh-word) that violate a particular constraint on movement in English. Specifically, in English an argument (such as the object of a preposition (2) or of a verb (3)) embedded in a relative clause (2) or an adjunct clause (3) cannot be extracted (moved from the position marked by ____ to the front of the sentence) to form a question
(2) *What did Sue read a book that talked about ___?
(3) *Who did this letter arrive after Sue called ____?
Importantly, in the corresponding, perfectly grammatical Chinese and
Indonesian wh-questions the wh-word is not moved. For example, the
Indonesian (4) expresses the same question as does (2).
(4) Siti baca buku yang menceritakan apa?
Siti read book RELATIVE ACT -tell-BEN what
(Literally) 'Siti read the book that talked about what?'
The same situation obtains in Chinese, where an unmoved wh-word (i.e.,
occupying the "_____" position) in a relative clause (5) , an indirect
question (6), a sentential subject (7) or a reason adverbial clause (8)
is licit (cf. Huang 1982) unlike the ungrammatical wh-movement analogs
in English:
(5) Relative Clause:
Ni zui xihuan shei xie de shu
You most like who write REL book
*'Who do you like the books that ____ wrote?'
(6) Indirect Question:
Ni xiang-zhidao shei lai-bu-lai?
You wonder who come-not-come
*'Who do you wonder whether ____ will come?'
(7) Sentential Subject:
Lisi mai shenme zui hao
Lisi buy what most good?
*'What is that Lisi buys ____ best?'
(8) Reason Adverbial Clause:
ni [ yinwei wo shuo-le shenme] er bu gaoxing?
you because I said what then not happy
*'What are you unhappy because I said ____?'
Where there is a mismatch in surface-string grammaticality between the
L1 and the L2, a possible hypothesis involving analogy is that learners
"analogize grammaticality" from the L1 "onto" the L2. That is, if L2
acquisition diverged from L1 acquisition in that analogy were the
determining factor in establishing knowledge of grammaticality (as
Bley-Vroman's proposal seems to imply), one would expect L2 learners to
judge wh-question types that are grammatical in their L1 as also being
grammatical in the L2. However, Martohardjono's results indicate that
this is not the case. Both the Chinese and the Indonesian subjects
rejected English wh-questions exhibiting extraction out of relative
clauses, adjunct clauses and sentential subjects as ungrammatical. This
suggests, contra Bley-Vroman, that "analogy" (however this inexplicit
term is to be defined) is not central in L2 acquisition, and can
therefore not be what critically distinguishes L2 acquisition from l1
acquisition.8
On the other hand, it also appears to be the case known that many patterns of acquisition match for both L1 and L2 acquisition suggesting that similar processes underlie these two types of acquisition.9 These patterns do not derive from knowledge of the L1, nor do they result by analogy to some other forms in the L2 input. This conclusion is well documented in much early literature, most notably in works that focused on the acquisition of certain functor morphemes in English such as articles, past-tense endings, and possessives. Results of extensive study indicated parallels in orders of relative difficulty for child L2 learners (Dulay & Burt 1974a; 1974b) and adult L2 learners (Bailey, Madden, and Krashen 1974) for the acquisition of these functors; moreover, these patterns paralleled those isolated for L1 acquisition of English.
There is a highly consistent order of relative difficulty in the use of functors across different language backgrounds, indicating that learners are experiencing intra-language difficulties..... Children and adults use common strategies and process linguistic data in fundamentally similar ways. ( Bailey et al.. 1974, p. 235).
Other work (e.g., Cook 1973) isolated common patterns of errors among both child L1 learners and adult L2 learners in their elicited imitation and comprehension of relative clauses in English.
To sum up, both groups [L1 and L2 learners] seemed to have tackled the tasks of imitation and comprehension of relative clauses in much the same manner. The similarities are: (i) the low proportion of word perfect imitations; (ii) the omission of 'that' even when required grammatically; (iii) the replacement of 'that' by grammatical alternatives; (iv) the addition of relative pronouns to sentences where grammatically possible; (v) the tendency to find clauses qualifying the 'object' more difficult to imitate than those qualifying the 'subject;' (vi) the fact that neither group showed a markedly different pattern of imitation with sentences differing in case grammar; (vii) the 'encoding' of syntactic structure was found in both groups; (viiii) the low proportion of correct answers to comprehension questions in both groups. (Cook 1973, p. 27).Similar results can be found in the early work of, for example, d'Anglejan and Tucker (1975) and Cooper, Olshtain, Tucker, and Waterbury (1979). We will return to a discussion of these data as well as more recent work in section 3.2.4.
The same arguments hold with respect to structure dependency in general. Several studies have shown that L2 knowledge, like child L1 knowledge is indeed structure-dependent; that is, L2ers have knowledge of, i.e. have acquired the L2 rules generating, the hierarchical syntactic structures associated with linear word-strings. As noted by Zobl (1983), Jenkins (1988), Flynn (1983), Flynn and O'Neil (1988), Felix (1988), Thomas (1991), Flynn and Martohardjono (1991), Uziel (1991), and Martohardjono (1993), an infinite number of structure-independent error types are never made by L2 learners.10 For example, Jenkins argues that L2 learners do not utter strings such as *Is the dog which in the corner is hungry. and they presumably judge them ungrammatical. If language learners were simply choosing a structure-independent rule that scans the string of words looking for the first occurrence of is in the linear word-string, and then preposes it to form a question, on an analogy with the formation of simpler questions in English (e.g., John is here => Is John here?), we might expect from L2 learners utterances of the form *Is the dog which ___ in the corner is hungry?. However, the absence of such errors strongly suggests that L2 learners have structure-dependent knowledge; specifically, knowledge of the "Head Movement Constraint": a restriction on the movement of certain structurally defined categories, e.g., verbs such as "is", disallowing *Is [the dog which ___ in the corner] is hungry? but allowing Is [the dog which is in the corner] ___ hungry? (see also Chomsky 1975, chapter 1 and Crain and Nakayama 1987) Similarly, the Head Movement Constraint blocks *Have John will __ breakfast? but allows Will John __ have breakfast? (see Travis (1984), Crain 1992; Freidin 1992 for more extensive discussion of this constraint).
L2 learners also exhibit the creative aspect of language use. Like children, adults are not limited to mimicry of what they have heard (Felix 1988; Flynn & O'Neil 1988; White 1988). L2 learners from all L1s thus far investigated achieve mental states for the L2 that go well beyond available data and beyond any explicit teaching; they can understand and produce utterances they have not seen or heard before.11
In sum, the data elicited from L2 learners cannot feasibly be accounted for in terms of inductive learning procedures for language that are implicit in models that reject UG.
2.4 A Case study: Bley-Vroman's version of the no-access hypothesis, The Fundamental Difference Hypothesis
An often-cited version of the no-access hypothesis is Bley-Vroman's (1989) Fundamental Difference Hypothesis (hereafter FDH), which states that the difference between L1 and L2 acquisition is "internal, linguistic and qualitative" (p. 50). It is internal, because it is "caused by differences in the internal cognitive state of adults vs. children, not by some external factor or factors... such as insufficient input, for example" (p. 50). Furthermore, "it is linguistic, in that it is caused by a change in the language faculty specifically, not by some general change in learning ability" (p. 50). Finally, it is qualitative, not just quantitative, in that in L2 acquisition "the domain-specific acquisition system is not just attenuated, it is unavailable." These differences in the knowledge sources for the child L1 learner and the L2 learner are schematized in the following table from Bley-Vroman (hereafter B-V) (1989, p. 51):12
Child language development Adult foreign language learning A. UG A. Native-language knowledge B. Domain-specific learning B. General problem-solving systems procedure
When we consider the details of the FDH, certain inconsistencies will appear, and it will turn out that its most central claim, that the mental representations of the L1 and L2 grammars are fundamentally different, is substantially weakened. We will show that the FDH does not in fact deny domain-specific linguistic knowledge in the L2 learner. At the same time, it paradoxically argues that the domain specific language faculty is inaccessible to L2 learners. In order to resolve the paradox and account for the domain-specific linguistic knowledge that L2ers acquire, the FDH ends up according native-language knowledge a privileged role in L2 acquisition. B-V writes,
(9)
[T]he learner will have reason to expect that the language to be learned will be capable of generating an infinite number of sentences; a language of finite cardinality will not be expected. The learner will expect that the foreign language will have a syntax, a semantics, a lexicon which recognizes parts of speech, a morphology which provides systematic ways of modifying the shapes of words, a phonology which provides a finite set of phonemes, and syllables, feet, phonological phrases, etc. Universals of this sort are available to the foreign language learner merely by observing (not necessarily consciously) the most obvious large-scale characteristics of the native language -- no deep analyses are necessary -- and by making the very conservative assumption that the foreign language is not an utterly different sort of thing from the native language. (p. 52)
These qualifications rather drastically modify the original and most fundamental claim that knowledge given uniquely by the language faculty is unavailable to the L2 learner. First, it is not clear that knowledge of phonemes, syntax, morphology, and so on--the most fundamental concepts provided by UG, which moreover are domain-specific in the sense that they are pertinent only to language--can be attributed to the L2 by the learner without "deep analysis" of the L2. Second, for the learner to assume that the architecture of the L2 is "not utterly different" from the architecture of the L1 is tantamount to saying that the learner is able to identify and presumably treat the L2 as an object that falls under the domain of the language faculty.13 This in turn amounts to an admission that the L2 learner is in large part approaching the task of L2 acquisition with the same domain-specific knowledge that underlies L1 acquisition. The significance of this is intended to be weakened by the statement that this knowledge comes to the learner "via the L1." It is not at all clear, however, whether and how the claim that domain-specific linguistic knowledge applied to the L2 "comes from the L1" can be empirically distinguished from the claim that it "comes from" the language faculty itself, since it is this very faculty that specifies universally determined knowledge of the L1 in the first place. This is particularly relevant when the knowledge is identical in the L1 and L2, as we will indicate below.
Given that the ideas expressed in (9) are widely entertained and fundamentally alter the original claim of nonaccessibility expressed by the FDH, let us consider (9) in more detail, beginning with the first sentence: "[T]he learner will have reason to expect that the language to be learned will be capable of generating an infinite number of sentences; a language of finite cardinality will not be expected. [our emphasis]" Here (and elsewhere) there seems to be a confusion between "language" and "grammar". A standard--and, we believe, necessary--assumption is that the learner is in fact acquiring a grammar (in standard theory, a grammar is a rule system that generates sentences; in current theory, a set of universal principles with parameter values determined; in theory-neutral terms, a particular set of linguistic laws (e.g., "The subject precedes the verb"; "The verb agrees with the subject")). Crucially, the learner is not "learning a language"; that is, he is not learning some infinite set of sentences (e.g., SEnglish = {Bill left, Sue leaves, a man is leaving, a tall woman has left,...}). The grammar is the knowledge-state rule system, the generator capable of generating an infinite number of sentences; the language is the infinite set that is generated. Thus, Bley-Vroman uses the term language to mean these two crucially different things, and the first assertion in the quotation must therefore be restated as "[T]he learner will have reason to expect that the grammar to be learned will be capable of generating an infinite number of sentences."14
Our first question then is this: Why/how is it that upon exposure to (among other data) some finite (therefore, proper) subset of the foreign language (i.e., foreign primary linguistic data), human (and presumably only human) foreign-language learners expect (know) that--like all natural-language grammars-the L2 grammar has infinite generative capacity and that the L2 grammar has other universal and very fundamental properties including a multicomponential internal structure "exhibiting a syntax, a semantics, a lexicon,...a morphology,...a phonology,...etc."? In other words, Why/how is it that under the FDH, the L2 learner "expects" (knows) that the L2 consists of precisely those subcomponents that all natural-language grammars display, as specified by UG?15 As seems reasonable, we argue that this "expectation" comes to the learner via the language faculty itself.
Note, in this regard, that the FDH implies something much stronger than the (very weak) assertion that the L2er expects the L2 to have for example, "a syntax" (see again (9)). There are an infinite number of possible "syntactic systems" (means of concatenating symbols), including systems such as "Freely concatenate any symbols in any order." Presumably, what is intended is not that the L2er expects the L2 to have "a" (i.e., any ) syntax. In fact, B-V explicitly claims that the L2er expects a syntax with infinite generative capacity, hence an infinite number of possible syntactic systems; namely, all those having finite generative capacity, are not "expected" by the L2er. It appears then that the intended meaning is
"The learner will expect that the foreign language [i.e., the L2 grammar] will have a NATURAL LANGUAGE syntax (as defined by UG), a NATURAL LANGUAGE semantics (as defined by UG), a NATURAL LANGUAGE morphology (as defined by UG), etc."
The same arguments hold with respect to the claims in (9) made about the learner's L2 phonology and morphology. Thus, B-V proposes that the L2 learner expects very specific component-internal atomic-category and phrasal-category inventories, for example, "a finite set of phonemes, and syllables, feet, phonological phrases, etc." Here again, the theoretical constructs B-V specifically mentions are precisely UG-specified sets of phonological constructs common to all natural language grammars.16 How/Why does the L2er know ("expect") such UG-specified constructs?
B-V also proposes that the L2 learner expects component-internal structure-building (symbol-concatenating) operations (or rules), for example, "a morphology which provides systematic ways of modifying the shapes of words." This statement in fact imputes far more specific knowledge to the L2 learner than the phrase "a systematic morphology" conveys. Clearly the claim is not that the learner merely "expects" any morphology that systematically modifies the shapes of words, such as a morphology incorporating rules like the following (found in no human grammar to date):
(i) Form plurals by adding the suffix -s to all and only those nouns that refer to an object weighing less than 40 tons.We believe L2ers do not ever hypothesize (perfectly systematic) morphological rules such as (i) (even though they are entirely consistent with, i.e. correctly generate, primary linguistic data such as cat/cats, dog/dogs, stick/sticks, etc.). Again, we presume that the intended meaning is that the learner expects a natural-language morphology as defined by UG--in other words, a morphology that excludes an infinite number of systematic rules like (i). Again, why/how is it that the learner "expects" (knows) that such UG-specified constraints characterize the L2?
Crucially, this has been the recurrent question throughout our discussion of (9). The FDH claims that second grammar acquisition is not constrained by UG; that is, although no empirical evidence is given against access to UG, the FDH claims that UG need not be assumed to account for the L2 learner's knowledge, because the same knowledge is available elsewhere, from a purported UG-external source--namely, from observation of the L1. Nonetheless, we have argued, under the FDH the L2 acquirer has knowledge that is in fact specified in UG; that is each property of the L2 that the L2 learner is claimed to 'expect' is in fact a property of grammar specified by UG, a property common to all natural-language grammars. Thus, we claim that the FDH attributes UG-specified knowledge to the L2 learner after all, albeit in the form of the L2 learners "expectations" about the nature of the second "language."
Notice, at this juncture, that any empirical distinctions between an account that attributes to L2ers "UG-specified knowledge" and one that attributes "UG-specified knowledge via the L1" are extremely subtle, since in the cases discussed the knowledge attributed to the L2er is identical in the two accounts (e.g., knowledge of "phonological phrase"). The difference is the source of the knowledge: UG versus UG-via-L1.17 Since much of the argument regarding UG accessibility rests on this distinction, let us consider it more closely. Recall that the FDH makes the following claim:
Universals of this sort are available to the foreign language learner merely by observing (not necessarily consciously) the most obvious large-scale characteristics of the native language--no deep analyses are necessary---and by making the very conservative assumption that the foreign language is not an utterly different sort of thing from the native language. (Bley-Vroman 1989, p. 51-52)
As discussed earlier, "universals of this sort" are universal properties of grammars (not languages) as specified by UG: infinite generative capacity, a finite set of autonomous computational subsystems (e.g., "a" syntactic component), component-internal categorial inventories (e.g., phonemes, each a bundle of UG-specified feature specifications), and so on. The question is, How are these acquired by "mere observation"? Specifically, What is an observable, obvious, large-scale characteristic of the native language requiring no "deep" analysis? One might imagine the following type of response: "An observable, obvious, large-scale characteristic requiring no deep analysis is, for example, that the native language has words." But of course "observing the large-scale characteristic 'word'" is equivalent to "analyzing the primary linguistic data in terms of the abstract construct 'word,'" a construct provided by UG and common to all natural-language grammars. Thus, humans can "observe" 'words' and find this "large-scale characteristic" of the continuous acoustic speech stream "obvious" only because the construct "word" is biologically provided (by UG). That is, organisms not genetically equipped to analyze human speech in terms of the UG-specified abstract construct "word" will be unable to "observe this obvious, large-scale characteristic" of the acoustic disturbances that humans, by virtue of our biological constitution, recognize as speech. Surely, then, such universal properties of grammar, are not "available by mere observation" of the L1. If in B-V's view no "deep" analysis is necessary, are 'shallow' analyses sufficient and if so, what exactly is the shallow analysis and why is it that only humans can posit it?
Under the FDH, once the L2er completes his "mere observation of the most obvious large-scale characteristics of the native language," he then makes the assumption that the foreign language is not an utterly different sort of thing from the native language. Leaving aside the unclarity of the phrase "not utterly different," the crucial question is, Where does this assumption come from? Clearly, the L2er is not told or taught to make the assumption. Rather, it seems to be a property of the L2er himself; that is, it seems to be a property of humans that upon exposure to primary linguistic data from a new language, they just "[assume] that the foreign language is not an utterly different sort of thing from the native language."
In fact, it does not seem to be the case that the L2er "assumes" the L2 and the L1 are "not utterly different." Rather, the L2er appears to analyze an L2 as fundamentally the same as an L1, and we argue that this is the case precisely because UG is still available to him to make this analysis possible. For example, even under the FDH, exposure to finite, fragmentary, and "degenerate" primary linguistic data from the L2 results in the internalization of an L2 grammar that, just like the L1 grammar, is constrained by what are in fact posited principles of UG.18
But how is this possible without UG? In other words, if not by virtue of UG, how is it that the L2 is "assumed" to be not utterly different from the L1? Here again, the FDH analysis appears to appeal directly to UG-constrained L2 acquisition.
In summary, we hope to have established the following points concerning this representative version of the no-access hypothesis. First, it incorporates a fundamental and hence serious confusion between "language" and "grammar." Second, it attributes not only fundamental but also highly specific knowledge of linguistic universals to the L2er. Therefore, the empirical distinction between the FDH and a UG-constrained theory of L2 formation is highly subtle, since they both attribute what seems to be the same (UG-specified) knowledge to the L2er (again depending on B-V's implicit assumptions in (9)--e.g. what is "a syntax", "a semantics"?--What is meant by "etc."?--What is the finite set of phonemes, syllables, feet, phonological phrases?--What explicit definitions of these terms are being assumed, but not provided?). Where they differ is in the source of the UG-specified knowledge. We believe the L2er has UG-specified knowledge by virtue of 'having' UG. By contrast, B-V asserts that the L2er obtains this very same UG-specified knowledge from purported UG-external sources namely (a) mere observation of the native language, and (b) the assumption that the foreign language is not 'utterly different' from the native language. We have argued, however, that these assumptions are highly inexplicit and that once they are rendered explicit, each in fact appeals directly to UG. Thus, the no-access hypothesis would seem to entail access after all.
2.5 Clahsen and Muysken's version of the no-access hypothesis
Another formulation of the no-access hypothesis, which confronts potential difficulties different from those facing the FDH, is found in the work of Clahsen and Muysken (1986) and Clahsen (1988), who compare the development of word order in the L1 and L2 acquisition of German. They claim that although child grammars are constrained by UG, adult grammars are not. Clahsen states that
the observed differences between L1 and L2 learning can be explained by assuming that child L1 acquisition falls under the parameter theory of language development, whereas the acquisition strategies used by adults in L2 development may be defined in terms of principles of information processing and general problem solving. (1988, p. 22)
The claim that adult L2 grammars are not constrained by UG is based on evidence concerning differences in the developmental sequences observed in L1 and L2 acquisition of German word order. The authors claim that the observed differences follow from the assumption that only children have access to UG. Of course, these (like all other such) conclusions crucially depend on the particular analysis, in this case, the analysis of German, that is assumed. Alternative explanations of these acquisition data have been proposed by several researchers, who have argued that a different (and arguably empirically preferable) linguistic analysis of German syntax reveals evidence for the role of UG principles in the L2 acquisition of German word order (e.g., duPlessis, Solin, Travis, & White 1989, Tomaselli & Schwartz 1990).
To illustrate briefly, any analysis of German must account for the fact that in main clauses the tensed verb is found in second position after the subject or a topicalized element, as in (10).
(10) a. Die Kinder haben das Brot gegessen.
the children have the bread eaten
'The children have eaten the bread.'
b. Das Brot haben die Kinder gegessen
the bread have the children eaten.
c. Gestern haben die Kinder das Brot gegessen.
yesterday have the children the bread eaten
(duPlessis et al, p. 58)
In embedded clauses, in contrast, the finite verb is found in final
position, as in (11).
(11) Ich glaube dass die Kinder das Brot gegessen haben.
I believe that the children the bread eaten have.
(duPlessis et al, p. 58)
In brief, Clahsen and Muysken account for these facts by adopting an
analysis of German that assumes that the verb-second effects can be
accounted for if the underlying order of German is TOPIC
COMP(lementizer, e.g., da') S(UBJECT) O(BJECT) V(ERB) and if two
obligatory movement rules apply in main clauses. The first rule moves
the inflected verb (in (10), haben) to the COMP position and the second
moves some phrasal category into the TOPIC position immediately
preceding the inflected verb. That the verb does not move to second
position in embedded clauses "is accounted for by the fact that the
complementizer has filled the position to which the inflected verb
would normally move, thereby preventing this movement" (duPlessis et
al. 1989, p. 58).19
Assuming this analysis of German, Clahsen and Muysken review various empirical studies of the development of verb position in German L1 and L2 acquisition. As a result of this review, they postulate two different stage grammars: one for L1 learners of German and one for L2 learners of German. They argue that the observed differences between L1 and L2 acquisition of German word order are due to the L2er's internalization of "unnatural" rules proscribed by UG. More specifically, they argue that, at a certain stage of development, German adult L2 learners, but not child L1 learners, move elements "such as particles, participles and infinitives" to sentence-final position. Because such movement is prohibited under the particular analysis of German that Clahsen and Muysken assume, they conclude that adult L2 acquisition of German is not constrained by UG.
This conclusion regarding the role of UG in adult L2 acquisition can be challenged in several ways (see more detailed discussion in Flynn and O'Neil, 1988a).20 duPlessis et al., for example, demonstrate that the adult L2 German results can be accounted for by a UG-consistent analysis by assuming a different independently motivated analysis of German, that of Travis (1984). This account of German also posits a verb movement rule; however, the verb may appear in three positions, namely, in V(erb), in Infl(ection), and in COMP(lementizer).21 Verb movement in embedded clauses is not absolutely prohibited as in Clahsen and Muysken's account since there is an available landing site for it, namely Infl.
2.6 Conclusion
In sum, we know of no version of the no-access hypothesis that displays even descriptive adequacy, since invariably these analyses fail to even specify the L2er's knowledge of language as represented in a grammar. Furthermore, the various versions of this hypothesis remain speculative and arguably contradictory, since in order to account for the knowledge L2 learners patently have, such hypotheses are forced to retreat (often implicitly) from the fundamental claim that only "nonlinguistic principles" (whatever they may be) govern L2 acquisition. Thus, in the end, such hypotheses characteristically attribute domain-specific linguistic knowledge to the learner--the very knowledge they presume to be unavailable in L2 acquisition.
3 The partial-access hypothesis
3.1 Overview of the partial-access hypothesis
In contrast to no-access hypotheses, partial-access hypotheses claims that UG knowledge is not totally unavailable, but that it is limited in very specific ways that vary according to the particular version of the hypothesis. One representative proposal argues that only an L1-instantiated UG remains available to the adult (cf. Figures 2 and 3). This would mean that only the invariant principles of UG i.e., those that characterize grammars of all languages would remain accessible to the L2 learner. Under this proposal, the child L1 learner has available the entire range of options provided by UG, namely, all invariant principles and all parameterized principles with settings still available. However, the L2 learner has direct access only to the invariant principles of UG (those that characterize the grammars of all languages). With respect to the parameterized principles for which values were set during the learner's L1 acquisition, only the parametric values instantiated in L1 are available to the learner during L2 acquisition. Where the L1 and the L2 differ in a particular parametric value, the L2 learner--having immutably set the L1 value during L1 acquisition--is hypothesized to be incapable of assigning a new value to this parameter to coincide with the target grammar; or, if a particular value of a parameter is necessary for the acquisition of the L2 but this value is not realized in the L2 learner's L1, then the L2 learner will not be able to acquire this value. One particular version of this proposal, the Window-Of-Opportunity Hypothesis (Schachter 1989) is outlined in (12).
(12) Window-Of-Opportunity HypothesisAccording to the proposal in (12), UG as it is available to a child L1 learner does not constrain the L2 learner's hypotheses. Grammar construction by the L2 learner is constrained not by principles and parameters of UG but by principles of UG and the immutably set parameters of the particular L1 grammar. Thus, this theory predicts that L2 grammar construction will differ significantly from L1 grammar construction. Further, completeness in L2 acquisition is predicted to be impossible in a wide variety of cases, namely, whenever one or more parametric values do not match or whenever certain L1 principles are not instantiated in the L2.22[A]ll that remains as part of the knowledge state of an adult native speaker of a language is a language-specific instantiation of UG, that of the first language. UG in its entirety will not be available as a knowledge source for the acquisition of a second language. Only a language-specific instantiation of it will be.... If, however, it turns out that in the acquisition of the target some instantiation of principle P is necessary and P is not incorporated into the learner's L1, the learner will have no language-internal knowledge to guide him/her in the development of P. Therefore, completeness with regard to the acquisition of the target language will not be possible. (Schachter 1989, p. 13-14)
A similar proposal is also made by Strozer (1992). Strozer argues that only the invariant principles of UG are available to the L2 learner and that parameter setting is impossible in adult L2 acquisition. Although this approach might seem to provide an appealing account of L2 acquisition, we argue that this position can be seriously challenged in at least two ways: first, by demonstrating that L2 learners are able to construct grammars incorporating parameter settings not instantiated in the L1, and, second, by demonstrating that other nonparametric grammatical options not instantiated in the L1 are available to the L2 learner.
3.2 Evidence against the partial-access hypothesis
3.2.1 New parameter settings23
The partial-access hypothesis entails that UG fails to constrain those aspects of L2 acquisition where there is a mismatch between the L1 and L2 grammars. This claim amounts to the hypothesis that an L2 learner has a type of hybrid grammar: one part UG-constrained and one part not.24 If, on the other hand, the L2 learner has continuous (full) access to UG, we expect that regardless of the match or mismatch between L1 and L2 language specific principles, the L2 grammar will be constrained by principles and parameters of UG. More specifically, we expect that the L2 grammar will, for example, obey certain universal constraints on the movement of certain elements, such as wh-words in questions. We also expect that an L2 learner will be able to assign a new (i.e., non-L1) value to a given parameter.
Results from several empirical studies of Japanese speakers' acquisition of English support the hypothesis that L2 learners are able to assign new parametric values in the construction of the L2 grammar when there is a mismatch between the L1 and the L2 (e.g., Flynn 1983; 1987; 1991; 1993; Flynn and Martohardjono 1992, 1994). Importantly, these learners do so in a manner consistent with predictions made by the theory of UG.
Let us consider evidence regarding the head-direction parameter mentioned in section 1.2. The particular formulation we focus on concerns a functional category head, a C0, as the head of a CP (complementizer phrase) and the adjunction direction that correlates with it: left-headed C0 correlates with right-branching adjunction and right-headed C0 with left-branching adjunction (see Lust 1992 for detailed discussion). Given this formulation, results of several empirical studies (e.g., Flynn 1983; 1984; 1987) investigating the role of this parameter in adult L2 acquisition of anaphora indicate that from early stages of acquisition, Japanese speakers learning English as a second language (ESL) are able to acquire the English value of the parameter. (Recall that Japanese is a head-final language in that the head of a phrase comes at the end and that English is head-first in that the head of a phrase comes first.
Extensive cross-linguistic L1 acquisition studies have demonstrated that young children establish the head-direction order for their L1s at early stages of acquisition (see review in Lust 1986). It has been argued that children use this knowledge to constrain their hypotheses about other aspects of the language-particular grammar they are constructing, one such aspect being anaphora direction, that is, the relative linear ordering of a pronoun and its antecedent.
In an elicited imitation and comprehension study, Flynn (1983, 1987) investigated the acquisition of the head-direction parameter by 21 Japanese speakers learning ESL (mean age: 30;00 years; ESL placement: High score 42 [range 0-50]) (for more detailed discussion see Flynn 1983, 1987). They were tested on structures such as those shown in (13) and (14).
PRE- AND POSTPOSED SUBORDINATE ADVERBIAL CLAUSES
(13) PREPOSED: [[complementizer When ] [the actor finished the
book, [the woman called the professor]]].
(14) POSTPOSED: [The worker called the owner [[complementizer
when ] [the engineer finished the plans]]].
These structures varied in terms of the pre- and postposing of the
subordinate clause in relation to the main clause. The preposed
left-branching structure in (13) correlates with a right-headed C0; the
postposed, right-branching structure in (14) correlates with a
left-headed C0.
We hypothesized that if L2 learners had access to their L1 parameter values alone, then the Japanese speakers tested in this study would have access only to a head-final parameter value. If this were so, we would expect to find no evidence that these learners were able to identify and assign a new value to the head-direction parameter for the L2, English, in a manner observed for child L1 acquisition. We might even expect that those structures that follow from the L1 parameter setting would be more accessible to the Japanese learner than those that follow from the L2 parameter setting; that is, they might show a preference for preposed sentence structures rather than postposed structures.
Results of these studies and other related experiments reveal two very important overarching findings. First, at early stages of acquisition, Japanese adult L2 learners of English do not find preposed sentence structures significantly easier either to imitate or to comprehend. That is, they treat both of the sentence structures in (13) and (14) similarly in production and comprehension, suggesting that they "know" that English and Japanese differ in head direction and are working out the consequences of a new parameter setting for English. Second, later in acquisition (at the highest proficiency level tested), the Japanese speakers showed a significant preference for postposed sentence structures (sentence (14)) over preposed ones (sentence (13)). This finding is consistent with that for L1 learners of English and suggests that these L2 learners had assigned a value to the head-direction parameter in conformity with the English value. These results suggest that UG remains available to the L2 learner. They have been replicated with other sentence structures and with other language groups, specifically, Chinese and Spanish speakers learning ESL (e.g., Flynn & Espinal 1985).25
3.2.2. Differentially instantiated UG principles
Another empirical test of the partial access hypothesis investigates non-parametric variation. For example, if in the L1, a UG principle applies vacuously to a certain construction, the partial access hypothesis predicts that the learner will not be able to apply this principle non-vacuously to the corresponding construction in the L2. Several studies have investigated this version of the partial access hypothesis by testing L2 learners' knowledge of principles constraining syntactic movement. The effects of such principles on the acquisition of wh-questions has been of particular interest to researchers because in some languages the wh-phrase moves in the formation of questions and in others it does not (as briefly discussed above in section 2.3). Furthermore, when syntactic movement of the wh-phrase occurs, certain UG principles, generally known as "movement constraints", are applicable. Consider, for example, question formation in English from a declarative containing a relative clause (RC), as in (15a). Although it is possible to extract a wh-phrase across the relative clause (indicated by the square brackets), as in (15b), extraction out of the relative clause results in ungrammaticality, as seen in (15c).
(15) a. The girl [RC who bought the book] introduced
the boy to the clerk
b. Who did the girl, [RC who bought the book]
introduce ___ to the clerk?
c. *What did the girl [who bought ___] introduce
the boy to the clerk?
This is known as a "Subjacency" effect, since it is Subjacency, a constraint on syntactic movement, that is violated in (15c). In languages without syntactic movement, that is, where question-formation occurs without fronting the wh-word (e.g. Japanese, Chinese, Indonesian), movement constraints like Subjacency are not applicable, and in those languages, the sentence synonymous with (15c) is grammatical. L2 researchers investigating movement constraints have been interested in whether or not speakers of languages lacking syntactic movement attain knowledge of the ungrammaticality of sentences such as (15c) in a target language like English.
Movement constraints are of particular interest to L2 researchers because the knowledge tested involves rather esoteric sentence-types which are assumed to be unavailable in the (L2) input. For L1 acquisition, it is similarly assumed that knowledge of the illicitness of such sentences is not determinable from the input: First, it is generally assumed that speakers of the L1, who provide the primary linguistic data for the L1 learner, simply do not utter such sentences. In addition, if sentences like (15c) were available, as for example in speech errors, this would complicate matters even more for the language learner, since he would by hypothesis not be given any indication of the ungrammaticality of this sentence. Critically, negative evidence is required to determine ungrammaticality from the input alone. Since negative evidence of this type is not generally made available to the learner, knowledge of Subjacency and other negative constraints is hypothesized to be biologically determined, that is, given by UG.26
This argument concerning underdetermination of the data (or poverty of the stimulus) can also be shown to hold in L2 acquisition. While in L2 acquisition, especially in the case of classroom learning, negative evidence is arguably more readily available than in L1 acquisition, it is highly unlikely that negative evidence concerning these particular structure types (i.e., ungrammaticality resulting from illicit wh-extractions) is ever explicitly or implicitly given (cf. Cook 1988; White 1985). Furthermore, knowledge of this type of ungrammaticality is not a function of amount of input, that is, no matter how long a learner has been exposed to the L2, s/he, like the native speaker, is unlikely to come across these types of sentences, and is furthermore unlikely to be told of their grammaticality status.
With regard to knowledge of movement constraints, several researchers have shown that L2 learners of English from various L1 backgrounds accept violations like (15c) to a significantly higher degree than do native speakers of English. For one group of learners in one study (Schachter 1989), their ability to correctly judge Subjacency violations as ungrammatical only barely surpassed 50%, or what is considered chance level. On the basis of results such as these, it has been argued that UG is not available to the L2 learner in the same way that it is to the child L1 learner.
In a recent study, Martohardjono (1993) proposes a different approach to evaluating L2 learners' knowledge of UG principles governing syntactic movement. She argues that percentages of rejection of sentences constituting violations might not be the best indicator of learners' competence, since, in general, L2 learners perform at lower success rates across tasks, when compared to native speakers. Thus, their performance on judging grammatical sentences, which are often used as controls in these studies, is also typically lower than that of native speakers. Instead of comparing absolute rates of rejection of individual wh-constructions between native and non-native speakers, Martohardjono tested whether L2 learners' judgments of violations conform to grammatical systems allowed by Universal Grammar. That is, although L2 learners may lag behind native speakers with regard to accuracy rates, their judgments of wh-structures may still derive from their knowledge of UG principles, and conform to a pattern predicted by UG. The relevant principles were movement constraints, in particular Subjacency and the Empty Category Principle (ECP). These principles restrict movement out of certain subordinate clauses, such as relative clauses and wh-islands (so-called because they constitute clauses headed by a wh-word/phrase, such as "whether", "why", etc.).
Martohardjono included several types of movement violations predicted by UG to vary in relative strength. Extractions of wh-phrases out of relative clauses, adjuncts and sentential subjects, for example, are predicted to be strong Subjacency violations, when compared to extractions out of clauses headed by "whether" or noun-complement clauses, which constitute weak Subjacency violations. The gradience in acceptability is exemplified in (16a)-(b):27
(16) a. ??Which book did John hear [NP a rumor [CP that
you had read __?
b. *Which book did John meet [NP a child [CP who
read __?
Similarly, certain subject extractions result in strong violations when
compared to object extractions, because typically subject extractions
involve the effects of violating two movement constraints, Subjacency
and the ECP, while object extractions involve only Subjacency. This is
exemplified below:
(17) Subject Extraction, violates Subjacency and ECP
(STRONG):
* Which neighbor did John spread [NP the rumor
[CP that __ stole a car?
(18) Object Extraction, violates Subjacency (WEAK):
??Which car did John spread [NP the rumor [CP
that the neighbor stole __?
Evaluating L2 learners' relative rates of rejection across these various constructions, Martohardjono found that the same pattern emerged across all L2 groups tested, regardless of whether their L1s instantiated wh-questions with or without movement. For example, she found that strong violations, such as wh-extraction out of relative clauses and adjunct clauses were rejected at a higher rate than weak violations, involving wh-islands and noun complements. Similarly, subject extractions, involving the violation of two principles were in general rejected at a higher rate than object extractions, involving the violation of only one principle. Importantly, the patterns that arose bore striking similarity to the patterns that the native speaker control group showed for these constructions. Finally, and most significantly, the patterns that arose for both L2 learner groups as well as the native English control group was precisely the one predicted by UG in terms of relative strength of the violation.28 Crucially, for two of the L2 groups (native speakers of Chinese and Indonesian) the L1 could not have been the knowledge source, either for the ungrammaticality of the sentences, or for the particular pattern which arose, since the corresponding constructions in the L1, (displaying wh-in-situ within relative clauses, sentential subjects, etc.) are grammatical. Since knowledge of ungrammaticality could not possibly have been derived from the L1 grammar, Martohardjono concluded that UG principles constraining syntactic Wh-movement must still be available to these learners.
The subjects for this experiment are described in (19). Examples of stimulus sentences are provided in (20).
(19) Subjects:
L1 # of Ss ESL PLACEMENT LEVEL
Chinese 16 advanced
Indonesian 17 advanced
Italian 11 advanced
English Control 10
(20) Examples of stimulus sentences:
a. Extraction out of relative clause (strong violation)
Subject: **Which man did Tom fix the door that ___
had broken?
Object: *Which mayor did Mary read the book that
praised ___ ?
b. Extraction out of Adjunct Clause (strong violation)
Subject: **Which waiter did the man leave the table
after ____ spilled the soup?
Object: *Which soup did the man leave the table
after the waiter spilled ___?
c. Extraction out of wh-island (weak violation)
Subject: **Which sister did Pat know where ____ had
hidden the candy?
Object: *Which patient did Max explain how the
poison killed ____ ?
d. Extraction out of noun complements (weak violation)
Subject: **Which neighbor did John spread the
rumor that ____ stole a car?
Object: *Which car did John spread the rumor
that the neighbor stole____ ?
As shown in (21), which gives the overall results, L2 learners from these L1 backgrounds are, in general, quite able to correctly identify ungrammatical wh-questions involving these different syntactic domains.
(21) Overall results
Mean percentage correct
Chinese 65
Indonesian 74
Italian 82
English 92
The interlanguage variability in these results can be explained if we
assume that the L1 plays a role in L2 acquisition. In particular, the
languages differ in the way the principles under examination are
instantiated. In Chinese and Indonesian, for example, wh-questions are
not generated by syntactic movement, resulting in the non-application
of principles like Subjacency and the ECP. Italian, on the other hand,
is similar to English in that wh-questions are generated by syntactic
movement, triggering both Subjacency and the ECP. This
cross-linguistic L1 difference is predictably reflected in the groups'
L2 accuracy rates. While it is clear from these figures that all
groups were able to recognize movement violations in English, the two
groups having an L1 which does not instantiate the principles under
examination (Chinese, Indonesian) show a lower accuracy rate than the
two groups having an L1 which does instantiate these principles
(Italian, English). This type of interlanguage variability is
expected, and indeed predicted by a parameter-setting theory of L2
acquisition (see Flynn 1987). Of greater importance in this study,
however, are the strikingly similar patterns which arise across all the
language groups as seen in figure 5: As discussed above, extractions
out of those domains which are predicted by UG to result in strong
violations were rejected at a higher rate than extractions predicted to
result in weak violations. An ANOVA showed this differential treatment
of strong vs. weak violations to be significant for all groups at p=
.0001.
The Indonesian subjects in this particular study provide a striking example of UG specified knowledge independent of grammaticality status in the L1. In addition to being able to recognize violations of movement constraints in general, they evidence a more subtle knowledge of degree of unacceptability in the L2 which is independent of their knowledge of the L1. As noted earlier, Indonesian questions do not involve overt syntactic movement, and therefore no movement violations occur in these structures. However, questions involving wh-islands and NP complements in this language turn out to be unacceptable or highly marginal for other reasons.29 Examples of these constructions are shown in (22).
(22) a. Wh-islands
*Siti ingin tahu dimana Adik menyembunikan apa?
Siti want know where Adik ACT-hide what?
'What does Siti wonder where Adik hid ___?'
b. NP complement
?*Siti dapat kabar bahwa saudaranya pergi kemana?
Siti receive news that relative-POSS go where?
'Where did Siti receive the news that her relative
went ___?'
In other words, the acceptability status of the corresponding sentences
in Indonesian is precisely the reverse of the weak/strong effects for
English. Questions involving adjuncts and relative clauses, which in
English constitute strong violations are perfectly acceptable in
Indonesian, since in this language the wh-phrase does not move in these
sentence-types and hence does not violate movement constraints. In
contrast, questions involving wh-islands and NP complements, which in
English constitute only weak violations, are highly marginal, if not
unacceptable in Indonesian. These facts, however, do not seem to affect
the L2 learners' ability to differentiate correctly between the two
types of violations when learning English. As the results in (23)
indicate, their judgments of these sentences pattern in the same way as
those of native speakers. That is, they weakly reject NP complements
and wh-island violations, and they strongly reject extractions out of
relative clauses and adjuncts.
(23) Mean percentage correct
Strong Weak
Indonesian 88 46
English 99.0 78.0
We argue that this pattern is not a coincidence,30 but that it results from the speakers' knowledge (both those for whom English is the L1 and those for whom it is the L2) of the differential effects of movement constraints, on sentences generated by syntactic movement.31
3.2.3 Error data from adult L2 acquisition
A third piece of compelling evidence against the partial-access hypothesis emerges from error data collected from the same set of Japanese speakers who were tested on pre- and postposed sentence structures (see section 3.2.1). Elicited imitation of sentences such as (24) was extremely difficult for these learners of English, as shown in figure 6.
(24) When the doctor received the results, he called the gentleman.Sentence (24) displays a preposed, left-branching adverbial adjunct clause as well as forward pronoun anaphora. In forward pronoun anaphora structures such as (25) below, the antecedent (Taroo) precedes the null pronoun form (interpreted as "he" in this sentence). Sentences such as these do not involve a surface contrast in the relative order of: (i) the antecedants and the proform and (ii) the subordinate and matrix clauses. These sentences are in accord with Japanese as a head-final (or right-headed CP) language.
(25) Taroo wa nyuusi no kekka o kiita toki
Taroo-topic entrance exam-possessive result-accusative heard when
/ hahaoya ni denwa sita
pro mother-dative telephone did.
'When Taroo heard (found out) the results of the entrance exam,
(he) called his mother.'
The extremely high error rate on these sentences is expected given a UG parameter setting model of grammar acquisition. As noted earlier, the Japanese speaker learning English must assign a new parameter setting for head direction. In L1 acquisition, the setting for this parameter has been argued to involve a correlation between the resultant configuration (head-initial or head-final) and anaphora directionality. When head-complement order is principally head-initial, forward directionality (i.e., antecedent precedes pronoun) is productively licensed.
If the Japanese L2 acquisition of English is constrained by the same parameter as is child L1 acquisition, and if the L2 learner draws the same empirical consequences as the L1 learner, then the Japanese L2 learners' difficulty with sentences such as (24) can be explained in the following way. Sentence (24) is not only inconsistent with the new parameter setting (to head-initial direction) of English as a head-initial language, but it also violates the empirical consequence regarding anaphora direction that correlates with this parameter setting. In this case, forward anaphora correlates with head-initiality. Results such as these are left unaccounted for within a model disallowing access to principles and parametric values independent of the L1 grammar.
3.2.4 Non-transfer of language-specific aspects
Results of a study investigating the L2 acquisition of the English "control" verbs promise, remind and tell by adult Spanish speakers also suggest that the L2 learner is constrained by UG rather than by the L1 alone (see Flynn, Foley & Lardiere 1991).32
L1 acquisition of the control structures exemplified in (26a) has been widely studied (e.g., C. Chomsky 1969; Sherman 1983; Sherman & Lust 1993; Maratsos 1974; Tavakolian 1978; McDaniels & Cairns 1990). The results from these and other studies indicate that: i) in comprehension, L1 learners of English interpret all control verbs as if they were object control verbs and ii) in both production and comprehension, L1 learners of English show a general overall preference for the infinitive structures exemplified in (26a) below when compared to their tensed counterparts illustrated in (26b).
Similarly, early L2 acquisition studies focusing on the comprehension of the structures in (26a) indicate a developmental pattern in which adult L2 learners interpreted subject control verbs as if they were object control verbs at early stages of acquisition (d'Anglejan & Tucker 1975; Cooper et al. 1979). In addition, these comprehension studies found no evidence that the L2ers attempted to "translate" or to "map" their native language structures onto those of the L2 even when the L1 would have provided a correct response for the English structures tested.
Given the comparable patterns of acquisition that L1 and L2 learners of English showed in comprehension of control verb structures, the next question asked was whether their patterns would also match for production. To test this, 21 adult speakers of Spanish learning ESL were tested in their production of the structures exemplified in (26aPb).33 The Spanish speakers tested were at either an Intermediate (n=9) or an Advanced (n=12) level of competence as measured by the standardized placement test from the University of Michigan.
As exemplified in (27aPb), Spanish clearly has both infinitives and tensed clauses. Spanish infinitivals also involve a null pronoun as in English (PRO) (27ai); however, the tensed counterparts involve a pronominal form (pro) (27biPiii) not allowed in (subject) position in an English tensed clause. All three control verbs tested in this study (promise, remind, and tell) can occur in Spanish tensed clauses, as in English. However, infinitives are not allowed with tell or remind in Spanish; that is, Spanish does not have counterparts to (26aiiPiii). The method used in this study was an elicited imitation task, in which subjects were asked to repeat sentences verbatim as given by the experimenter. The sentences administered are illustrated in (26).
(26) a. Infinitives
i. John promises Henry PRO to go to the
store.
ii. John reminds Henry PRO to go to the
store.
iii. John tells Henry PRO to go to the store.
b. Tensed finite
i. John promises Henry that he will go to the store.
ii. John reminds Henry that he will go to the store.
iii. John tells Henry that he will go to the store.
(27) a. Infinitives
i. Juan le promete a Henry PRO ir a la tienda.
Juan him promises to Henry PRO to go to the store
'Juan promises Henry PRO will go to the store.'
b. Tensed finite
i. Juan le promete a Henry que pro ir a la tienda.
Juan him promises to Henry that pro will go to the store
'Juan promises Henry that he will go to the store.'
ii. Juan le dice a Henry que pro vay a la tienda.
Juan him tells to Henry that pro go to the store
'Juan tells Henry to go to the store.'
iii. Juan le recuerda a Henry que pro ir a la tienda.
Juan him reminds to Henry that pro will go to the store
'Juan reminds Henry that he will go to the store.'
As shown in figure 7, Spanish speakers significantly preferred infinitives over finite that-clauses for the verbs tested (promise, remind, and tell ) when they were asked to repeat both types of sentence in an elicited imitation test. Importantly, these results replicate those for L1 acquisition of English (see review in Sherman & Lust 1993).
Given the results in figure 7 we reason that if only the learner's L1, and not UG, constrained L2 acquisition, we would expect only L1-instantiated lexical properties to be available to the L2er. That is, at the very minimum, we would expect the Spanish speakers to have found remind and tell sentences with finite that-clauses to be significantly easier to acquire than the infinitive structures for these verbs. However, these results indicate that this is not the case. These results, along with: (i) those from Japanese and Chinese speakers--which match those for the Spanish speakers--and (ii) those from earlier comprehension studies that indicate that L2 learners, like L1 learners, interpret both subject and object control verbs as object control verbs, have been used to argue that a general principle of locality plays a role in the L2 acquisition of English. (See Sherman and Lust 1993 for details of a similar proposal in L1 acquisition.) Briefly, we argue that regardless of their L1s, L2 learners prefer object-controlled, infinitive structures because in these structures an object antecedent minimally c-commands PRO; that is, the nearest c-commanding element that appears in these structures controls PRO.34 In the infinitive structures in English, there is a control domain within which the object is the minimal c-commander of the proform, PRO. In the finite that-clauses, (under their analysis) neither the subject nor the object minimally c-commands the subject pronoun in the subordinate clause. Both L1 and L2 learners of English seem to rely on this locality principle at early stages of acquisition. Consequently, adults need to learn to override this linguistic principle (and it appears that they do), in order to learn the lexical language-specific subject control properties of, for example, promise. Again, the important result is that the patterns of acquisition isolated for the Spanish, Japanese and Chinese speakers seem inexplicable in terms of "transfer from the L1," but can be accommodated if
one assumes that UG remains available to the adult L2 learner.
3.2.5 Summary
In conclusion, the empirical results we have reviewed in this section do not support the partial-access hypothesis. L2 learners do seem to construct a grammar of the new target language under the constraints imposed by UG; the principles and parameters of UG carefully investigated thus far indicate that both those not instantiated in the L1 and those that apply vacuously in the L1 but are operative in the L2 can be acquired by the L2 learner.35
3.3 The L2 acquisition of functional categories: A case study
We now turn to a detailed consideration of a particular version of the partial-access hypothesis, one concerning the acquisition of syntactic categories--namely, the Weak Continuity Hypothesis proposed by Vainikka and Young-Scholten (VYS) (1991) (see also Eubank 1988, to appear; Schwartz 1993 for discussion).36
On the basis of data collected from several types of production tasks, VYS claim that certain syntactic categories (namely, functional categories; see below) are initially absent from the grammars of L2 learners and that these categories progressively emerge in discrete stages.37 Because VYS's proposal provides interesting arguments in support of the partial-access hypothesis, we focus on the theoretical (syntactic), empirical, and methodological (experimental) issues that it raises. In section 4 we will discuss possible alternative analyses of VYS's data..
3.3.1. Methodology
3.3.1.1. Experimental Tasks
VYS's version of the Weak Continuity Hypothesis posits L2er knowledge of only those syntactic categories that are represented by a certain percentage of phonetically correct forms in their subject's speech, collected from five elicited production tasks, each involving descriptive narration on the part of the subject. In two of the tasks "the test subjects were asked to tell the stories in comic strips which contained minimal or no text" (p.9). In the other tasks they were required to describe a picture, to describe an action that the experimenter had carried out, and to use the present and the present perfect tense to describe a particular process.
First, to rely on the data collected from these particular production tasks alone, is in our view problematic. Although inferences concerning linguistic knowledge must be based on performance data of some sort, the type of production data elicited in VYS's experiments do not provide a very reliable indication of learners' knowledge of grammatical categories. Production tasks of this kind make increased performance demands as compared to other tasks, given the lack of experimental controls. Thus, it is highly likely that the absence or incorrect instantiation of certain syntactic categories in these "naturalistic" types of production tasks is due to a deficiency in production or performance, rather than to a knowledge deficit. To take one example, contra VYS, it is not at all clear to us that failure to produce (complex) sentence embedding necessarily indicates the absence of the syntactic category characteristically employed in the analysis of embedded sentences (aswe will discuss in more detail below).
Moreover, even if a deficiency in the grammar (i.e., a linguistic-knowledge deficit) is in fact responsible for this particular performance/production failure, it is by no means clear that the deficiency resides in the syntactic component. Such deficiencies could, for example, be lexical. A learner may simply have not yet correctly associated the appropriate lexical item with its grammatical category. It is altogether possible that the absence of the German complementizer da' 'that' in some of the utterances of the L2 learners in VYS's data (which VYS interpret as critical evidence for the absence of the syntactic category COMP(Lementizer)) is instead due simply to the L2 learners' failure to acquire this particular lexical entry (word).
Similarly, the observed deficiencies may in principle be phonological. For example, to attest the presence or absence of tense in English, one would presumably monitor L2 learners' production of verb-final /-s/ or /-d[-ed]/. However, if absent, the absence of these morphemes (suffixes) could in fact be due to a reduction of unstressed phonemes or clusters in certain morphological environments--in this case, word final position. In order to eliminate this as a possibility, one would have to monitor the environments in which these morphemes were present or absent, as has been proposed for L1 acquisition (see Demuth 1992).
In addition, these types of naturalistic production tasks do not incorporate the controlled manipulation of target structures relevant to the factors being investigated. Critically, the experimenters had to wait for the appearance of target structures in the speech stream. Testing grammatical knowledge in this manner is not always efficacious or ultimately reliable, and yields data that are arguably inconclusive. It is also potentially problematic to employ an experimental task in which subjects must narrate in a particular tense (e.g., present, as in VYS's study). First, the subject might have difficulty with that particular tense, resulting in high error rates throughout. Second, the present tense in English, for example, provides an infelicitous (or inaccurate) characterization of a given event (e.g., a single event depicted in a comic strip), since present tense expresses habitual action in English. Thus, showing a subject a picture of, say, a boy walking toschool and asking the subject to narrate in the English present tense ("A/The boy walks to school") arguably forces the subject to make infelicitous (or perhaps false) statements (as noted in Hamburger and Crain 1982, p. 256). In German, present tense is ambiguous between habitual and present action. Finally, presenting subjects with comic strips containing "minimal or no text" (VYS, p. 9) is potentially problematic: Which comic strips had which texts? Which comics had no text? How do the text-comic pairings differ? How do these differences in the (very complex pictorial and orthographic) stimuli presented influence or affect the subjects' linguistic performance in the these quite different elicited production (descriptive narration) tasks?
3.3.1.2. Use of percentage correct as an indication of grammatical knowledge
An additional point, although one that is not unique to this study, concerns the sole use of percentage correct in the context of natural-speech data to substantiate the (non-) existence of knowledge of syntactic categories. VYS set >60% correct usage of a particular form as a criterion that it has been acquired. We argue that this inferential method is, for several reasons, unsound.
First, as we have already suggested, the tasks used to evaluate grammatical knowledge in these studies do not lend themselves to frequent occurrence of the target structures. Second, even if a structure is isolated by such a method, in our view the likelihood of its being correct is low, given the performance demands inherent in the tasks themselves, which might well mitigate against their correct use. Third, it is simply not clear whether there is a correlation between any percentage of correct usage of a particular aspect of grammar and knowledge of that aspect. Thus, it is conceivable that a learner may in fact know the target language (i.e., the learner's grammar generates all the target structures) but the learner never uses certain structures, or uses them incorrectly, for performance reasons. As Chomsky (1991, p. 19) notes, this is certainly true for native L1 speakers. An infinite number of grammatical sentences in any language are not usable--for example,all grammatical sentences (of which there is an infinite number) exhibiting three or more center-embedded relative clauses.
To illustrate, 28(aPc) are all grammatical. However, (28a) and (28b) are both usable (parsable), but (28c) (containing a triply center embedded relative clause) is not.
(28) a. The rat died.
b. The rat the cat chased died.
c. The rat the cat the dog bit chased died. (meaning:
"The rat which was chased by the cat (which was, in
turn, bitten by the dog) died.")
Given this crucial distinction between grammaticality and usability (a
particular subcase of the crucial competence (knowledge) vs.
performance (behavior) distinction), it is particularly problematic to
attribute a knowledge deficit to a subject merely because he fails to
use a particular construction, especially in naturalistic production
tasks not specifically designed to elicit the particular structures
under investigation (such as the tasks employed by VYS). As Poeppel
and Wexler (1993) succinctly put it: "Absence of evidence is not
evidence of absence."
Of course, it is an empirical issue whether a particular subject in a particular experimental study is omitting (or less frequently using) a given construction for performance reasons. If the omission is due to performance, and not to a knowledge deficit, then decreasing experimental task-demands should result in increased occurrence rates (see section 5 below)--an unexpected correlation if a knowledge deficit is involved. Needless to say, it is an empirical issue which, if any, performance factors are in fact implicated (and in what way they are implicated) in any given experimental result (see Lust, Chien & Flynn 1987).
3.3.1.3. Use of correct morphology
A final point in the context of methodology involves reliance on correct morphology as evidence of syntactic knowledge (regardless of the percentage correct assumed to be criterial for acquisition).
For example, the correct use of subject-verb agreement suffixes requires special morphological knowledge, not just knowledge of the syntax, an issue to which we will return (see also related discussion in Hyams and Safir 1991). Suppose, for purposes of illustration, that an L2 learner of English had the agreement morphology (i.e., suffixal) analysis of English present tense "completely backwards"; that is, suppose the learner thought that 3rd person singular agreement in the present tense displayed no overt agreement suffix (e.g., He eat_) and that all other forms exhibited the -s verbal agreement suffix (e.g., They eats). In such a case, the subject would never use agreement morphology/suffixes correctly, as in (29a) and (29b).
(29) a. He/She/It like_ the boy.
b. I/We/You/They likes the boy.
Although strongly suggesting a morphological (suffixal) misanalysis,
such 100% incorrect use (which we will show was not the case for VYS's
subjects) does not necessarily reflect incompetence regarding the
syntax of subject-verb agreement.38
Notice first that our hypothetical learner, although having the morphology mixed up, does seem to know that the grammar of English incorporates a rule of subject-verb agreement--that is, the learner has not posited any of a number of possible but non-English rules such as object-verb or object-preposition agreement. Given such knowledge and given that "subject" (a relational syntactic notion, i.e., "subject of a sentence") and "verb" (a syntactic category of a particular type, distinct from, say, noun or sentence) are purely syntactic notions, it is clear from this alone that ignorance of the morphology (i.e., ignorance of correct verbal agreement suffixes) does not entail (total) ignorance of the syntactic analysis. Thus, requiring correct morphology (correct verbal agreement suffixes) is a highly questionable criterion to employ in assessing the learner's syntactic knowledge.
Moreover, notice that our hypothetical learner, who has the morphology wrong 100% of the time, could also have more detailed native syntactic knowledge beyond the syntactic knowledge of the English rule/law that the syntactic subject agrees with a particular type of syntactic category, namely, the verb. First, our hypothetical learner could also know that subject-verb agreement is in a certain sense "local"; that is, the subject agrees not with just any verb in the sentence, but with a/the local (nearby) one.
(30) a. *He say they likes Bob
b. He says they like Bob.
Second, our hypothetical learner could possess the syntactic knowledge
that subject-verb agreement is blocked whenever sentential negation
(not) appears, exactly as is the case in the target grammar English.
(31) a. He likes the boy
3s 3s
b. *He not like+s the boy (Direct subject-verb agreement is
blocked)
3s 3s
c. He does not like the boy
3s 3s
That is, our "morphologically backward" learner could nevertheless
evince the target grammar knowledge that (1) the syntactic
(transformational) rule of do-insertion (Chomsky 1965) is obligatory
whenever sentential negation appears, (2) do is inserted after the
subject but before negation, not (a syntactic (i.e., word order) fact),
and (3) it is do, not the main verb, that bears the agreement suffix (a
morphologically incorrect agreement suffix for our hypothetical
subject).
(32) Incorrect morphology, correct syntax
a. Incorrectly, 3s = no suffix; correctly, do required with
negation
He/She/It do_ not like the boy
3s 3s
(not: *He/She/It not like_ the boy)
b. Incorrectly, 3pl = -s suffix; correctly, do
required with negation
I/We/You/They does not like the boy
3pl 3pl
(but not: *I/We/You/They not likes the boy)
Thus, even though displaying the wrong morphology (verbal agreement
suffixes) 100% of the time, the learner would seem to have the
following native-speaker syntactic knowledge--specifically, knowledge
(33) a. of a distinct set of syntactic categories including
the category verb, which displays agreement morphology,
as distinct from, say, prepositions, which do not
b. that English is a subject- (not, e.g., object-)
agreement language
c. that English exhibits the syntactic order
Subject-(do-Neg-)Verb-Object
d. that subject-verb agreement is "local"
e. that do-insertion is obligatory when sentential
negation appears and do is a verb inserted into a
position preceding negation but following the
subject (a word order, syntactic fact)
f. that when do-insertion applies, it is do and not
the main verb that bears the verbal agreement suffix,
in the Present Tense.
g. In the Present Tense, there is one verbal suffix
for 3s and one for all other person, number (and
gender) forms.
One way of providing a unified syntactic analysis of such knowledge
regarding the varying distribution (syntax) of the agreement suffix
(the knowledge that in some cases the subject agreement verbal suffix
appears on the main verb (like in (29)) but in other cases it appears
separate from the main verb, on do (as in (32)), is to postulate that
all cases derive from a single type of underlying (abstract)
D-structure (DS) representation of the sentence, in which the agreement
suffix invariably appears separate from the main verb (Chomsky 1965;
1991c).39
(34) a. DS [Sentence [Subj He] [Agreement -s] [verb
like] [the boy] ]
3s 3s
b. DS [Sentence [Subj He] [Agreement -s] not
[verb like] [the boy] ]
3s 3s
This unifying abstract type of representation clearly does not undergo
"rules of pronunciation"; that is, it is not an accurate representation
of the pronounced order of elements. Rather, D-structure undergoes
transformational rule application (i.e., application of rules which
move, delete, or insert syntactic categories (e.g., NP and V), yielding
what is called "an S-structure representation" that (for our purposes)
does undergo pronunciation. In the S-structure representation, the 3rd
person singular agreement suffix -s must become affixed to a verb
(Lasnik 1981). Thus, neither D-structure representation in (34) can
"survive" as a well-formed S-structure (pronounced) representation,
since both contain the unaffixed affix -s, an ill-formed object with
respect to pronunciation. In order to generate the desired S-structure
(pronounced) forms, transformational rules (in particular movement
rules) must apply to the D-structure representation and attach the
affix to a verb. In (34a), the rule of "affix-hopping" (Chomsky 1965)
can apply, moving the affix -s over, and attaching it to, the immediate
right of the verb like. The result is the following S-structure
representation containing the complex verb likes, consisting of the
uninflected verb like plus the 3rd person singular agreement suffix
-s.
(35) SS [Sentence [Subj He] [Agreement ] [verb like]+[Agreement
-s] [the boy]]
3s 3s
In (36), interestingly, the appearance of not prevents the application
of affix-hopping.
(36) SS *[Sentence [Subj He] [Agreement] not [verb like]+[Agreement
-s] [the boy] ]
3s 3s
Given that affix-hopping is blocked whenever sentential negation not
appears, in order to meet the S-structure requirement that the verbal
agreement affix -s appear affixed to a verb, the so-called dummy
auxiliary verb do is inserted by a transformational rule to "support"
the "stranded" affix, yielding the (pronounced) S-structure
representation (37).
(37) SS *[Sentence [Subj He] [Aux do] [Agreement -s] [Negation
not] [verb like] [the boy]]
3s 3s
Given that our "morphologically backward" hypothetical learner could indeed know all these (as well as other) relevant (and quite complex) syntactic facts regarding English agreement, while at the same time having the incorrect pairing of subjects with suffixes, we would attribute the target grammar (English) syntactic analysis of subject-verb agreement to him, while at the same time hypothesizing that his morphological analysis was indeed incorrect.
The methodological point, we hope, is clear. Use of nontarget morphology (e.g., incorrect agreement suffixes), even incorrect use 100% of the time, by no means entails a nontarget analysis of the syntax of agreement. In fact, we will strongly suggest that such a distinction between morphological and syntactic knowledge is not restricted to our hypothetical learner, but may well accurately characterize certain subjects in VYS's study. We will argue that experimental data VYS interpreted as indicative of a nontarget syntactic analysis are more readily interpretable as evidence of target-grammar syntactic analysis co-occurring with a misanalysis of the morphological facts. Thus, VYS's reliance on correct morphology as an indicator of syntactic competence may well result in a significant underestimation of their subjects' acquisition of the target syntactic analysis. That is, VYS's results may simply indicate that their subjects have not yet mastered the morphology of verbal agreement. Importantly, such a result would be unsurprising, if not expected, since agreement morphology is cross-linguistically idiosyncratic; in other words, it is not a universal, invariant property of human language but a property of particular languages and therefore must be "learned" (by L1 learners as well).
We agree with VYS that the subjects investigated in their study have not acquired "the full agreement paradigm." Crucially, though, we take this to mean only that they do not know the following idiosyncratic, nonuniversal, morphological facts about verbal agreement suffixes in German:
(38) -e = 1s
-o = 1s (colloquial)
-st = 2s
-t = 3s and 2pl
-n = 1pl and 3pl
There is one final methodological point to be made in this context:
VYS's studies indicate a high occurrence of -n in their data; they
suggest that it represents only the infinitive ending and therefore
argue that its appearance is consistent with their hypothesis that
these subjects' grammars lack agreement. We do not agree with this
conclusion. As (38) indicates, -n is indeed an agreement suffix in
German, in addition to being the infinitive suffix. Since it serves
this dual function, it may well have the highest distribution of the
suffixes and would therefore constitute the "best guess" by a subject
who has not mastered the paradigm in (38). A learner might well know
that an agreement suffix is needed, but not know which one, and
therefore pick -n--another case where the learner knows the underlying
morpho-syntax but not the surface morpho-phonetics.
The methodological questions we have raised significantly weaken the theoretical claims made in the Weak Continuity Hypothesis studies. Nevertheless, we will consider several of the (methodologically independent) theoretical arguments developed in VYS 1991. We will consider these arguments in terms of each of the nontarget syntactic stages postulated by VYS in their account of the acquisition of German as an L2.
We first outline what we tentatively assume constitutes the target syntactic analysis of German. Having done this, we can then isolate more precisely the formal differences between the target syntactic analysis and the nontarget misanalyses that VYS attribute to their subjects learning German as an L2. This approach is necessary (though, unfortunately, not ubiquitous) in all acquisition research. That is, before one can ask "How is this (aspect of) grammar (e.g., the syntax of subject-verb agreement) acquired?", one must first (tentatively) adopt an explicit analysis (e.g. the affix-hopping analysis outlined above) of what this (aspect of) grammar is. (Readers unfamiliar with X' theory and the hypothesized class of syntactic categories and projections may wish to read the Appendix before turning to the next section concerning those aspects of the analysis of German relevant to our discussion of VYS.)
3.3.2. The target analysis of German
German displays the following word orders, among others (from Haegeman 1991, Sec. 11.2, p. 524):40
(39) a. ... dass Hans das Buch kauf+t
Complementizer Subj. Obj. verb + inflection
... that Hans the book buy+s
'...that Hans buys the book'
b. Kauf+t Hans das Buch?
buy + s Hans the book
'Does Hans buy the book?'
c. Das Buch kauf+t Hans.
the book buy+s Hans
'Hans buys the book'
In an embedded declarative clause introduced by the complementizer da'
(39a), the inflected verb occurs sentence-finally. In an unembedded
yes-no question (39b), the inflected verb is sentence-initial. In an
unembedded declarative clause (39c) the verb is in so-called "second
position"; that is, it appears immediately after a sentence initial
phrase, here the noun phrase das Buch. This is known as the
"verb-second" or "V2" phenomenon.
All three word orders can receive a unified analysis by virtue of being derived from one and the same type of underlying (D-structure) representation in which (unlike English), V is final in V' and Agr is final in Agr'. That is, we will assume here that German has underlying Comp-Subject-Object-Verb-Inflection order, as represented in (40), conforming to the X' theory of phrase structure.41,42
Thus, it is hypothesized that in German, unlike in English: (i) the direct object precedes the verb and (ii) Agr follows the VP. As noted earlier, in connection with Japanese, although the hierarchical (dominance) relations imposed by X' theory are assumed to be universal (e.g., each head projects to a single-bar and to a phrasal projection, with specifier and complement optional), the order of the head and its complement is not universal, but instead varies cross-linguistically.
The S-structure representation of the embedded clause in (39a) is derived from the D-structure representation in (40) by applying head-to-head movement; that is, the verb kauf, a head, is moved to the inflectional suffix -t, which occupies the head position Agr, so that the suffix is properly affixed to the verb as is required in the S-structure representation. Example (39b) is derived by applying the same movement transformation, yielding [kauf+t], but then [kauf+t] undergoes additional head-to-head movement from Agr to the Comp position (which is empty in this case, da' being absent), yielding the verb-initial word order. Example (39c) is derived by applying the same two (head) movement transformations, plus (phrase) movement of the direct object NP das Buch into Spec CP (a phrasal position). Applying these three movements (as indicated by the arrows) to the D-structure representation (40) produces the S-structure representation of (39c) shown in (41) (adapted fromHaegeman's (16e), p. 529).
Thus, cases of V2 like (39c) are analyzed by positing V-movement first to Agr and then to (the "second position") Comp, with a phrase (here the direct object NP das Buch) moving into the preceding "first position," Spec CP. Unification is achieved to the extent that all the different surface word orders are derived from a single (invariant) type of X'-consistent (unpronounced) D-structure representation like (40), by application of independently motivated movement rules.
This movement analysis of V2, which hypothesizes that the verb's appearance in second position is derived from a unitary D-structure representation by movement of V into an unoccupied Comp position, additionally makes the empirical prediction that V2 is blocked when the Comp position is filled (by, e.g., the complementizer da'). The prediction is confirmed by the obligatory appearance of non-V2 (and instead V-final) order in embedded clauses like (39a), introduced by da'.
3.3.3. The evidence for stages: Theoretical assumptions
In this section we consider the existence of the stages (i.e., grammars) proposed by VYS (1991) in the L2 acquisition of German. We argue that even if one completely accepts their methodology (reviewed in detail in 3.3.1 above), their results still fail to provide compelling support for the existence of the stages of acquisition that they postulate.
VYS argue that there exist at least the following three discrete and ordered stages of syntactic development in the acquisition of German as an L2:
(42) a. Stage 1: The VP-stage analysis. The grammar lacks
the functional heads C, I, Agr, and their corresponding
single-bar and phrasal projections. Consequently
sentences are analyzed as:
b. Stage 2: The IP-stage analysis. The grammar lacks
the functional heads C and Agr and their corresponding
single-bar and phrasal projections. Consequently
sentences are analyzed as:
c. Stage 3: The AgrP-stage analysis. The grammar
lacks the functional head C and its corresponding
single-bar and phrasal projections. Consequently
sentences are analyzed as:43
At each stage, sentences are analyzed as instances of the tree
provided. A learner in the VP-stage analyzes the German sentence as a
VP; in the IP-stage as an IP; and in the AgrP-stage, as an AgrP.
Crucially, each stage is simply a type of grammar and to say "Person P
is in/at a particular stage (e.g., the VP-stage)" is to say no more and
no less than "Person P's knowledge of language L is characterized by
grammar G exhibiting the formal properties F." For example, at the
VP-stage, it is assumed that the categorial inventory of the grammar
altogether lacks the functional categories Infl, Agr, Comp, and their
corresponding X' projections (VYS 1991, p. 18). VYS postulate that it
is this property of the grammar, namely, an 'impoverished' categorial
inventory (not, e.g., some imaginable performance effect), that
accounts for the purported VP-analysis of German sentences by L2ers in
this stage.
VYS's hypothesis that a given subject is at the VP-stage entails that such a subject has no knowledge of the functional syntactic category Agr. Then, because all syntactic operations concerning subject-verb agreement (e.g., English affix-hopping or German V-movement to Agr, discussed earlier) crucially refer to the syntactic category Agr, one empirical prediction is that such a subject has no knowledge regarding the syntax of subject-verb agreement. Clearly, this prediction is a very strong one that, if incorrect, should, in principle, be readily falsifiable.
3.3.4. The VP-stage
As noted, at the VP-stage the grammar is assumed to lack functional categories, and sentences are analyzed as VPs (as in (42a)). VYS assign three of their seventeen subjects to the VP-stage: Aysel, Memduh, and Changsu.44 We argue that the evidence VYS present does not support their claim that the L2 German grammar of these speakers altogether lacks functional categories. To support the existence of such a grammatical stage, one would minimally have to find a speaker whose utterances in the corpus consistently lack evidence of an IP.45 However, as VYS themselves point out, none of the speakers they assign to the VP-stage fits this description. Instead, all three speakers produce some sentences, albeit few in number, that by VYS's own criteria must be analyzed as containing an IP. For Aysel (see VYS, p. 22 and table 4), 12% of his sentences display a word order that in VYS's view is amenable to an analysis in which these sentences "represent the subsequent stage, with verb raising [to Infl] being the first indication of this." With respect to the sentences that Memduh produced, VYS note that "6% involve an agreement form of the copula or a modal, presumably in Infl"[one of the categories the VP-stage is presumed to lack] (our emphasis). For Changsu, the proportion of sentences indicating the existence of Infl was 16%.
VYS do not discount these data; nor do they claim that these subjects are in the next (IP-) stage (if they did, there would be no evidence supporting the existence of the VP-stage). Rather, to account for these sentences, VYS suggest that they are "precursors of the next stage of acquisition" (p. 23; see also p. 24) and state that "we find some indication...that these speakers might be in the process of entering the next stage, with a higher functional projection" (p. 22).
What does it mean to claim that these VP-stage speakers might be in the process of entering the IP-stage? If stages are simply grammars (i.e., "at the VP-stage" means "has a grammar the categorial inventory of which lacks functional categories"), what can it mean when a VP-stage speaker (whose grammar by definition lacks functional categories) is "in the process of entering" the IP-stage (the grammar of which has at least one functional category, Infl)? This suggests that such a speaker is in some (unspecified) intermediate stage equivalent to neither the previous nor the following stage. But in the case at hand, if Aysel, Memduh, and Changsu are neither in the VP-stage nor in the IP-stage, they cannot provide evidence for a VP-stage (because they are not in it). Hence, all the evidence for such a stage becomes questionable.
Similarly, trying to account for particular data by suggesting that they are "precursors of the next stage" fails to address the empirical, linguistically significant question regarding the subjects' knowledge of German: does the syntactic inventory of the subjects' grammar of German contain functional categories, or doesn't it? VYS themselves presume that 12%, 6%, and 16% of these three subjects' sentences are to be analyzed with a category "in Infl." They write that "it is certainly possible that these speakers have at least an IP" (p. 24) and (as noted earlier) "we find some indication...that these speakers might be in the process of entering the next stage, with a higher functional projection" (p. 22). But if any syntactic analysis attributed to them is presumed to include an Infl node, then it must also be presumed that their grammar contains Infl. Otherwise, how could they construct an analysis containing this category? But if their grammars contain Infl, itcannot be presumed that these subjects are in the VP-stage. Again, such subjects can therefore provide no evidence for the existence of this stage.
Why do VYS not attribute functional categories to the categorial inventory of these subjects' grammars? They themselves concede that the empirical support for the VP-stage versus a grammar incorporating functional categories is tenuous, writing, "It is difficult to distinguish between this approach and our approach based on empirical data" (pp. 24-25). One reason why they do not attribute functional categories in these three cases may be statistical. VYS justify attributing a VP-stage grammar to these three subjects since "the majority of the sentences (i.e. those containing a verb) produced by these three speakers look like bare VPs" (p. 24). Further, they state that "if there is an IP projection in all of their VP sentences, we would have to explain why in most sentences there is no evidence for the IP" (p. 24). But clearly, the implication here is that (by VYS's own criteria) some sentences do constitute evidence for the presence of IP.
The remaining question is, "What exactly are the data that are interpreted as indicating that these subjects 'might be in the process of entering the next stage, with a higher functional projection'?" The data in question indicate that the subjects' grammars do indeed have functional categories. If the data is reliable, the evidence for the VP-stage consequently disappears. If the data are unreliable, the next question is, "Were the only unreliable data precisely the data supporting the evidence of functional categories?" If so, a principled account of this must be provided. If not, we must determine precisely which data are reliable and which are not (see section 3.3.1, and below, for discussion of methodology).
In summary, we hope to have shown that, even accepting VYS's methodology, the arguments they provide for the existence of a VP-stage, at which the L2er's grammar of German altogether lacks functional categories, are arguably uncompelling.
Before we turn to the IP-stage, it should be noted that the VP-stage grammar, as postulated by VYS, is in large part formally underspecified. Consequently, our discussion of it has been (necessarily) quite restricted in i