Universal Grammar Model & Second Language Acquisition PDF

9 The Universal Grammar Model and Second Language Acquisition The model of language acquisition most associated with SLA research in recent years has been the Universal Grammar (UG) model developed by Chomsky, which uses the principles and parameters syntax described in the previous chapter. This association between linguistics and SLA research reveals some of the potential and some of the dangers involved in the adoption of a complex and specialised learning model. The present chapter introduces the main themes of current UG-related SLA research at the time of writing. The area is, however, vast and rapidly changing. Recent research on UG lines can usually be found in the journals Second Language Research and Studies in Second Language Acquisition. 9.1 THE UNIVERSAL GRAMMAR MODEL OF LANGUAGE ACQUISITION Chapter 1 introduced the Language Acquisition Device (LAD) model of language acquisition, which has been implicit in many theories we have discussed, for exampIe in Krashen's Input Hypothesis. The LAD model in Chomsky (1964) is essentially as follows: Language Input ----+ Output (a grammar Acquisition Device (language data) of a language) Acquisition is the process through which language data goes into the LAD 'black box' and a grammar comes out; the LAD evaluates alternative grammars to see which best fits the incoming data. The UG theory fleshes out this model by establishing the crucial features of the input, the contents of the black box, and the properties of the resultant grammar. The major change is a shift to describing grammars in terms of principles and parameters, which are purpose-designed to account for Li acquisition, rather than ruIes. The principles are unchanging regard- less of the actual language involved. Taking examples from the last chap- ter, according to the principles of X-bar syntax all phrases have two levels; according to the Projection Principle alilexical items project their syntactic 200 Universal Grammar Model 201 specifications on to the phrases of which they are heads. These principles form part of the language faculty of the mind - Universal Grammar. The claim is that it is impossible for a human mind to know a language without knowing the X-bar and Projection principles, since they are already pres- ent inside it. UG allows for variation between languages through para- meters; languages can only vary within the pre-set limits for a particular parameter. The parameter itself is universal but the values it may take vary from one language to another. Any language a human being knows must, among other things, be either pro-drop or non-pro-drop, must have a setting for each of the parameters affecting word order, and must have one of the possible settings for the goveming category parameter of Binding Theory. The answer to the knowledge question is that the language knowledge of adult human minds takes the form of universal principles and variable settings for parameters, whatever the language they have leamt. In addi- tion, knowledge of language contains the lexical items of the language, with detailed specifications of how each item may project on to the structure of the sentence. The full schema is a set of principles and parameters arranged into modules such as Binding Theory; it is a 'compu- tational system' that ranges from the component of Phonological Form to the component of Logical Form, that is, from 'sound' to 'meaning', as described at greater length in Cook (1988c). The principles and parameters framework makes no pretence to cover all aspects of language. The whole complex apparatus is concemed with the crucial central area of syntax defined as core grammar; much of language is peripheral, that is to say, idiosyncratic and linked to UG principles and parameters in a looser way. UG is maximally relevant to the core elements that can be described through its principles and parameters, such as Binding and pro-drop, minimally relevant to other aspects of language, say constructions such as "The more the merrier". As has often been pointed out, for example in Cook (1988c), while this provides a useful delimitation of the areas right- fully considered part of UG, it also provides a convenient way of disposing of elements that do not fit current analyses of UG by pushing them out into an undefined periphery. How then does someone acquire language? To put the UG theory of acquisition in Chomsky's words, 'what we "know innately" are the prin- ciples of the various subsystems of So [the initial state of the child's mind] and the manner of their interaction, and the parameters associated with these principles. What we leam are the values of the parameters and the elements of the periphery (along with the lexicon to which similar consider- ations apply)' (Chomsky, 1986a, p. 150). First the principles. These are not in a sense 'acquired' from outside since they are already present as part of UG inside the mind; instead, they become attached to the person's knowl- edge of a particular language together with values for parameters. This is 202 Universal Grammar Motkl and SLA not to say that a child's speech will show immediate signs of, say, the Projection Principle, as he or she may be unable at first to handle the length or type of sentence which would show it in action - you can't project much on to the sentence if you're only able to say one word at a time. Next, the setting for parameters. Even if you have the pro-drop para- meter present in your mind, you still need to detect which setting is right for it - whether your parents are speaking a pro-drop or non-pro-drop language. The parameter has to be 'triggered' by something in the lan- guage input the child hears. This is seen predominandy as caused by positive evidence - things that are actually present in the input. So the word order parameters for English may be triggered by hearing sentences such as "John ate an apple", the word order parameters for Japanese by hearing "Ojiisan to obaasan ga koya ni sunde imashita" (old man and old woman cottage in lived). The evidence for triggering the setting for the pro-drop parameter is more problematic since pro-drop languages have sentences both with and without overt subjects, and since even non-pro-drop lan- guages have occasional null-subject sentences, "Can't buy me love" in English, for instance. Hyams (1986) originally claimed the crucial evidence to be the presence of stressed subject pronouns and 'expletive' subjects in non-pro-drop languages; once children see that "there" and "it" are necessary as dummy expletive subjects in sentences such as "It's snowing" or "There's a black cloud on the horizon", they can set the English value for the pro-drop parameter to non-pro-drop. As with the principles, there is some dispute over whether all parameters are present in the mind from the outset or whether some come into being at a particular age. The role of language input is to 'trigger' the appropriate setting for each parameter; though input is necessary to the process of acquisition, its nature does not affect its outcome, provided only that an adequate range of natural sentences is encountered. Instead of language being 'leamt', para- meters are 'triggered'. There is no theory of language leaming to supple- ment the principles and parameters contents of the black box apart from the notion of triggering (Saleemi, 1992), with the possible exception of the Subset Principle described in the previous chapter. The model is 'determi- nistic' in that the input simply triggers a setting (Atkinson, 1992). A problem with the deterministic view is that LI children may switch parameter-setting, for example from pro-drop to non-pro-drop in the acquisition of English; why does the input not immediately set off the right setting? Ways round this are to suggest that the child may not be able to take in crucial aspects of the input at a particular stage or that the child may at first not have certain aspects of UG available. Atkinson (1992, p. 210), however, discusses whether the concept of triggering actually differs from the concept of leaming, concluding that it is difficult to find 'a distinctive role for parameter setting as a developmental mechanism'. The UG model sees litde value for the child in negative evidence, Universal Grammar Model 203 consisting either of direct correction by parents - "You mustn't say that, Jimmy" - or of indirect evidence of what doesn't occur in the language - null-subject sentences do not occur in non-pro-drop languages. Direct correction is rarely found in transcripts of actual conversation with chil- dren, and it is not provided by all parents in all cultures; since essentially all human children acquire human language, this rules out direct negative evidence as a main source of language evidence for children, whatever its potential might be. Furthermore, even when correction of children's lan- guage does occur, children seem to ignore it. The role of indirect negative evidence in acquisition is also problematic as it is hard to assess how or what the child notices is missing - in a sense very many things are 'missing' from what the child hears; Saleemi (1988) suggests indirect negative evi- dence is only useful when the child can supplement it with positive evidence. One of the interesting questions for the UG model concems the initial setting for parameters. There might be no initial setting, so a child can adopt any setting with equal ease. Or there might be adefault setting consisting of one or other of the possible settings. Hyams (1986) showed that the early speech of children leaming English has null subjects. Hence, so far as the pro-drop parameter is concemed, the child starts with a particular value, namely pro-drop, rather than having a neutral setting; the setting can remain the same if the child is leaming Spanish but must change if the child is to acquire English. There is, however, some controversy whether Hyams's position accurately reftects the facts of acquisition (Hulk, 1987; Radford, 1990; Cook, 199Oa) and whether children leaming their L1 necessarily start from the pro-drop setting. This distinction between the initial default setting and the leamt setting is sometimes referred to as 'markedness', though this term has many confusing senses: the initial setting is said to be the 'unmarked' setting; the setting that can only be acquired through experience is 'marked'. If the Subset Principle of leaming is true, the unmarked default setting should always define a language that is smaller than the marked setting; as we saw in the last chapter, this applies with difficulty to Hyams' first formulation of the pro-drop para- meter as pro-drop languages are 'larger' than non-pro-drop languages rather than 'smaller'. A further controversy surrounds the issue of whether all the principles and parameters are present in the mind to start with, or whether they come into being over time. In other words, are the principles and parameters like the heart, which is structurally complete at birth, or like the teeth, which grow and are replaced over many years? Borer and Wexler (1987) pro- posed that the UG properties themselves 'grow' in the mind over time. For instance the lack of inftections such as possessive "-s" from children's early speech coupled with the lack of modal auxiliaries like "can", tense forms like "-ed", and other 'functional' parts of the sentence may show that the 204 Universal Grammar Model and SLA IP and CP are not yet present in the child's mind (Radford, 1986; 1990); the reason for the comparatively crude structure of the child's sentences may be that the language faculty in the mind has not yet fully come into being: UG itself develops over time rather than being constant from birth. There is then dis agreement between those who accept a 'growth' model of UG which develops over time and those who accept a 'no-growth' model that remains constant over time, except for differences due to other performance factors. As weIl as knowledge of syntax, the child also has to have a knowledge of the vocabulary of the language. As we saw in Chapter 8, the Projection Principle makes properties of individuallexical items dictate the syntactic environments in which they can occur. Hence the learner has to acquire a vast number of words, each with its own idiosyncratic syntactic behaviour. Within a different syntactic theory, Gross (1990) analysed 12,000 'simple' verbs in French and found that no two of them could be used in exactly the same way. The learner has to learn the syntactic specification of large numbers of idiosyncratic vocabulary items. In some ways, because of the comparative sparseness of the syntax in this theory, the learning load falls increasingly on vocabulary, a trend that reaches its maximum in Chomsky's startling remark, 'there is only one human language, apart from the lexicon, and language acquisition is in essence a matter of determining lexical idiosyncrasies' (Chomsky, 1989, p. 44). Input UG principles parameters LI grammar The substance of the change is that the black box now has definite contents; it produces the grammar of the language through instantiating principles, and setting values for parameters; this necessarily depends on acquiring vocabulary items with their specifications. The LAD box is no longer a set of processes along with a way of evaluating alternative gram- mars but an inventory of principles and parameters. Versions of this redrawn model are often used in the literature, for example Haegeman (1991, p. 15) and White (1989, p. 5). It does nevertheless present certain problems as a metaphorical model, as we shall see later. More detailed critiques of the principles and parameters theory of LI acquisition can be found in Goodluck (1991) and Atkinson (1992). Universal Grammar and Second Language Acquisition In principle the same model could be applied direcdy to L2 learning, as in the following scheme: Universal Grammar Model 205 L1 Input a grammar of LI UG principles parameters L2 Input a grammar of L2 It might be that L2 grammars are not describable in principles and para- meters terms. But, if the L2 user's knowledge of language indeed consists of principles and parameters, the principles are no more leamable in an L2 than in an LI. Ifthe UG argument is correct, no grammar should ever exist in the human mind that breaches principles of UG or that has variation in core grammar not covered by the settings for parameters, except for any qualifications that have to be made for a 'growth' model or for perform- ance reasons. If L2 grammars are indeed independent L2 grammars leamt by 'natural' language acquisition, being human grammars, they should manifest the same principles and choose from the same parameter settings as any other language. Rather than having any conceivable grammatical system, they are constrained by principles - the class of model Keil (1981) calls 'structural change govemed by constraints'. Let us first then see whether L2 grammars may in fact be described in principles and parameters terms. The evidence that L2 grammars obey principles is discussed in White (1989) and Cook (1985b). One principle that has been used as a test-case is structure-dependency. This has often been employed in an acquisition context as an example of a principle that could not be acquired from positive evidence, for example in the Chomsky/ Piaget debate (Chomsky, 1980b); more detailed explanation can be found in Cook (1988c). English allows: "Is Sam the cat that is black?". The movement in this question depends upon the position of the "is" in the hierarchical structure of the sentence; it is the "is" within the relative clause that is moved, not the "is" in the main clause. English, however, forbids: *"Is Sam is the cat that black?" because the structure of the sentence has been ignored; you can't make a question in English by moving any "is", only the one in the right position in the structure. The principle of structure-dependency claims that all human languages rely on structure in this way; they never move arbitrary words or words in the first, second, or nth position in the linear order of the sentence, but always move an element that plays a particular part in the hierarchical sentence structure. While it is seen here as a limitation on movement, structure-dependency is also needed for Binding Theory and other areas of the grammar. The question is whether L2 leamers use 206 Universal Grammar Model and SLA structure-dependency in circumstances when they do not need it in their first language; if they do, this ability must be contributed by their own minds rather than transferred from the L1. Japanese, for example, does not have syntactic movement and so does not require structure- dependency for movement; if Japanese learners of English know structure- dependency, it could not come from their L1. White (1989) cites research by Otsu and Naoi (1986) as providing some evidence for use of structure- dependency by L2 learners. Otsu and Naoi (1986) tested 11 teenage Japanese schoolgirls who had just been taught English relative clauses; they were asked to change sentences with relative clauses into questions. All of them bar one succeeded without breaking structure-dependency. Structure-dependency is perhaps a poor example since it seems in some way a 'macro-principle' of the whole grammar rather than a specific principle of principles and parameters theory; indeed it predates the theory by some years (Chomsky, 1971). Comrie (1990) points out that there are documented languages that occasionally break it, for example Serbo- Croatian, and that it does not apply to phonology, for instance stress- assignment rules. Nor is structure-dependency strictly 'missing' from Japanese, merely not used in movement. Furthermore the need for structure-dependency in the grammar may now be redundant in that such violations would not be permitted by other principles of the theory (Culi- cover, 1991; Freidin, 1991). Research that shows that other principles are present in the L2 is cited by Cook (1985b) and White (1989). Felix (1988), for example, tested several UG principles with German learners of English; results indicated that 'adult L2 learners do have consistent intuitions about grammaticality con- trasts involving principles of UG' (Felix, 1988, p. 285). On the other hand similar research with classroom learners by Felix and Weigl (1991) showed none of the effects associated with UG; 'whatever the basis for their judgments was, it was almost certainly not UG' (Felix and Weigl, 1991, p. 176). It is then the classroom context that is crucial. So, though the evidence that interlanguages do not breach UG is sparse, one should point to the lack of evidence for such breaches; no one appears to have produced clear evidence of sentences from L2 learners that breach principles and parameter settings of UG, say, the Projection principle, or Binding principles, frequent as such sentences should be if they were not universal. Hence the knowledge question for L2 learning still seems answerable in terms of principles: L2 learners' grammars indeed conform to the built-in principles of the mind. But can L2 grammars also be described within the types of variation captured in parameters? Some of the research presented in the last chap- ter, such as Finer and Broselow (1986) and Broselow and Finer (1991), with the settings for the governing category parameter, has indeed expli- citly concerned itself with showing that the L2 learner is working within the Universal Grammar Model 207 bounds of the parameter settings available in UG. The issue that is often debated in this context is how L2 learning relates to the marked and unmarked setting of the parameters. One possibility is that the L2learner starts with the default unmarked setting regardless of the L1; another is that the L2 learner transfers a particular setting from L1 to L2. As we saw in the previous chapter, White (1986) showed that L2learners of English who spoke French, a non-pro-drop language, found null subjects less grammatical in English than did learners who spoke Spanish, a pro-drop language. This suggests that the L1 setting for pro-drop is active in the L2. However, Liceras (1989) showed that English learners did not transfer L1 settings to Spanish. Going from the unmarked to the marked setting seems to present more problems for L2 learners than going in the reverse direction. Second Language Acquisition and the Poverty-of-the-Stimulus Argument The heart of the UG model of acquisition is the argument that, because people demonstrably know aspects of language that they could not have learnt from the speech they have heard, these must already be built in to the human mind, that is, be innate. This is known as the poverty-of-the- stimulus argument, or 'Plato's problem': 'How do we come to have such rich and specific knowledge, or such intricate systems of belief and under- standing, when the evidence available to us is so meagre?' (Chomsky, 1987). The basic form of the argument is that, by comparing what goes in to UG and what comes out, we can establish what is in the middle - the knowledge of language that is necessarily part of the human language faculty. The structure of this argument, as summarised in Cook (1991b), is as folIows: Step A. A native speaker of a particular language knows a particular aspect of syntax. For example, a speaker of English knows Binding Theory. That is to say, the grammar that comes out of the black box contains the binding principles linking pronominals, anaphors, and lexical NPs with their antecedents. Step B. This aspect of syntax could not have been learnt from the language input typically available to children. English children never hear sentences that breach binding principles, nor do they ever produce such sentences so that they could learn by being corrected; indeed, if children actually produced mistakes with binding, adults would not notice them since they would assurne correct binding was intended, as we saw in the previous 208 Universal Grammar Model and SLA chapter. In other words, the input to the black box is not sufficient to show the children that the binding principles are necessary. Step C. This aspect of syntax is not leamt from outside. Consequently this aspect of linguistic knowledge cannot be a response to any environmental factor outside the leamer's mind. If binding principles are present (step A), but are not leamt (step B), they cannot come from outside the child's mind through any form of leaming. Step D. This aspect of syntax is built-in to the mind. As there is no extemal source for this aspect of language, it must be an innate part of the human mind. That is to say, binding principles are part of the black box itself. Once the force of this argument has been accepted, it can be tested on any area of syntax, not just on binding principles. The argument does not depend on any particular syntactic point or type of description; 'if some fact of grammar is unleamable, then it must be related to UG, independent of whether or not we are able to provide an exact and complete UG- account of that fact' (Felix and Weigl, 1991, p. 168). To test the argument for a particular point, it is not necessary to look at the actual stages that children go through in developing language. Instead, the knowledge of any native speaker, as shown by single sentence evidence of possible sentences in the language, can be used to take step A, from which the other steps logically follow. To repeat the discussion of single sentence evidence from Chapter 1, if the single sentence "John loves freesias" is clearly EngIish, and no one would dispute it, then that is all the evidence we need that an English-speaking person knows the structures involved in it. Single- sentence evidence is the basic evidence type used in I-Ianguage linguistics. The strength of the poverty-of-the-stimulus argument partly depends on the types of evidence available to the child under step B. Any aspect of language that the speaker knows must either be leamable from positive evidence, that is to say, through exposure to sentences of the language, or be part of the innate equipment of the human mind. So the syntactic description proposed for it should either be leamable from actual sen- tences encountered or be built in to the mind. Within this conceptual framework, the theory of syntax is integrated with the theory of acqui- sition; all aspects of syntax have to be considered from the perspective of acquisition. While actual syntactic descriptions often make no more than a nod in this direction, this is nevertheless the clear premiss of the principles and parameters model. The UG theory of acquisition is not ultimately dependent on evidence from children's speech; its main prop is single sentence evidence of what native speakers know, not the history of how such knowledge came into being. It therefore implicitly makes a distinction Universal Grammar Model 209 between language acquisition and language development (Cook, 1985a). 'Acquisition' refers to language learning within the poverty-of-the-stimulus argument in its abstract and logical form - how the speaker can in principle acquire a grammar from language data - without looking at actual stages of development or specimens of children's language. 'Development' refers to language learning within a framework of how such knowledge actuaHy develops over time in the learner; its study is based largely on children's performance, whether observational data or experiments. Development is concerned with the chronology of emergence; acquisition pays no attention to time. Development cannot set aside the interaction of language with the psychological and social aspects of the child's development, while acqui- sition can. Let us see how the poverty-of-the-stimulus argument works for second language acquisition; while the steps above are fairly conventional, the interpretation for SLA is more idiosyncratic and is described more fuHy in Cook (1991b). Step A applies in the same way to L2 learning as to L1 acquisition; that is to say, it is necessary to show that the grammar contains some specific aspect of syntax. Showing, say, that L2 learners observe the binding principles, as described in the previous chapter, allows us to take this first step and to describe apart of their grammatical knowledge. A qualification on stage A in SLA, however, is that, while, according to this theory, native speakers aH acquire the same L1 competence, few L2 learners reach a level in the L2 that is equivalent to their knowledge of the L1 and their L2 knowledge varies widely. Stage A in first language acqui- sition describes knowledge common to all speakers, that is, it is a constant from one person to another; Stage A in second language acquisition has to treat this knowledge as varying between L2learners and not having a single fixed form. Step B involves seeing whether this aspect of syntax could be learnt from the evidence available to L2 learners. Unlike L1 acquisition, there are sources other than positive evidence from which such knowledge could come. One available source is transfer of parameter settings from the L1, which is not of course possible for the child acquiring the first language. The work with the pro-drop parameter in Chapter 8 provided an example where the learner seems to start with the L1 setting as a basis for the L2, though this is muddied by the controversy over the normal initial setting; arguably in this case and in others, at the initial stage the positive evidence from outside is balanced by the internal evidence of the knowledge of the L1. But some L2 learners, particularly those in classrooms, also encounter types of evidence unavailable to the L1 child. For example, many teachers frequently correct L2 students, unlike parents in the L1. Teachers also often explain particular grammatical points to students, which provides metalinguistic evidence if not negative evidence proper, although few of their explanations are likely to concern those areas of syntax covered by 210 Universal Grammar Model and SLA principles and parameters theory. In first language acquisition, variation in language input is irrelevant: all children manage to leam human language irrespective of the input they hear. Hence the poverty-of-the-stimulus argument takes into account only the evidence that is available to all LI children, that is to say, positive evidence, and ignores evidence types occasionally available to some children, for example correction. But L2 leaming cannot follow the same route: L2 leamers do not all achieve the same level of competence. Two causes suggest themselves. One is that the prior knowledge of the LI inhibits progress. The other is that L2leaming is more sensitive to the type of evidence available than LI acquisition; successful progress in L2 leaming may depend on the nature of the evidence to a greater extent than in LI acquisition. Steps A and B of the poverty-of-the-stimulus argument both need qualification in SLA research, as there is variation in both knowledge of language and types of evidence available. Hence the modifications to steps A and B mean that steps C and D are not so inevitable in L2 leaming. Indeed, this fact alone may raise some doubts about the relevance of UG to L2learning. Nevertheless, the basic core of the argument still remains: if some aspect of language that is known by L2 learners could not have been learnt from the types of evidence available to them, its only other source must be the innate properties of the human mind. Access to UG in Secood Language Acquisitioo The question that has been most often considered by L2 researchers is whether UG is actually involved in L2 leaming. This has been posed as a choice between three possibilities (Cook, I985a), set out in Figure 9.1. Universal Grammar Other mental faculties j direct j 00 access I I I Figure 9.1 Forms o[ access to UG in L21earning In the no-access position, L2 leamers acquire the L2 grammar without any reference to UG; the grammar is leamt through other faculties of the human mind, and so probably be ars Httle resemblance to usual linguistic competence; it is leamt in the same way as any other aspect of knowledge- Universal Grammar Model 211 cookery, physics, or whatever. In a direct-access position, L2learners learn in exactly the same way as LI learners; they set values for parameters according to the L2 evidence they encounter without any other inftuence. In an indirect-access position, L2 learners have access to UG through what they know of the LI, but they start with the LI parameter settings rather than the initial neutral or default state. Most of the L2 research has treated access to UG as part of a diachronic process of language acquisition rather than as part of the synchronie state of the mind - the sand suspended in the stream rather the sand deposited on the bottom of the lake, to use Weinreich's metaphor (Weinreich, 1953, p. 11). Let us look at some general arguments that have been made against continued access to UG before taking up two syntactic examples that have been used in this debate. (i) Knowledge 0/ L2 is not so complete or so good (Schachter, 1988; Bley-Vroman, 1989). 'Most proficient ESL speakers do not have fully formed determiner systems, aspectual systems, or tag question sys- tems' (Schachter, 1988, p. 224). L2 learners rarely attain the same level of knowledge in their L2 as in the LI. This is claimed to indicate no access to UG. (ii) Allianguages are not equally easy to learn as second languages, as they seem to be as Lls. If UG were available, any L2 would be as easy to learn as any other. According to Schachter (1988), this forms 'an insurmountable barrier to the "back to UG" position'. (iii) Learners get 'fossilised' at some stage, rather than progressing inevita- bly to full native competence (Schachter, 1988). Schachter's example is the overuse of the present indicative tense by English learners of Spanish. As no LI children get stuck in this way, UG is claimed to be unavailable for learning the L2. (iv) L2 learners vary in level 0/ success and in ways o/learning, while LI children do not (Bley-Vroman, 1989). Again, a fully available UG would help all learners equally; therefore it is claimed to be not available. These general arguments are taken to support a no-access, or at best restricted access, position in L2 learning. Bley-Vroman (1989) puts for- ward the Fundamental Difference Hypothesis that L2 learning differs from L1 learning because of changes in the language faculty with age. He proposes that, while the child uses UG and language-specific processes to learn the LI, the adult L2 learner uses the LI and general problem-solving processes; 'the function of the innate domain-specific acquisition system is filled in adults (though indirectly and imperfectly) by this native language knowledge and by a general abstract problem-solving system' (Bley- Vroman, 1989, p. 50). This resembles the division made by Selinker (1972) 212 Universal Grammar Model and SLA between the 'latent language structure' (LAD) and the 'latent psychologi- cal structure', mentioned in Chapter 1. To some extent, these general arguments attack a straw man by overstat- ing the claims of the va model. va is only concerned with the acquisition of highly abstract and complex principles and parameters of core grammar via the poverty-of-the-stimulus argument; it has no brief to explain lan- guage acquisition outside this area by venturing into areas of the periphery, of performance, or of development. It does not stand or fall on areas that are tangentially related to it at best. The lack of completeness (claim i) or of success (claim iv) has nothing to do with the va model unless L2 knowledge is less complete in the core areas covered by va; if specific principles or parameter settings are missing, this would indeed be counter- evidence. But an overall lack of success and of completeness is a far cry from core grammar. va does not encompass all , or even most, of L2 learning by any means; claim (ii) for the comparative difficulty of learning Chinese compared to Italian for an English speaker, true as it may weIl be, is irrelevant unless it is related to the principles and parameters of va rather than to the many other differences between the two; for example Chinese differs from English in being a tone language rather than an intonation languages, and in having a character-based writing system rather than one based on the alphabet. Even if this claim were conceded, it would refute only the direct access position in which all L2 learners start equal, rather than the indirect access position in which they use their LI knowledge. The fossilisation of L2 learning (claim iii) and the variation between learners (claim iv) would be telling if it were expressed in prin- ciples and parameters terms in the core va area. If the va model indeed claimed to be true of all aspects of language learning, these points could be fatal. But it only makes claims for precise syntactic areas, to which much of the evidence for these no-access claims does not relate. 9.2 ACCESS TO ua AND SUBJACENCY IN L2 LEARNERS There are nevertheless areas where more appropriate research has been carried out to test the no-access position. A key area is subjacency, a principle of movement in what is known as Bounding Theory. Vnfortunately, subjacency is highly technical and exists in several versions both historically and simultaneously; the following tries to give an in- terpretation that will allow the reader to see the point of the L2 research, without going into the full technical details. Haegeman (1991) provides an up-to-date account in more detail. Essentially subjacency restricts how far elements may move in the sentence - the bounds within which movement Access 10 UG anti Subjacency 213 can take place, hence the name of Bounding Theory. Elements can move across only one 'bounding node' in the sentence but not more; in English the bounding nodes are IP and NP; the variation between languages over what constitutes abounding node will be dealt with later. This restriction originates from UG. Some languages do not have syntactic movement, such as Korean. If Koreans who leam English show that they know subjacency, this demonstrates they have access to UG, as it could not be transferred from their LI; conversely, if they do not show signs of subja- cency, then they do not have UG available to them. Let us first expand the concept of movement beyond that used so far. In the principles and parameters theory, as we have seen, movement is the relationship between the d-structure and the s-structure of the sentence. Thus the sentence: "What did you see?" has the s-structure: "What did you see t" This contains a trace (t) left in the position from which the item comes. The s-structure is related by movement to the d-structure: "You PAST see what" which gives the basic structure projected by the lexical heads in it. The only reason for having distinct d- and s-structures is so that the relations hip called movement can be accounted for. As always, this terminology does not mean that an element of the sentence literally 'moves' from one place to another to form an s-structure out of a d-structure; movement is a convenient metaphor for talking about declarative structural relationships between levels of structure, not an actual process. Where does the moved element go to? The answer is the empty parts in the structure of the Complement Phrase (CP or C"). In X-bar theory, the CP has the same phrase structure of specifiers and complements as other phrases, namely: CP (C'') I I I specifier c' I I I C complement (IP) 214 Universal Grammar Model antI SLA The head of CP, known as C or COMP, may be a complementiser, such as "that", found in subordinate clauses: "I said that she was here." But CP need not have a head at aß; both the specifier and head positions of the CP can be empty places in the d-structure of the sentence, ready for the elements that are moved to fit into them. In: "What will he ask?" "what" therefore has moved to the empty specifier of CP: 1 1 WhatWiII fie ask t The auxiliary "will" has also moved from its position as head of the IP so that it precedes the subject "he"; it has landed in the empty head position of C, as shown in: In 1 What will he t ask t This question therefore involves two types of movement into the vacant positions in CP; "what" moves into the specifier position, "will" into the head position, seen in the following simplified diagram of the s-structure: CP (C") I I I what j C' I I I C IP I will i t·1 he ask t·J In more complex sentences, movement proceeds in aseries of hops, going from an empty space in one CP to one in the next, as in: 1 11 "Which book did you say t2 that Peter was reading t1 ?" 1 Access 10 UG antI Subjacency 215 "Which book" moves to the front in two steps, using as astaging post the specifier of the embedded ep and leaving behind it a further trace 12. The reason for this staged movement will become apparent below. Movement from one clause up into another is known as 'long' movement. The Subjacency Principle says in effect that an element that is moved from one position in the sentence to another must not be moved across too many 'bounding nodes'. The bounding nodes for English are IP and NP; they do not include ep. In the above sentence, the wh-phrase "which book" moves from its original position 11 in the sentence to 12 (the empty specifier of the embedded ep) crossing the bounding node IP b then from the new position 12 to the specifier of the main ep, crossing the bounding node IP2 , as seen in: 1 [cp Which book did [IP2 11 he say [cp 12 that 1 hp 1 Peter was reading 11] ] ] ]? On each hop "which book" crosses over only a single IP, resulting in a grammatical sentence. Let us switch to the ungrammatical sentence: *"What did he say Joe told her Bill's guess that Ken liked I?" "What" originates in the most embedded clause on the right. Showing only the relevant IP and NP bounding nodes, the s-structure of this becomes: 1 I I What did he say 13 hp Joe told her [NP Bill's guess 12 that Ken liked 11 ] ] 1 "What" moves from the position 11 to the front of the sentence via the stepping stones 12 and 13. The reason for its ungrammaticality is apparent if we count the number of bounding nodes for the hop from 12 to the next possible resting place in a ep 13: "what" has to cross not only the NP [Bill's guess... ] but also the IP [Joe told... ], that is to say, two bounding nodes in one hop. The Subjacency Principle states that 'Movement may not cross more than one bounding node'; an element that is moved can cross only one bounding node (IP or NP) in the sentence. This sentence therefore violates subjacency because the movement of "what" crosses an IP and an NP in one step. This represents the basic analysis of subjacency used in most of the L2 research. Later developments, as we shall see, introduce a parameter over what constitutes abounding node. Each of the pieces of L2 research uses a 216 Universal Grammar Model and SLA slightly different range of structures to which subjacency applies, such as relative clauses, NP complements, embedded questions, and so on; while they will be mentioned briefly, to explore the complexities of these sub- types would go way outside the possibilities of this chapter. The L2 research again concentrates on comparing L2 learners with Lls that have movement with learners with Lls that do not. A language that does not have syntactic movement, such as Korean, will not require subjacency. So subjacency can provide a test of L2learners' access to UG. A Korean who knows the Subjacency Principle in English cannot have transferred this knowledge from the LI since it is no part of Korean, but must have derived it directly from UG; alternatively, a lack of subjacency in Korean learners would be a sign that neither indirect nor direct access was available. A key experiment was performed by Bley-Vroman, Felix, and Ioup (1988), which tested 92 Korean advanced learners ofEnglish and 32 native speakers of English. The learners were selected to be daily users of English and were all advanced, as measured by the TOEFL test (mean score 581), so that they were able to cope with the difficult sentences involved. The technique was a grammaticality judgement task asking the subjects whether each sentence 'sounds possible in English'. Ten sentences violated subjacency, two did not; seven violated Empty Category Principle (ECP) (not covered here), six did not. A typical sentence which violated subjacency was: *"What did Sam believe the claim that Mary had bought?" Its ungrammaticality is due to the long movement of "what" crossing two bounding nodes, NP and IP, without a stopping place: 1 *What did 11 1 hp Sam believe [NP the claim 12 that Mary hp had bought 11?] ] ] A test sentence that respected subjacency was: "What did Bill think that the teacher had said?" Here the movement of "what" crosses only one bounding node at a time (both IPs in this instance): 1 "What did 11 1 hp Bill think 12 that [IP the teacher had said 11?] ]" Access 10 UG anti Subjacency 217 The control sentences had other types of movement that do not break subjacency, such as: "Where is the person that I want you to talk to?" In terms of all sentences tested, native speakers scored 92.2 per cent correct overall compared to the non-natives' 74.9 per cent. The more detailed results are presented in Bley-Vroman et al. (1988) sentence by sentence. A sample of the two grammatical and two ungrammatical com- plex 'factive' NP sentences is as follows: L2learners (%) Natives (%) "What did Bill think that the teacher 54 94 had said?" "What did John realise he could not sell?" 55 97 *"What did Sam believe the claim that 79 100 Carol had bought?" *"What did John hear the news that the 72 97 mayor would do?" ludgements 0/ complex 'factive' NPs (Bley-Vroman et al., 1988, p. 19) The native scores were way above the non-native scores for all four sentences. The L2 learners found the grammatical sentences more difficult than the ungrammatical sentences to a large extent. The overall results for all sentence types similarly show a deficit for L2 learners and an overall superiority for ungrammatical rather than grammatical sentences. Bley-Vroman et al. (1988) also looked at the learners' consistency of judgement of pairs of grammatical and ungrammatical sentences, such as "Who does John want to see?" versus: *"Where did Bill want to know who put the book?" Control sentences were paired with different ungrammatical sentences to get six pairs. Again, results are presented sentence by sentence. Turned into percentage of subjects, not done by Bley-Vroman et al. (1988), the results for subjacency are: 218 Universal Grammar Model and SLA Both Both Both Backwards (good sentences sentences sentences sentence deemed correct deemed bad deemed good bad, and vice versa) (%) (%) (%) (%) Natives 92.3 3.3 3.3 o Non-native 54.6 26.5 12 6.8 Percentage 01 subjects correctly judging six sentence pairs differing in grammaticality (adapted {rom Bley-Vroman et al., 1988, pp. 22-3) The natives were nearly all consistently right (92.3 per cent); half the non- natives were right (54.6 per cent), though 26.5 per cent consistently marked both sentences as wrong. - Similar tables compare pairs of sentences with the same structure that are both wrong, such as: *"What did Sam believe the claim that Mary had bought?" and: *"What did John hear the news that the mayor would do?" Calculating the subjacency results for the two resulting pairs yields: Both bad Both good 1st Bad 2nd Bad (correct) (%) (%) (%) (%) Natives 91 o 1.5 7.5 Non-natives 69 13.5 2.5 14.5 Consistency 01 subjects at sentence pairs (adapted from Bley-Vroman et al., 1988, pp. 22-3) Nearly all the natives were correct (91 per cent); two-thirds of the non- natives were correct (69 per cent); the rest were evenly spread between 'Both Good' (13.5 per cent) and '2nd Bad' (14.5 per cent). Both these sets of figures support the claim that the L2 leamers have access to the Subjacency Principle, but to a lesser degree than native speakers. They conclude that 'slightly over half of the non-native speakers typi- cally exhibit the correct UG-based judgements on any given UG effect' (Bley-Vroman et al., 1988, p. 24), which is way above the chance level they claim of 25 per cent but also far below the level of native speakers. So 'adults appear to have some sort of access to knowledge of UG'; their short-fall compared to the natives is due either to UG being 'attenuated' or to the use of 'a general problem-solving system' (Bley-Vroman et al., 1988, Access to UG and Subjacency 219 p. 27). Evidence for this attenuation is provided by the non-natives' predilection for rejecting grammatical sentences. Research summary: R. W. Bley-Vroman, S. Felix, and G. L. Ioup, (1988) 'The Accessibility of Universal Grammar in Adult Language Leaming', Second Language Research, 4 (1), 1-32. Aim: to investigate access to UG principles by L2leamers Learners: 92 Korean advanced leamers of English, 32 native controls Aspect of Ianguage: the principle of subjacency conceming how far elements may move in the sentence, not utilised in languages like Korean that have no syntactic movement Data type: grammaticality judgements Results: 'slightly over half of the non-native speakers typically exhibit the correct UG-based judgements on any given UG effect' (p. 24) Conclusions: 'adults appear to have some sort of access to knowledge of UG' (p. 27) but not as complete as natives Other experiments have followed the same logic. (a) Schachter (1989) used a similar research design to test knowledge both of structures with subjacency violations, such as: *"What party did for Sam to join shock his parents?" and of equivalent structures without Wh-movement, such as: "That oil prices will rise again this year is nearly certain." If subjacency is available to the leamers, those who know the unmoved structures will also know that the subjacency violations are incorrect; thus they should lai! both sentences if they don't know subjaeeney, pass both if they do. The leamers she tested were 'highly proficient in English' and eonsisted of 20 Chinese, 21 Korean, and 20 Indonesian students of linguis- ties or 'freshman English' in the USA, as weIl as 19 native speaker eontrols. She used a grammatieality judgement test in whieh leamers heard 24 grammatieal sentenees without wh-movement and 24 with equivalent sub- jacency violations. These divided into four syntaetic types: subjeets, rela- tive clauses, NP complements, and embedded questions. A footnote mentions nine additional grammatieal wh-question eontrols and nine ungrammatical eontrols involving rightward movement. 220 Universal Grammar Model and SLA The overall results were as follows, set out here for the relative clause type: N A B C D + Syntax -Syntax + Syntax -Syntax +Subjacency +Subjacency -Subjacency -Subjacency Natives 19 17 1 0 1 Indonesian 20 6 1 9 4 Chinese 20 10 0 6 4 Korean 21 5 0 8 8 Numbers 01 subjects adopting particular solutions (adapted from Schachter, 1989, pp. 83-4) Nearly all the natives (15 out of 19) passed both the declarative sentence sub-test and the relative clause violation sub-test; thus they knew both syntax and subjacency. Only 10 out of 20 of the Chinese passed both sub- tests, 6 out of 20 passing the syntax but not the relative clause sub-test; many of them therefore appeared to know the syntax but not the subja- cency principle. The highest number of Indonesians (9 out of 20) fell into the category of passing the syntax but not the subjacency. Most Koreans either knew the syntax without subjacency (8 out of21) or knew neither (8 out of 21). Thus the UG prediction that knowing the syntax means knowing the prineiple is not eorreet. Turning to the overall results for all four sentence types, most natives performed consistently eorrectly on both syntax and subjacency, namely 56 out of76 (73.6 per cent); their main problem was with sentences containing noun phrase complements, such as that in *"Who did the police have evidence that the mayor murdered?" where only 10 out of 19 (52.6 per cent) were correct. L2 learners were much less successful; those who knew both syntax and subjacency num- bered 76 out of 244 (31.1 per cent), the Koreans' total of 13 out of 63 (20.6 per cent) being particularly low; 111 out of 244 (45.5 per cent) knew the syntax but not the principle. Despite some differences between the Lls in terms of subjacency, there were no significant differences between the three groups. Schachter argues that the high proportion of learners who knew the syntax but not the principle provides 'a major difficulty' for 'those who believe that the principles of UG are available and accessible to post- puberty language learners' (Schachter, 1989, p. 85). (b) White (1985) reports a 'pilot' study testing a parameterised version of subjacency. In recent years it has been commonly accepted that the Access 10 UG antI Subjacency 221 subjaeeney prineiple includes a parameter of variation over wh at eonsti- tutes abounding node. In Freneh one may say: "Combien as-tu vu de personnes?" while the English equivalent is ungrammatieal: *"How many did you see of people?" The reason is that the Infteetion Phrase (IP) does not eonstitute abounding node in Freneh so that "eombien" has to eross only one bounding node to get to the front of the sentenee, as in: 1 "Combien as 1 hp tu vu [NP t de personnes?] ]" But IP does constitute abounding node in English, so "how many" has to cross both an NP and an IP. 1 *How many did [IP you see 1 [NP t of people?] ]" The English bounding nodes are therefore IP and NP, the French are CP and IP. White (1985) tested 73 adult L2 leamers of English, divided into groups from beginners to advanced according to a placement test; 19 spoke French, and 54 spoke Spanish, which has the same setting for subjaeency as French; there was a control group of 11 native speakers of English. A grammatieality judgement test was given on several constructions inviting 'eorrect' or 'incorrect' as possible answers and they were asked to rewrite incorrect sentences; the relevant subjacency sentences consisted of six grammatieal sentences with movement and seven ungrammatical sentenees in which IP was not treated as abounding node, such as: *"How many did you buy of the books?" (In fact some English speakers find such forms grammatical when they include adefinite article: "How many have you got of these old books?") Overall results summed in terms of percentage correct were: 222 Universal Grammar Model and SLA Ungrammatical (% ) Grammatical (%) Natives 89.7 81.8 Non-natives 57.9 81.8 Grammaticality judgements 0/ subjacency (adapted from White, 1985, p.9) Native scores were high on both ungrammatical and grammatical sen- tences, the chief exception being a low of 36 per cent for the apparently grammatical sentence: "What programme did you say that John watched last night?" L2 learners were poor at spotting the ungrammaticality of sentences that used IP as abounding node, being only 57.9 per cent correct. There is improvement with level for the ungrammatical sentences among the L2 learners, showing they gradually acquire IP as abounding node. So 'the results partially support the claim that the LI parameter will be carried over' (White, 1985, p. 12), that is, that the learners are affected by the type of bounding node in their Lls. However, this is blurred by the number of 'indecisive' learners who varied their responses. She concludes 'L2learners cannot approach the L2 as if starting totally from scratch where UG is concerned' (White, 1985, p. 14). White (1988) carried out a similar experi- ment on French learners of English, concluding again that 'the lower performance of Adults (1) [low intermediate] on these sentences [wh- islands] can be attributed to the LI parameter'. (c) Johnson and Newport (1991) argued that a sign of access to UG in L2 learning would be that those aspects of language to which UG applies would be easier to learn than those to which it is irrelevant. They asked 23 Chinese learners of English to carry out a grammaticality judgement task. As Chinese has no syntactic movement, it does not require subjacency. Grammaticality judgements tested three types of stroctme: NP comp- lements, relative clauses, and wh-complements. Each strocture was given in four different forms, each form having twelve test sentences. Taking the NP complements as an example, the fom types were: declarative sentences, as in: "The teacher knew the fact that Janet liked math." subjacency violations, as in: *"What did the teacher know the fact that Janet liked?" Access to UG antI Subjacency 223 control sentences consisting of a similar, but grammatical, question: "What did the teacher know that Janet liked?" no-inversion sentences in which the auxiliary and subject remain uninverted: *"What the teacher did know that Janet liked?" In addition 12 'simple' questions were included. The 'universal' ungram- matical sentences with subjacency violations should utilise UG, the language-specific ungrammatical sentences with no-inversion should not. An active UG should make it easier for the learners to detect the ungram- maticality of sentences that break universal principles than of those that have language-specific errors. Results were as follows, here derived from a graph and expressed as the number correct out of a maximum score of 36 (the total number of sentences for each type): Simple Declarative Control No- Subjacency questions sentences questions inversion violations Natives 36 34 32 35 35 Non-natives 34 31 24 26 22 Native and non-native speakers' judgements 0/ subjacency (adapted from Johnson and Newport, 1991, p. 23) The L2 learners' success with the parallel declarative sentences and simple questions demonstrates that it was fair to test them on subjacency. They were notably below native levels on the subjacency sentences, even if their scores were above chance and they did reject them more than the control sentences. Crucially their scores on the subjacency sentences were as bad as on the no-inversion sentences, rather than better. UG is little help to the learners. Johnson and Newport (1991) used the same test in a second experiment with 21 Chinese learners of English who had arrived in the US at different ages between 4 and 16 and had lived there an average of 9.6 years; these were compared with the adults in the first experiment. The results showed a straightforward relationship with age of arrival: the younger the learners had arrived the better they were at subjacency. Furthermore the no-inversion sentences showed a very similar decline in correctness of judgement for age of arrival. Thus access to UG seemed to be progress- ively cut off for older learners: 'the changes that occur between child- hood and adulthood in language learning seem to affect all aspects of 224 Universal Grammar Model and SLA grammar acquisition, including access to UG... ' (Johnson and Newport, 1991, p. 44). Issues with Subjacency Research In some ways the subjacency research literature demonstrates the interest of the principles and parameters approach to SLA research and its appro- priateness for tackling issues of UG. However, the syntactic theory that underlies it is constantly changing. Tbe first discussion of a predecessor to subjacency calIed the Right Roof Constraint was in Ritchie (1978); the subjacency principle itself was used in Bley-Vroman et al. (1988) and Schachter (1989); parameterised subjacency comes in White (1985) and Johnson and Newport (1991). But the story has not stopped there, bound- ing nodes being, for example, redefined in Chomsky (1986b); an introduc- tion to such later work can be found in Haegeman (1991). Tbe concept of subjacency is now much less important to the theory, being largely sub- sumed under the Empty Category Principle, a Pandora's box we shall not open here. As always, the SLA research lags behind the changes in the theory, which, as we saw in the last chapter, inevitably happens when it tries to use the most up-to-date theory available. Tbe advantage of being in touch with current syntactic theory, as exemplified in the subjacency research, is inseparable from the complexity of the syntactic ideas involved and their refusal to stand still. The methodology of subjacency experiments is entirely based on gram- maticality judgements, a general discussion of which is found later in this chapter. Clearly this limits what can be concluded from the research. Other criticisms can be made of the research design of these studies. Tbe follow- ing list starts with four points made by Johnson and Newport (1991): Bley-Vroman et al. (1988) used too small a set of sentences for subjacency in their experiment (Johnson and Newport, 1991), num- bering ten if all types of violation are included; the design of Schachter (1989) did not test subjects on questions that observed subjacency, only on violations; 'to obtain these results, then, the subjects only needed to respond yes to all of the test items, independent of the structure being tested at the time, and similarly independent of any real knowledge of English syntax' (Johnson and Newport, 1991, p. 11). Schachter (1989, p. 81) does, however, mention that nine grammatical controls were given; 'it would not be possible for a subject to infer that all declaratives were grammatical and all questions ungrammatical'; actual results to support this statement are not provided; Schachter (1989) categorises results by subjects rather than by errors, so there is no indication of how welllearners did (Johnson Access to UG antI Subjacency 225 and Newport, 1991, p. 11). The standard to pass each sub-test seems high at 5 out of 6 (83.3 per cent); the natives in White's experiment, for instance, scored 81.8 per cent on ungrammatical sentences, thus failing this criterion; Schachter (1989) compares sentence type X for the syntax task and sentence type X + movement for the subjacency test, in a paradigm going back to Otsu (1981). But movement places an additional memory load on processing, as we have seen earlier. Sentences with movement are intrinsically more difficult to process regardless of violations and should have been compared with equivalent sen- tences with movement without violation, figures for which are not provided. At the moment it is not clear that the students knew movement itself without the factor of subjacency, in particular 'long' movement from one clause to another; while Johnson and Newport (1991) tested the subjects for long movement, we do not know if the 24/36 subjects who scored such sentences correctly were essentially the same as the 22/36 who spotted subjacency violations - which would imply that 91 per cent of those who knew 'long' movement knew subjacency; in other words the control test should be the control questions, not the simple questions. Crain and Fodor (1984) argue that subjacency violations may be caused by a parsing factor called right attachment (add the item to the structure of the sentence as far down on the right as possible) that confticts with subjacency, even for native speakers; Bley-Vroman et al. (1988) compare various sentence pairs. Of the six pairs on subjacency, some have different structures in the two sentences, for example, "What did Bill think that the teacher had said?" has a Verb complement sentence, while: *"What did Sam believe the claim that Carol had bought?" has an NP complement. Furthermore, the six grammatical members of the pair have an average length of 9.1 words, the ungrammatical members 10 words, perhaps inevitable but nevertheless a possible difference between the two; a crucial factor in the claims of Johnson and Newport (1991) is the difference between the non-Ianguage-specific ungrammatical sen- tences with subjacency violations and the language-specific ungram- matical sentences with no-inversion, such as *"What the teacher did know that Janet liked?" 226 Universal Grammar Model and SLA This sentence is indeed ungrammatical. But why is it ungrammati- cal? It is not without movement since "what" has dearly moved into CP; rather , it is without auxiliary-movement. The 'real' 'uninverted' form of the sentence would be: *"What the teacher knew that Janet liked?" Where does the "did" come from in their sentence? "Do" in English is inserted in an auxiliary-Iess question to carry Tense and Agreement. "Do" has here been given Tense and Agreement as if it were a question, but the final movement into CP has not taken place; in its present position "did" can only be related in some curious fashion to the "did" that carries special emphasis. So the Johnson and Newport sentences are bizarre. Whether such ungram- maticality is language-specific is a moot point; it seems unlikely that any language would insert dummy forms such as "do" unnecessarily into the sentence. 1 cannot remember encountering such sentences naturally from L2leamers, apart perhaps from double past marking ("I did liked hirn"). The major difference between the UG-driven and the language-specific forms is not as dear-cut as this experiment requires. Subjacency had to be inc1uded in this chapter because of its importance to the UG-related SLA research. The criticisms above are more niggling than elsewhere because much of the research and methodology is not dearly described, compared to the other areas we have seen. At face value the subjacency research denies that L2leamers have total access to the UG principle of subjacency and supports the no access position. Looked at more c1osely, the implications are less sure. It is not so much that the principle is unavailable as that it is less available; L2leamers show signs of knowing it to some extent. Bley-Vroman et al. (1988) have L2leamers who are 82.4 per cent successful, White (1985) 57.9 per cent, Johnson and Newport (1991) 61.1 per cent. Moreover, while natives were better, they were far from perfect, except on the Johnson and Newport test; Bley- Vroman at al. (1988) 92.2 per cent, White (1985) 89.7 per cent on ungram- matical sentences, 81.8 per cent on grammatical; on Schachter's test only 73.6 per cent of natives 'passed' overall, down to 52.6 per cent on NP complements. The poverty-of-the-stimulus argument still applies: as there is no other way the leamers could have acquired this knowledge, they must have access to UG, partial as it may be. They possess the principle, even if they show it erratically. The problem is not therefore that access to UG is completely absent but that it is partially blocked. We need to explain why Access to UG antI German Word Order 227 L2 learners do not show this principle of UG fully rather than why they don't possess it at all. The same problem is encountered in first language acquisition: to get round LI children's apparent lack of Binding principle B for pronominals, Grimshaw and Rosen (1990) distinguish knowledge of a principle from obedience to a principle. Most of the L2 research shows knowledge of subjacency but partial obedience to subjacency. What is needed is not so much explanations for lack of access as explanations for lack of obedience. The most likely relate to the performance nature of grammaticality judgements; though not relying primarilyon the speech or comprehension processes of actual conversation, grammaticality judge- ments are none the less aspects of performance. Whatever account of mental parsing one assumes, the comprehension of subjacency involves storing an element in the mind to link to something later in the sentence: it puts a load on memory. The way that subjacency is couched in Berwick and Weinberg (1984), for example, is in terms of counting; knowing that crossing more than one barrier is not permitted means counting how many barriers are involved, hence a storage problem. Presumably the storage system involved also has some effect on the processing of movement in grammaticality judgements. Another performance factor is the sheer length of sentence overloading the memory system; Bley-Vroman et al. (1988, p. 20) point out 'Non-native speakers do seem to reject at a high rate grammatical examples which are long, especially with more than one embedding.' If it is worthwhile persevering with subjacency research given its problematical syntactic status and its methodological difficulties, future research will need to evaluate the relationship between knowledge and obedience more precisely. In my view, attention should be turned to the many contemporary areas of principles and parameters theory that are simpler to research and yield more clear-cut answers. 9.3 ACCESS TO UG AND GERMAN WORD ORDER Other syntactic-based arguments for the no-access position are based on the type of movement used in German. In Chapter 5 we saw that German has a Subject-Object-Verb (SOV) word order in subordinate clauses such as "Ich sage dass ich Englander bin" (I say that I English am), but Verb second order in the main clause resulting in SVO, Adverb-VS, and so on. That is to say, as weil as SVO "Ich bin Englander" (I am English), it is also possible to have OVS "Englander bin ich" (English am I), and Adverb-VS "Jetzt bin ich Englander" (Now am I English), all with the Verb in second position after another constituent. Many syntacticians treat the SOV order as the d-structure and derive the orders found in the main clause by moving the Verb into second position in the s-structure: "Ich bin Englander" has a 228 Universal Grammar Model and SLA d-structure "Ich Englander bin". Translating this into the CP and IP framework introduced earlier, the finite Verb moves to the head of CP (called Inversion in Chapter 5), and elements such as the subject or Adverb move to the specifier of CP (called Adverb Preposing in Chapter 5). This fills up both empty places in the structure of the CP, namely the head and specifier, seen in the structure of the CP of the sentence "Morgen gehe ich weg" (Tomorrow go I away): CP I I I Adverb c' I I Morgen I I C IP gehe ich weg Some elements therefore move twice: an auxiliary such as "muss" first moves to the head of the IP (to get its AGR and TNS), then on to the head ofCP as in "Endlich muss ich gehen." (at last must I go) In subordinate clauses the verb cannot move in this way because the head position in CP is already filled by a complementiser such as "dass". This analysis extends the L2 leaming discussion of the last chapter by bringing in the analysis of functional phrases such as IP and CP, a main growth point in principles and parameters theory. Vainikka and Young- Scholten (1991) for instance postulate an order of L2 acquisition of German in which, simplifying slightly, L2leamers start by knowing only a VP, soon acquire the IP (gradually gaining inflections and subjects), and finally acquire the CP (shown by movement in Wh-questions, and so on). A further example is the discussion of null-subject sentences by Meisel (1990) mentioned in Chapter 8; this linked null subjects with the absence of inflected verbs, that is, with the emergence of an IP. Both these follow the type of analysis introduced by Radford (1986; 1990) in which most of the peculiarities of early child speech are ascribed to one simple cause: IP and CP and other functional phrases are missing from early grammars. All the features which are associated with such phrases, such as the subject NPs, past tense "-ed", articles, and so on, cannot be used. A clear discussion of the implications of this approach to first language acquisition can be found in Atkinson (1992). This type of approach is very intriguing for SLA. In Access 10 UG and German Word Order 229 one way it seems as if the omission of grammatical morphemes described in Chapter 2 may at last get, if not its missing explanation, at least a more coherent syntactic description. In first language acquisition, missing func- tional phrases can be ascribed to some type of maturation: some aspect of the children's ageing renders these functional phrases available to them. But, if the same effect is found in adult Second Language Acquisition, maturation cannot be the cause since the L2 learner is weH past the age of 20 months at which such features are usuaHy learnt in the LI. A consequence of the division into lexical and functional phrases is that the load between principles and parameters is now distributed differently. Rather than parameters belonging to principles, they belong to the heads of phrases, whether lexical items or the abstract heads of functional categories; this was foreshadowed in the discussion of parameterised Binding Theory in the last chapter where parameter values were attached to particular anaphors or pronominals, say "hirnself' or "zibun", rather than to the whole grammar. The approach by Tsimpli and Roussou (1990) treats principles as applying to any acquisition process, whether LI or L2, while parameters, once set in the LI, cannot be reset. This is the general background to their discussion of pro-drop in Chapter 8, where they claimed that apparent resetting in L2 learning was either LI transfer or use of UG principles. However, most work with the CP has been carried out within earlier syntactic frameworks. The key paper bearing on access to UG is Clahsen and Muysken (1986), who compared the learning of German word order by native children and foreign adults using several published studies. Four developmental stages in LI children learning German were derived from earlier studies: I No fixed order; verbs occur both in verb second and in sentence final position, but with preference for the verb final position about 60-70 per cent of the time; for example, "Ich Schaufel haben" (I shovel have). II Non-finite verbs occur regularly in final position, for example, "Deckel drauftun" (cover on-put); finite verbs occur in both verb second and final positions. In other words, the children are starting to restrict the non-finite verb to its correct position, since non-finite verb forms only occur in subordinate clauses. III Finite verbs occur only in second position, for example, "Die Schere hat Julia" (The scissors has Julia), and the finite auxiliary forms and the non-finite main verb are separated in the word order, for example, "Ein Schiff muss du erst jetzt bauen" (A ship must you first now build). The children have acquired Verb movement; Verb second increases from 40 per cent to 90 per cent over about a month. IV Finite verbs occur in sentence final position in subordinate clauses, as 230 Universal Grammar Model anti SLA soon as such clauses first appear, as in "Guck was ich in mein tasche hab" (Look what I in my pocket have). This stage is strikingly error- free and so confirms that children 'make use of learning strategies specijic to the language acquisition device' (Clahsen and Muysken, 1986, p. 103), that is, UG. These four stages are taken to be the natural order produced by UG; from the beginning German children have a tendency towards the underlying SOV order rather than SVO or verb second orders even though children are restricted to main clauses. Clahsen and Muysken's analysis of L2 learning produces very different results. These are based on research with L2learners of German with SVO Lls, namely Italian and Spanish, and with SOV Ll, Turkish, chiefly taken from the ZISA project outlined in Chapter 5. They briefly outline the following six L2 stages, which are basically the same as the Multidimensional Model outlined in Chapter 5; some stages have no examples in their paper, so examples from the Multidimensional Model have been repeated where possible. The stages of the Multidimensional Model will be numbered as in Chapter 5 (pp. 95-6): I Fixed SVO order, for instance, "Ich studieren in Porto" (I study in Porto). There is no Verb Final influence at this stage. This is the same as stage 2 (Canonical Order) in the Multidimensional Model. II Adverbials come before the Subject, that is to say, the order is Adverbial-SVO. In other words the learners have acquired Adverb preposing without Inversion, as in "Da Kinder spielen", Adverb- Subject-Verb (There children play) , equivalent to stage 3 of the Multidimensional Model. III Non-finite forms move to sentence final position. Since the learners have verbs in the middle of the sentence, they can separate the non- finite form and move it to the end rather than the reverse; "Ich habe ein Haus gebaut" (I have a house built), the same as stage 4 (Verb Separation) of the Multidimensional Model. IV The Subject occurs after the Verb in questions and after complemen- tisers. "Dann hat sie wieder die Knoch gebringt" , Adverb-Auxiliary- Subject-Adverb-Object-Verb (Then has she again the bone brought). They have now acquired the Inversion rule - stage 5 of the Multidimensional Model. V Adverbial VP. Adverbials now sometimes occur between finite verbs and the object. VI Finite verbs come at the end in embedded sentences , where they pre- viously occurred in SVO position. "Wenn ich nach Hause gehe, kaufe ich diese Tabac" (When I to horne go... ). Hence, unlike the Ll learners, the L2 learners at first get the word order in subordinate Access 10 UG and German Word Order 231 clauses wrong. This is the same as stage 6 (Verb Final) of the Multidimensional Model. So, while LI children start with SOV and gradually leam the verb second position (SVO) of main clauses, adults start with SVO and gradually leam when the verb is final (SOV). The L1 children deduce the underlying order of German and then leam how to modify it by movement; the L2 leamers cannot. The LI children gradually leam leftward movement from an SOV base, the L2 leamers rightward movement from an SVO base; this poses considerable difficulties for the description of the L2 leamers' grammar within a principles and parameters framework, for various technical reasons, for example there is no obvious CP available to the right for the elements to move into. Clahsen and Muysken (1986, p. 116) claim that 'by fixing on an initial assumption of SVO order, and then elaborating aseries of complicated rules to patch up this hypothesis when confronted with conflict- ing data, the L2leamers are not only creating a rule system which is far more complicated than the native system, but also one which is not definable in linguistic theory'. L2 leamers clearly have no access to UG as they are producing grammars that do not belong to it. Note that this argument depends on sequence of acquisition, not on astate of knowledge as in the subjacency research, or on the poverty-of-the-stimulus argument. Research summary: H. Clahsen, and P. Muysken (1986), 'The Availability of Universal Grammar to Adult and Child Leamers - a Study of the Acquisition of German Word Order', Second Language Research, 2(2), pp. 93-119. Aim: to investigate access to UG in L2leaming through word order in German Learners: chiefty ZISA project leamers, mainly 45 migrant workers in Germany Aspect of language: word order in German Data type: transcripts of corpora Method of analysis: construction of stages for LI and L2 acquisition of German Results: L1 children start with SOV and gradually leam Verb second and SVO; L2 leamers start with SVO and leam SOV Claims: 'L2 learners are not only creating a rule system which is far more complicated than the native system, but also one which is not definable in linguistic theory' (p. 116); UG is then not involved in L2 leaming 232 Universal Grammar Model and SLA A specific refutation of Clahsen and Muysken (1986) was made by DuPlessis, Solin, Travis and White, in the following year (DuPlessis et al., 1987). Their argument has three phases. (i) the syntactic analysis used by Clahsen and Muysken (1986) is inad- equate. A proper analysis of word order in German within the CPIIP framework should involve three parameters: (a) the 'headedness' parameter according to which heads are first or last in the phrase, that is, aversion of the head parameter discussed in the last chapter, (b) the Proper Government parameter whether or not COMP (that is, C, the head of CP) may govern I, and (c) the Adjunction parameter whether or not phrases may be adjoined to the IP at the beginning. (ii) This analysis explains the stages of the L2 learners in terms of setting of the different parameters. Let us lay this out in a table, numbered according to Clahsen and Muysken's L2 stages: Headedness Proper Government Adjunction parameter parameter parameter Stage 1 head initial COMP #- Governor Stage 2 " " yes to fronting Stage 3 head final " " Stage 4 " " ?no to fronting Stage 5 " " " Stage 6 " COMP as governor " Sequence of parameter-setting (according to duPlessis et al., 1987) At stage 1 the learners use SVO because the parameters are wrongly set. At stage 3 they discover the alternative headedness setting, and so have the verb in the correct final position; they do not have the COMP setting right, so the Verb Second analysis applies to both main and embedded clauses. In other words, the learners have switched to an SOV analysis from an SVO one. Around stage 5, their fronting is more limited but still not fully correct; 'our data suggest that they allow both types of fronting and do not reset the Adjunction parameter tilllater, if at all' (duPlessis et al., 1987, p. 68). Finally, at stage 6 they reset the Proper Government parameter so that COMP may be a governor, thus forcing Verb Final in embedded clauses. The order of acquisition for the L2 learners is claimed to be the product of sequential resetting of three parameters. DuPlessis et al. (1987, p. 74) contend 'that the interlanguages of L2learners fall within the range of grammars permit- ted by UG, rather than being unnaturat'. A later paper by Clahsen and Muysken (1989) accepts that the COMP parameter would explain some of the results but regards it as an ad hoc solution; 'the parameter Access 10 UG anti German Word Order 233 appears to be quite a strange thing' (Clahsen and Muysken, 1989, p. 17). They deny there is evidence for an SVO>SOV switch in stage 3. (iii) DuPlessis et al. (1987) present an additional experiment to support their multiple parameters view, based on evidence from student essays. A group of 28 L2 learners of German made few headedness parameter mistakes with word order but were still having problems with Adjunction and with the nature of COMP; hence duPlessis at al. (1987) do not feel that the Clahsen and Muysken sequence of Adjunction before COMP setting is necessarily right. A small group of five advanced L2 learners of Afrikaans similarly had few headedness parameter mistakes but encountered problems with COMP and with Adjunction. Hulk (1991) also looked at acquisition of grammaticality judgements of word order by Dutch learners of French at four different levels. Dutch is a Verb Second language, like German, while French is SVO. Hulk used a similar multi-parameter analysis, but changed the actual parameters slightly. This produced the interesting result that Dutch learners of French do not start with SVO, as Clahsen and Muysken (1986) and the Multidimensional Model insist, but transfer their SOV order to French. A parallel debate can be found over the interpretation of the acquisition of negation in German, for which Clahsen (1988) suggested that 'there was no possible grammatical analysis for the L2 data': Tomaselli and Schwartz (1990) apply an alternative multiple parameter analysis to claim that 'the L2 negative-placement data should be what tip the balance in favour of a UG-based approach'. To sum up, Clahsen and Muysken's claim for no-access is based upon one syntactic interpretation of movement in one language. However true it may be, it is scarcely by itself sufficient to disprove access to UG. Even if their stages are correct, there are other differences between LI children and L2 adults which may explain them. For example, the difference between the two orders is partly that adults get embedded sentences wrong at early stages before getting them right. But children may have been prevented from using subordinate clauses at early stages through perform- ance limitations on sentence complexity that do not apply to adults; if they had used them at the early stages of the adults, they might weIl have got them wrong. Indeed, children may start by being unable to distinguish subordinate from main clauses, and so use the Verb Final forms interchan- geably with the verb second forms; adults, on the other hand, may start by knowing from their LI that subordinate clauses exist and have sufficient processing capacity to use them. The adult SVO starting point may be a handicap compared to the children's initial ftexibility between Verb Final and Verb Second positions. Or it might be that the input to children and L2 learners is different; even if, say, the sentences directly addressed to 234 Universal Grammar Model and SLA children had mostly SVO main clauses, they would still overhear a large range of sentences with other orders; L2 learners, particularly in class- rooms, might hear a restricted range of SVO sentences. Interestingly, Issidorides and Hulstijn (1992) report that the speech addressed to both L1 and L2 learners of Dutch, also a Verb Second language, often has Object-Verb order. The final oddness about Clahsen and Muysken's argu- ment is that adults, who actually get the word order of declarative sen- tences right from the beginning, are claimed to have no access to UG, while LI children, who get the word order wrong, do. The adults are said to be using the canonical order strategy suggested by Slobin and Bever (1982), as described in Chapter 5; the children are not. Indeed, a necessary part of the argument in Clahsen and Muysken (1986) consisted of a refutation of the applicability of canonical order to first language acqui- sition because German children do not follow it, regardless of its virtues in second language acquisition, as discussed in Chapter 5. Alternatives to UG-access Some of those who believe in the no-access position have put forward psychologically-based alternatives to UG access of the type seen in earlier chapters. The general outline is as follows: L1 input UG principles LI grammar parameters Other mental L2 input processes L2grammar UG is unavailable in the L2, its role being taken by other mental processes. These researchers accept UG for LI acquisition but believe that it is bypassed or no longer in existence in L2 learning; the L2 grammar is produced by some other means, probably with some contribution from the LI. Three variants on this are: Bley-Vroman (1989) suggests that the L2 learner employs a combi- nation of 'native language knowledge' and 'general problem-solving systems'. The LI acts as 'a kind of surrogate for Universal Grammar'. The problem-solving system involves other capacities of the mind, similar to models of the mind produced in cognitive science such as Anderson's ACT* (Anderson, 1983), to be discussed in the next chapter. No details or evidence are given for either of these proposals. Access 10 UG anti German Word Order 235 Clahsen and Muysken (1989) claim that UG can function as a learning device in two ways: one is by making principles of language directly available to the LI grammar, indirectly to the L2 grammar via the LI; the other is by having parameters to be available for setting in LI acquisition but not in L2 learning. The differences in L2 learning are that principles are only available via the LI, para- meters are not available at all. L2 learning is 'language acquisition without access to parameter setting' (Clahsen and Muysken, 1989, p. 23). Similar suggestions have been made within the later syntactic theory by Smith and Tsimpli (1991) and Tsimpli and Roussou (1990). Felix (1987) developed the idea of two cognitive systems competing in the learner's mind - the language system (equivalent to UG) and 'the Piagetan-type of problem-solving cognitive structures' (Felix, 1987, p. 154). An L2 mayaIso be learnt via this problem-solving system, in Chomskyan terms a general faculty of the mind other than the specific language faculty. While the LI child has no choice but to use the language system alone, in adult L2 learning there is a competition between these two faculties of the mind, with the problem-solving faculty taking too much responsibility; L2 learning is variable because the problem-solving capacity is subject to indi- vidual variation and situational pressure in a way that the language system is not. Schachter similarly invokes 'a set of cognitive systems (or procedures or inference rules) available for second language learning but not specifically designed for second language learning, systems which have access to first language knowledge as weIl as the ability to abstract out regularities in the linguistic data' (Schachter, 1988). Felix (1991) suggests that the question of access cannot be settled in an aIl-or-none fashion. Some aspects of syntax such as subjacency may still be accessible in an L2; others, such as X-bar theory, may not. In particular he argues that this is related to age. Johnson and Newport (1991), for example, showed that older learners have less access than younger ones to subjacency. Felix and Weigl (1991) argue that access is also related to the input available in particular situations; classroom language teaching may in effect cut the learners off from UG; 'the sad conclusion may be this: you can't really learn a language successfully in the classroom' (Felix and Weigl, 1991, p. 178). As the evidence for the no-access position seems murky, it is not clear that it is worth considering in detail the alternatives that have been put forward to explain it. To repeat the general comment made in Chapter 5, there is no point in considering alternatives from within psychology that are simply a means to demolish linguistic work rather than developing 236 Universal Grammar Model and SLA alternatives in their own right. Showing that expected differences do not occur is testing the null hypothesis: that is to say, it shows negatively that some explanation does not work, but shows nothing positively about what does work. It would be interesting to show that some other pattern of results occurred; it can never be sufficient simply to show that expected differences do not occur. Once some general problem-solving procedure is shown to be capable of accounting for the type of grammatical knowledge that L2 learners have been shown to possess, it may be entertained as an alternative to the UG model. It is not enough to say that Slobin and Bever or Piaget or Anderson can solve the problems of L2 learning without providing concrete evidence how these theories are involved in second language acquisition itself. Indeed, these models are proposed by other people as alternative to UG in first language acquisition for similar reasons to those for which these researchers suggest adopting them for second language acquisition. It seems strange when a psychologically based theory is rejected for Lllearning, but the very same theory is accepted for L2learning; why not consider it for LI acquisition at least? While it may be the case that L2 learning is indeed not based on the same means of learning as LI acquisition, it seems an odd coincidence that the rejected L1 theory turns out to account for L2 learning, rather than some totally different theory. UG-related L2 research is interesting in the context of the concerns of this book because it has seen itself more or less as a subfield of linguistics. It has taken pains to explore the most up-to-date syntactic points to which it can gain access, ranging from pro-drop to subjacency. It has often attempted to feed its condusions back into linguistics, as Flynn and Lust (1990) argue. Sometimes it is undear whether a paper is a contribution to the theory of syntax itself or to L2 learning theory, or indeed if the author actually accepts this difference. While this dose relationship makes for excitement and for constant change, it does not necessarily add to SLA research itself, nor integrate it with other aspects of learning than core UG syntax. Most change seems to come from the development of UG theory rather than from progress in L2learning research, a criticism that is more true of the execution of the research in practice than of its relationship to UG theory in principle. Stages of development are less than crucial to the UG model of LI, since it is concerned with acquisition not development; the intervening stages the learner goes through are only one source of evidence. The same point applies to the L2 sequences of development for pro-drop, subjacency, and so on. The logical question is still whether the final knowledge of the L2 learner contains things that could not have been learnt from evidence - the poverty-of-the-stimulus argument. To quote Felix (1988), 'adult L2 learners are able to attain grammatical knowledge which can neither be learnt on positive evidence nor is generally explicitly taught in the foreign Evidence in UG-related Research 237 language classroom'; without invoking some even more mysterious source for this knowledge, all that remains is UG. 9.4 EVIDENCE IN UG-RELATED RESEARCH AND GRAMMATICALITY JUDGEMENTS As we have seen, there are considerable methodological problems with empirical L2 research related to UG because of the very nature of this theory. The crux of the UG model is the poverty-of-the-stimulus argu- ment, which starts from 'single sentence' evidence of what the speaker knows; it does not need to be justified by secondary evidence of sampies of speech, stages of acquisition, or grammaticality judgement tests, even if these provide helpful confirmation. The poverty-of-the-stimulus argument assurnes that the native speaker's knowledge is static and does not vary from one person to another, to all intents and purposes - the idealised speaker in the homogeneous community (Chomsky, 1965b). However, this is far from true of L2 leaming. Most L2 leamers are in various stages of development, and they differ from each other in many ways; there is no standard, finished, L2 leamer in the same sense that there is a monolingual native speaker; the idealisation to competence is different in the case of L2 leaming. The access to language knowledge afforded by single sentence evidence is not available to support stage A in the poverty-of-the-stimulus argument - the unquestionable knowledge that native speakers share; any individual's notions about an L2 sentence will be temporary and idiosyncratic. In Clahsen and Muysken (1986) the evidence is transcripts of leamers' speech, in Duplessis et al. (1987) student essays; Hilles (1986), Lakshmanan (1991), and Hilles (1991) draw on the same transcripts from the Cancino et al. (1978) study: all are sampies of natural spoken language. The discussion of observational data in Chapter 2 suggests that sampies of L2 leamers' speech

Universal Grammar Model & Second Language Acquisition PDF

Document Details

Tags

Related

Summary

Full Transcript

Upgrade to continue