Vivian Cook   Online Writings

Universal Grammar and the learning and teaching of second languages

Vivian Cook, University of Essex

In T. Odlin (ed.), Perspectives on Pedagogical Grammar, CUP, 1994, 25-48

This chapter has the general aim of describing the possible relationships between Universal Grammar and language teaching.  It first sketches an overview of the principles and parameters theory of syntax and shows how this relates to the Universal Grammar (UG) model of language acquisition; it then describes some of the issues in applying the UG model to second language learning; it concludes by drawing some implications for language teaching. 

Principles and parameters grammar

The Chomskyan UG model of acquisition is based on the theory of syntax known variously as principles and parameters theory or Government/Binding (GB) Theory, named after Chomsky's book 'Lectures on Government and Binding' (Chomsky, 1981a).  The basic concept is that language is knowledge stored in the mind.  This knowledge consists of principles that do not vary from one person to another and parameter settings that vary according to the particular language that the person knows.  Principles and parameters theory can be approached through an analogy to a video-recorder.  A recorder needs two elements in order to function: the unchanging equipment itself, which is the same in every set that is sold, and the variable tuning, which has to be set by the user to local circumstances.  When a new recorder is switched on, everything may be in working order, but nothing appropriate will show on the screen until the channels have been tuned to the local TV stations.  The combination of the two elements of permanent equipment and particular tuning allows the recorder to function in any situation. 

The human mind similarly has built-in language 'principles' that are part of its knowledge of any language.  But it also has 'parameters' within these principles whose values are set to the actual language it learns.  The principles are the permanent equipment in all minds; the parameters tune the principles to a particular language or languages.  A mind that knows English and one that knows French contain the same language principles; the main difference between them is the different settings for the language parameters. 

These principles and parameters are highly abstract and they interact with each other in complex ways.  The following discussion tries to illustrate these concepts without technical apparatus; however, behind these simplified descriptions stands a rigorous syntactic theory, outlined for example in Cook (1988a) and Haegeman (1991), substantially different from theories current before the 1980s, even if it shares certain features with other syntactic theories of the 1980s such as GPSG (Gazdar et al, 1985). 

A much-discussed example of a principle is the principle of structure-dependency (e.g. Chomsky, 1988).  This states an obvious but curious fact: in many languages, the structure of questions depends on the structure of the sentence itself rather than on the sequence of words in it.  The question:
     "Is Sam the cat that is black?"
is linked to a similar structure to that seen in:
     "Sam is the cat that is black"
Forming a question involves knowing which of the two examples of "is" can be moved to the beginning of the sentence to get the grammatical sentence:
     "Is Sam the cat that is black?"
instead of:
     *"Is Sam is the cat that black?"
The speaker of English knows that the "is" in the main clause must be moved rather than the "is" in the relative clause.  The ability to form English questions therefore relies on the speaker's ability to tell the subordinate clause from the main clause.  English questions always depend on knowledge of the structure of the sentence: they are structure-dependent.

Yet there is no real reason why questions should involve a knowledge of structure in this way.  Many other ways of forming questions can be imagined which depend on the sheer sequence of words in the sentence rather than on its hierarchical structure - say reversing the order of words or moving only the second word.  Such alternatives are logically possible and are indeed carried out by computers with ease.  But they do not occur in human languages.  The mind knows that, in order to form a question by movement, it must rely on the phrase structure of the sentence instead of the sheer sequence of words.  This applies, not just to questions, but also to all other constructions in which movement occurs in the sentence, such as passives.  All speakers of English know structure-dependency without having given it a moment's thought; they automatically reject *"Is Sam is the cat that black?" even if they have never encountered its like before.  How do they have this instant response?  They would accept many sentences that they have never previously encountered, so it is not just they have never heard it before.  Nor is structure-dependency transparent from the normal language they have encountered - only by concocting sentences that deliberately breach it can linguists show its very existence.  Structure-dependency is then a principle of language knowledge built-in to the human mind.  It becomes part of any language that is learnt, not just of English.  Principles and parameters theory claims that an important component in the speaker's knowledge of any language such as English is made up of a handful of general language principles such as structure-dependency. 

Let us now look at some parameters.  In English, declarative sentences must have grammatical subjects, such as "he", "it" and "there" in the following sentences:
  "He's going home"
  "It's raining"
  "There's a book on the table"
In Spanish subjects are not needed in the equivalent sentences:
  "Va a casa"
  "Llueve"
  "Hay un libro en la mesa"
This difference is due to the 'pro-drop' parameter.  Some languages, such as Spanish, Italian, Chinese, and Arabic, permit sentences without subjects, and are called 'pro-drop' languages.  Other languages, which include English, French, and German, do not permit sentences without subjects, and are called 'non-pro-drop'.  All languages fall into one or other of these groups.  The pro-drop parameter therefore has two values or 'settings' - pro-drop or non-pro-drop.  Any mind that knows a language has set the pro-drop parameter to one or other of these two values.  A person who knows English knows the same principles and parameters as a person who knows Spanish but has set the value of the pro-drop parameter differently.

Another recently studied parameter distinguishes English from French.  In English, it is possible to say:
  "John often drinks wine"
but not:
  *"John drinks often wine"
However in French the reverse is true in that it is possible to say:
  "Jean boit souvent du vin"
but not;
  *"Jean souvent boit du vin"
In other words, in English the adverb "often" precedes the verb, in French it follows it.  English also permits:
  "John does not drink wine"
where the negative element "not" precedes the main verb "read" rather than:
  *"John drinks not wine"
In French however it is correct to say:
  "Jean ne boit pas du vin"
The negative "pas" follows the main verb instead of preceding it; the other negative element "ne" indeed precedes the verb but is often left out in colloquial speech.  Furthermore English speakers say:
  "The workers all drink wine"
with the 'quantifier' "all" preceding the verb, but French speakers say:
  "Les ouvriers boivent tous du vin"
with "tous" following the verb. 

These consistent differences over the elements that may follow or precede the verb can be accounted for by a further parameter called 'opacity' (Pollock, 1989); in French certain grammatical elements must occur after the verb, in English before it.  A French-speaking person has set the parameter so that these elements must follow the verb; an English-speaking person has set it so that these elements must precede the verb.  They have tuned the parameter in different ways.  The two languages differ in a single overall factor that affects all these constructions - the opacity parameter - rather than in terms of rules about the position of adverbs, negative elements, and quantifiers like "all".

The speaker of course knows many other aspects of language as well as principles and parameters.  Knowledge of vocabulary is specially important to principles and parameters theory.  A person who knows the verb "faint" knows not only its meaning but also how it is used in sentences: "faint" usually has an animate subject in front of it but no grammatical object after it.  So that it is possible to find:
    "Peter fainted"
but not:
    *"The rock fainted"
where the subject is inanimate, or:
    *"Peter fainted Mary"
with a grammatical object.  Knowledge of words is closely tied in to the syntax; the native speaker has learnt how words behave in sentences as well as what they mean.  Many of the complexities of a language are now seen as having more to do with how particular words are used than with syntax.  An extreme version of this position is Chomsky's controversial claim that syntax is innate but vocabulary is learnt:

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). 

Grammar in the principles and parameters theory is concerned with the fundamental aspects of language knowledge - those aspects that are built-in to the mind and that vary within closely definable limits.  This 'core' grammar is what distinguishes human language from animal or computer communication.  Principles and parameters theory makes proposals chiefly about these 'core' areas; it has little or nothing to say about 'peripheral' areas outside their scope.  Hence much of the everyday grammar speakers use and need is beneath its notice. 

This chapter has outlined one principle and two parameters to give some idea of how the theory works.  While there is considerably more to it than this, the complexity arises more from the way in which the various principles and parameters interact with each other than from their sheer number.  The overview of the whole theory given in Cook (1988a) for instance makes use of about seven principles and five parameters.

The powerful type of description made available through principles and parameters theory has shed new light upon many aspects of grammar.  The pro-drop or opacity parameters for instance are intriguing and novel ways of capturing the differences between English and Spanish, and French and English, or indeed many other pairs of languages.  Actual syntactic descriptions in terms of principles and parameters have potential uses in syllabuses for language teaching and in teaching exercises.  The syllabus for teaching French to English people can now include the crucial opacity difference between the two languages; teaching exercises could be devised that unify the teaching of such formerly disparate constructions as the positions of negation, adverbs, and quantifiers; students could be guided to understand such phenomena through 'language awareness' (Hawkins, 1984) and 'sensitization' via the L1 (Riley, 1985), or through 'consciousness-raising' in the L2 (Rutherford, 1987). 

Hence any teaching program that utilises syntax has a new and rich source of ideas to call upon.  Syllabus design has by and large depended upon views of grammar no longer current in linguistics, primarily those of 'structuralist' grammar.  Principles and parameters syntax is only one among the contemporary theories of syntax that are crying out to be applied; others are described in the later chapters of this volume and in Cook (1989).  Syllabuses that fail to take on board current versions of syntax are ignoring information that may be extremely valuable.  It would be a shame if language teaching is cut off from exciting developments in syntax because teachers see them as too difficult or too remote from their interests.  As we have seen, at one level they can provide information about almost every sentence their students may want to say, whether concerning questions, the presence or absence of subjects, or the position of adverbs before or after the verb.

The Universal Grammar model of language acquisition

The principles and parameters theory may be utilised as a theory in its own right for the syntactic insights that it provides.  However its main function is within the Universal Grammar (UG) model of language acquisition.  This claims that principles of language do not need to be learnt as they are already built-in to the mind.  No child needs to learn structure-dependency because he or she already knows it in some sense; it is literally inconceivable for a human mind to know language in a structure-independent way.  The same applies to all the other principles of language. 

Naturally the precise way in which principles apply depends upon the particular language involved.  Japanese for example does not form questions by movement but uses question markers within the sentence.  For instance in the sentence:
    Kimi wa kono hon o yomimashito ka
    you     this book  read (Have you read this book?)
the question marker "ka" at the end of the sentence signals that it is a question.  So the arguments about structure-dependency employed so far do not apply to Japanese questions.  However the critical aspect for UG is that Japanese questions do not break this principle; they simply do not need it.  To disprove structure-dependency would take a language that breaks the principle rather than ignores it.  So far none has been found: all languages meet structure-dependency. 

While the parameters themselves are also built-in to the mind, their values need to be set; the channels need to be tuned.  Parameters are like electric switches that are moved to one position or the other.  A child learning English needs to move the switches to non-pro-drop; a child learning Spanish to move the switch to pro-drop; a child learning French to move the opacity switch one way, a child learning English the other.  Learning comes down to the setting of values for the parameters - to moving the switches.  Learning English means setting all the values for UG parameters to those for English, learning French to those for French, and so on.  This raises several issues:

- what is the initial setting for a parameter?  In other words, what is the position of the switch to start with?  It might be that a child starts from a neutral parameter setting and then adopts one or other of the possibilities - that the switch is initially in the middle instead of one way or the other. 

                                    |----------> setting A (pro-drop)
neutral initial setting -->|
                                    |----------> setting B (non-pro-drop)

A child learning English would start with a neutral setting for pro-drop and change it to non-pro-drop; a child learning Spanish would start from the same neutral setting and change it to pro-drop. 

     Or it might be that the switch starts in one or other of the two positions and has to be reset to the other position when necessary.  In this case the parameter has a default value, called the unmarked setting, that children will retain unless something makes them change it to the non-default value, or marked setting, of the parameter. 

   unmarked setting ----------> marked setting (if necessary)

The unmarked setting therefore is used unless children encounter evidence to the contrary.

To settle this means investigating which is the first setting that children use.  If children have a neutral setting to begin with, they will learn, say, English or Spanish with equal facility.  If one or other of the two settings is an unmarked default setting, all children start with this, but some children, such as those learning English, will have to switch away from it.  Hyams (1986) claims that young English children often produce sentences without subjects, such as "Want more bubbles" or "Now wash my hands", and gradually learn that the subject is compulsory.  They are initially treating English as if it were a pro-drop language like Spanish.  So pro-drop seems to be the unmarked setting from which all children start, non-pro-drop the marked setting. 

pro-drop setting (unmarked) ------> non-pro-drop setting (marked)

English children have to change the setting to non-pro-drop so that in due course they consistently produce sentences with subjects.  Spanish children need to do nothing as the parameter is already set to the right value.  There is, however, some controversy over Hyams' position; Hulk (1987) for example shows that French children do not have a pro-drop stage like the one found by Hyams in English children.

- what changes the value for a parameter?  Children have to hear sentences that tell them which way to set a parameter; in other words they need some language evidence to find out the correct value.  Often this evidence is fairly obvious; the child only needs a few sentences such as:
    "Mummy is phoning Daddy"
to know that English has Subject Verb Object order, a few sentences such as:
    Hanako wa tegami o kaita
    Hanako      letter      wrote (Hanako wrote a letter)
to know that Japanese has Subject Object Verb order. 

Sometimes the vital clue may be more subtle.  Hyams (1986) believes that the crucial evidence for setting the pro-drop parameter consists of sentences with 'dummy' subjects such as "it" and "there" as in:
    "It's snowing"
or:
    "There's a fly in my soup"
Such 'dummy' subjects are the give-away that English is a non-pro-drop language as they are not found in pro-drop languages. 

The types of evidence that the child encounters can be of two types; positive evidence of what actually occurs, and negative evidence of what does not occur (Chomsky, 1981a).  The dummy subject English sentences with "there" and "it" illustrate how the pro-drop parameter is set from positive evidence that the child actually hears.  Negative evidence consists of two sub-types.  One sub-type is correction by people who tell the child what not to do:
    "No we do not say 'is phoning', we say 'he is phoning'." 
The other sub-type is the absence of certain constructions from the input the child encounters; it might be that the fact the child never hears sentences without subjects in English tells him or her that English is non-pro-drop.  To make an analogy, you could learn the rules of snooker simply by watching other people play - positive evidence - or you could learn by actually playing with someone and making mistakes that are corrected by other players - negative evidence.

Linguists have been reluctant to give negative evidence a large role in first language acquisition.  On the one hand parents hardly ever seem to make the right kinds of correction of their offspring, on the other hearing what actually occurs is insufficient to tell them what can not occur, as it does not distinguish between things which they haven't heard because they are impossible and those they haven't heard by sheer chance.  Positive evidence of actual sentences the child hears has to suffice for language acquisition since it is the only type that can be guaranteed for any child in any situation. 

- do the principles and parameters change as the child develops?  So far the child's mind has been described as if it were static and unchanging.  But it might be that the child's mind develops over time; rather than all the principles and parameters being present from the beginning, they come into play at particular stages of development.  The fact that English children initially produce sentences without subjects may not reflect their particular setting for the parameter, but indicate that the parameter itself has yet to develop in their minds.  Various proposals have been put forward for so-called 'growth models' of acquisition in which the UG properties themselves 'grow' in the mind over time (Borer and Wexler, 1987).  For instance the lack of inflections such as possessive "-s" from early children's speech coupled with the lack of modal auxiliaries like "can", tense forms like "-ed", and other 'functional' parts of the sentence (Radford, 1986; 1990) may show that certain types of phrase are not yet present in the children's minds; the reason for the comparatively crude structure of the child's sentences may be because part of the language faculty in the mind has not yet come into play.

Access to UG in second language learning

How can the UG model be related to the learning of languages other than the first?  The usual demurral has to be made that, in L2 learning as in L1 acquisition, UG is only concerned with the central aspects of grammar.  The main interest for L2 learning has been in the role that UG plays in L2 learning.  This can be diagrammed as the three possible relationships given below.

    Universal Grammar                  Other mental faculties
             |....                                        |
             |   :...                                     |
             |      :...  direct                        |  no
             |         :... access                     |access
             |            :....                           |
            \ /               :....                      \ /
         _________                :....       _________
           L1                      :----->          L2  
         grammar                                grammar
         _______-   ----------------->
                         indirect access       

                     Access to UG in L2 learning

In a no-access model L2 learners acquire the L2 grammar without consulting the UG in their minds; the grammar is learnt through other mental faculties.  In a direct-access model L2 learners acquire the L2 in exactly the same way as L1 learners by using UG; they set values for parameters according to the L2 evidence they encounter.  In an indirect-access model L2 learners have access to UG through what they know of the L1 but they start with the parameters in their L1 setting instead of in their original state. 

Controversy still reigns over the choice between these alternatives.  Let us start with some of the general arguments for the no-access position, which claims that L2 learning makes use of other attributes of the mind than UG.
-  while L2 learners show some effects of UG, they do not use it as consistently as L1 natives (Bley-Vroman et al, 1988); hence some other factor than UG must be involved. 
-  the knowledge of L2 learners is not as complete as that of L1 learners and they are not as successful (Schachter, 1988; Bley-Vroman, 1989), so UG is not centrally involved. 
-  children manage to learn any L1 with equal ease; some languages are clearly much more difficult for L2 learners than others, for instance Chinese versus Italian for speakers of English (Schachter, 1988); therefore UG is not available. 
-  L2 learners become 'fossilised' at some stage rather than progressing inevitably to full native competence, hence UG is not involved (Schachter, 1988). 

These arguments against UG involvement in L2 learning are valid insofar as they are based on evidence from areas that are the proper concern of the UG model.  If the learner breaks principles of language or has impossible values for parameters, the UG position is discomfited.  To make these arguments for no-access pertinent, it would have to be shown that these core areas were different in L2 learning.  Yet, by and large, little of the evidence cited by these writers for these general arguments tackles principles and parameters issues.  Claims that L2 learners know less of their L2 than their L1 are certainly true in a general sense, with some demurrals to be made later; but little of the research shows that learners know less of core UG grammar.  If L2 learners knew principles partially, or if L2s differed in difficulty so far as central UG areas are concerned, or if principles and parameters were 'fossilised' in some way, the arguments would have some weight.  So far, however, no clear evidence has been produced that L2 learners do not conform to a principles and parameters system; they are for example no more likely than children to use syntactic movement that does not depend on structure.  L2 learners seem confined by the same UG as L1 learners; their grammars form 'normal' human languages. 

More interesting are arguments that attack the availability of UG on its own ground, that is to say, by dealing with areas of core syntax.  A further principle of grammar called subjacency needs to be introduced to illustrate this point.  Sentences involving wh-words such as "who" and "what" are regarded as being derived from other structures via 'wh-movement'.  So:
      "Who did he say that John liked?"
is based originally on an underlying structure similar to:
      "He said that John liked who?"
But in English it is ungrammatical to say:
     *"The task which I didn't know to whom they would entrust"
although this sentence is derived from an underlying structure similar to that of the grammatical sentence:
     "I didn't know to whom they would entrust the task"
Why does wh-movement not work in this case?  The reason is that the items must not be moved across too many 'barriers' in the sentence; the principle of subjacency says that an item can move across one such barrier but not across more. 

Research with subjacency by Bley-Vroman, Felix, and Ioup (1988) tested whether L2 learners who spoke an L1 that did not have subjacency showed signs of having acquired it in English; if they did, this would show that their UG was still available.  While a group of Korean learners of English indeed turned out to recognise subjacency, they were not so successful as natives.  So UG could be only partially available.  Schachter (1989) performed an experiment with a similar logic on L2 learners of English with Chinese, Korean, and Indonesian as L1s; she gave the learners both a syntax test to see if they knew the structure involved and a subjacency test.  The aim was to see whether those who knew the syntax automatically knew subjacency as well.  There are several problems with the methodology and statistics of this experiment.  However, Schachter argues that UG is unavailable to L2 learners since they were not using the principle for structures they already knew, that is to say those who passed the syntax test did not necessarily pass the subjacency test. 

Both experiments come up against the problem of the psychological processing of a second language.  L2 learners are known to have slower and less efficient cognitive functioning in the L2 than in the L1 in almost every respect, as reviewed for example in Cook (1991b).  Such a 'cognitive deficit' on processing is one possible explanation for subjacency results such those by Bley-Vroman et al (1988); L2 learners are less than perfect because their processing in the L2 is less effective.  Similarly Schachter (1989) compared sentence type X for the syntax task and sentence type X+movement for the subjacency test.  The additional memory load movement places on processing might easily provide a reason for the fall-off rather than subjacency itself.  Several psychological and computer theories of syntactic parsing have indeed been developed to account for the memory load imposed on the comprehension processes by movement (Marcus, 1980; Wanner and Maratsos, 1978).

Another aspect of movement in the sentence has also been put forward as a proof of no-access.  Clahsen and Muysken (1986) compared the learning of German word order by native children and foreign adults using many published studies.  German has a Subject Object Verb (SOV) word order in subordinate clauses such as:
  "Ich sage dass ich dich liebe" (I say that I you love)
but a word order in the main clause in which the verb comes second, that is to say as well as SVO:
   "Ich liebe dich" (I love you)
it is also possible to have OVS:
   "Dich liebe ich" (You love I)
and Adverb VS:
   "Immer liebe ich dich" (Always love I you). 
Many linguists treat the SOV order as the norm, and derive the orders found in the main clause from it by moving the verb into second position.  Clahsen and Muysken (1986) found that L1 children learning German first of all prefer to put verbs at the end of the sentence, although they have some in other positions; gradually they learn to move verbs with tense forms to second position; when subordinate clauses begin to appear in their speech, the verb comes in final position without any mistakes. 

Adult L2 learners of German start with a fixed SVO order, then learn how to put Adverbials at the beginning of the sentence (i.e. Adverbial SV) before moving some particles to the end of the sentence and putting the verb second (i.e. Adverbial V S...).  Finally they learn subordinate clauses with the correct SOV order.  So, while L1 children start with SOV and gradually learn the verb second position (SVO), adults start with SVO and gradually learn when the verb is final (SOV).  This leads Clahsen and Muysken (1986) to suggest that:

'by fixing on an initial assumption of SVO order, and then elaborating a series of complicated rules to patch up this hypothesis when confronted with conflicting data, the 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' (Clahsen and Muysken, 1986, p.116). 

In later work Clahsen and Muysken (1989) suggest that L2 learners are still constrained by the principles of UG in some ways, but cannot set parameters as all the values have already been set for the L1; adult L2 acquisition is 'language acquisition without access to parameter setting'.

This claim for differences between L1 and L2 learners of German is scarcely by itself sufficient to disprove access to UG by all L2 learners for all aspects of syntax.  Even accepting the Clahsen and Muysken stages as correct, other differences between the L1 children and the L2 adults may explain them.  For instance adult learners have a larger memory processing capacity than children; for this reason children may start by not distinguishing subordinate from main clauses, and so use the verb final forms interchangeably with the verb second forms.  Alternatively adults may start by knowing from their L1 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 flexibility between verb final and verb second positions.  The differences apparent between L1 children and L2 adults may not be caused by lack of access to UG by adults but by the fact that adults have already learnt one language and already possess mature cognitive systems; they necessarily have a different starting point for language acquisition, whatever the access to UG may be.

The principles and parameters theory is also changing rapidly in the areas of both subjacency and word order.  For some time it has been claimed that subjacency includes a parameter which permits variation in the definition of the appropriate barriers to movement (Chomsky, 1981b; Rizzi, 1982; Sportiche, 1981).  The ungrammatical English phrase:
    *"The task which I didn't know to whom they would entrust"
is perfectly acceptable in Italian:
    "L'incarico che non sapevo a chi avrebbero affidato"
One barrier to movement that exists in English is not present in Italian.  It is therefore hard to interpret research such as Schachter (1989) and Bley-Vroman et al (1988) as their syntactic analysis allows for no variation between languages.  Lydia White has carried out research with L2 learners with L1s that reflect this difference in parameter setting and has indeed found they have difficulties with subjacency in English (White, 1988; 1989). 

The analysis of word order in German is also far from fixed in the form given by Clahsen and Muysken (1986).  Duplessis et al (1987) provide an alternative analysis, based on current trends in the theory, which bases German word order on the interaction of three parameters; they argue that this accounts for the apparent peculiarity of the L2 learners' developmental order in German and they produce supporting data from an experiment of their own.  The case for no-access can be regarded at best as non-proven, at worst as irrelevant to the actual claims that the UG model is advancing.

UG and L2 acquisition

The choice is therefore between direct and indirect access to UG in L2 learning.  The same questions that were outlined under L1 acquisition will be looked at with reference to L2 learning before developing a specific notion of L2 learning. 

- what are the initial L2 parameter settings?  This question has been the acid test between direct and indirect access models.  If direct access is correct, L2 learners would start with the same values for parameters as L1 children and would set each parameter from scratch.  If indirect access is correct, the starting point for L2 learners is the values of their first languages, which may or may not be the unmarked settings for L1 acquisition.  They would have to reset those that were different in the L2, perhaps back to the unmarked value.

Evidence about initial settings for parameters is reviewed in White (1989).  So far as pro-drop is concerned, French learners of English differ from Spanish learners in judgements about sentences without subjects (White, 1986); the difficulty is that this does not apply to all the effects of the pro-drop parameter, an argument used by Bley-Vroman (1989) as support for the no-access model.  Intermediate French learners accepted ungrammatical English sentences that go against the English setting for subjacency but that conform to the setting for French (White, 1988), thus showing the French setting was still present in their knowledge of English.  Hence such results seem to exclude both the possibility that the initial L2 parameter-setting is neutral and the possibility that there is a uniform initial setting.  On this view L2 learning is unlike the acquisition of the first language because parameters already have a value - the switch is already in one position.  To phrase this in terms of markedness, discussed earlier, this L1 value might be the same as the initial unmarked value (e.g. the pro-drop setting for Spanish children), or it might be the marked value learnt from positive evidence (e.g. the non-pro-drop setting for English children), depending on which L1 the learner speaks. 

There may, however, be a problem of direction; a person going from an L1 that has the unmarked value, say the Spanish pro-drop value, to an L2 that has a marked value, say the English non-pro-drop value, may find it easier than someone going from marked to unmarked.  An analogy can be made to cookery; once you've boiled an egg, it's too late to decide you wanted to have fried egg, as you can't return the egg to its original state.  Revoking an initial parameter setting from marked to unmarked may involve similar problems.  Nevertheless Liceras (1989) found English and French L2 learners had little difficulty with subjectless Spanish sentences, despite the fact they were going from apparently marked to unmarked values of the pro-drop parameter.

- what changes the L2 value for a parameter?  The issue of evidence is rather trickier for L2 learning than for L1 acquisition.  In L1 acquisition the argument that only positive evidence is necessary is based eventually on the fact that all children learn language, but not all children get negative evidence; variation in the type of evidence cannot matter: the researcher looks for the lowest common denominator in the types of evidence available to the child.  In L2 learning, very few learners master the L2 to the same extent that they master their L1.  They also encounter a greater range of situations than the child does; in classroom situations, language itself will be the focus of the situation, which it seldom is in the L1. 

The causes of this apparent deficiency have long been the subject of debate.  One possibility is that the type of evidence is more crucial in L2 learning than in L1 acquisition; classrooms can be considered ways of optimising the evidence available to the learners; the varieties of evidence available to the L2 learner include many possibilities not available in the L1, whether habit-based structure drills, grammatical explanation, or controlled communication games.  All or any of these may give the learner reasons for changing a parameter setting, which are unavailable in L1 learning.  L2 learning research has to go beyond the logical problem of which types of evidence are available to the learner to the empirical question of which type is best for the learner. 

The other possible cause of L2 deficiency is bound up with the concept of marked and unmarked parameter settings.  Let us assume that L1 learners start from unmarked values which they reset, if necessary, from positive evidence.  What happens in the L2 if they have to reset from a marked value to an unmarked value?  According to a principle of learning known as the Subset Principle, often now used to complement the UG model, this is impossible from positive evidence (Wexler & Manzini, 1987).  Once learners have made too 'large' a guess, they cannot jump back to a 'smaller' guess as the positive evidence will never show this is necessary.  If L2 learners are to be able to shift in this way, they will have to encounter evidence of different kinds that may be unnecessary in L1 learning. 

- do the principles and parameters change as the L2 learner learns?  The access to UG might, depend upon the learner's age.  On the one hand the development of the L2 in children might be in step with the development of the L1; accepting a growth model, if a principle or parameter has not become available to them in the first language, it is presumably not available to them in the second.  Or there might be a delay between the UG property manifesting itself in the L1 and the L2.  No evidence seems to be available on this issue. 

On the other hand the choice between the three models of access to UG might depend on the age of the learner.  The Critical Period Hypothesis claimed that, for one reason or another, the ability to learn language in the normal fashion disappears after about the early teens (Lenneberg, 1967).  Again no real evidence on issues relevant to UG is available.  In general the decline of L2 learning ability with age has been less well-supported by researchers in recent years, often now turning into a contrast between short-term gains by adults and long-term gains by children (Long, 1990), though there are certain methodological problems with the research base for this distinction (Cook, 1986).  The account of age effects by Singleton (1989) sums this up in the statement:

'The one interpretation of the evidence which does not appear to run into contradictory data is that in naturalistic situations those whose exposure to a second language begins in childhood in general eventually surpass those whose exposure begins in adulthood, even though the latter usually show some initial advantage over the former.'

Inasmuch as this is related to UG, it suggests that UG is more accessible in younger children.  De Houwer (1990) indeed shows that a small girl in Belgium learnt both English and Dutch as if she were learning two language independently.

Let us assume for present purposes that the L2 learner has finished developing and so possesses the mature adult grammar of the first language.  The difference between growth and non-growth theories does not apply to L2 learning as the learner already possesses UG in its entirety, whether present from birth or developed over time.  The same is true for the distinction between acquisition and development.  The learner's language will not be affected to the same degree by physical or cognitive development, for good or for ill, since this has already taken place. 

Multi-competence

The last section represented a reasonably consensus view of L2 learning.  The present section develops ideas of my own that take the L2 ideas in a particular direction (Cook, 1991; ta).  So far the starting point of the acquisition argument has been taken to be the acquisition of one language - the first; the point of UG is how a mind comes to acquire a grammar of one language in the form of the language principles and the values for parameters.  But L2 learning is predicated on the fact that the mind can learn two grammars, both obeying the same principles but having different settings for parameters, even if the L2 grammars are rarely as 'full' as those of monolinguals.  In terms of statistics, there are probably more minds in the world that contain more than one grammar than there are pure monolinguals.  It has been argued that the choice of the monolingual's knowledge of language as the primary object of linguistic study is a justifiable idealisation:

'We exclude, for example, a speech community of uniform speakers, each of whom speaks a mixture of Russian and French (say, an idealized version of the nineteenth-century Russian aristocracy).  The language of such a speech community would not be "pure" in the relevant sense, because it would not represent a single set of choices among the options permitted by UG but rather would include "contradictory" choices for certain of these options.' (Chomsky, 1986, p.17). 

However, this idealisation may state the problem of acquisition in the wrong terms.  Putting acquisition in terms of how the mind comes to learn a grammar with particular parameter settings implies a fixedness and permanency about the settings: the output of language acquisition is the finished grammar with all its settings fixed.  L2 learning is a matter of resetting these to produce a new grammar.  If the mind is seen as potentially knowing more than one grammar from the outset, each parameter can have two settings; the mind switches from one to the other more or less from moment to moment.  The foundation of the theory has to be this ability to know two settings simultaneously.  Starting from the monolingual person commits the theory to a particular fixed architecture of the mind with the switches permanently set.  The model has to recognise that minds with more than setting for a parameter are the norm instead of the exception.

The state of the mind with two language has been termed 'multi-competence', that is to say 'the compound state of a mind with two grammars' (Cook, 1991a).  The mind of the person who knows two languages should be taken as a whole rather than as equivalent to two minds that know one language each.  The problem of language acquisition is how one mind acquires one or more grammars from input.  This can be seen either as separatist multi-competence in which the two languages are effectively separate or as wholistic multi-competence in which they form a total system at one level or another.  The evidence that the two languages can form a single system in the mind comes from many sources and is documented in Cook (ta): in terms of the lexicon, Beauvillain & Grainger (1987) showed that French/English bilinguals had access to both meanings of a word "coin" (English 'piece of money' versus French 'corner'); in terms of phonological processing, Altenberg & Cairns (1983) found L2 users differed from monolinguals in their assessment of non-words such as "gurch", showing 'both sets of constraints are simultaneously available to the bilingual during processing'.  Hence the argument for the no-access model that L2 learners have different grammars from L1 learners is beside the point; the total system of the L2 user is involved and it is hardly surprising if that part of it dealing with the L2 is different from an L1 grammar known in isolation.

UG and language teaching

The usual qualifications on the scope of UG must be made when discussing language teaching.  UG is concerned with knowledge of language in the human mind.  It has nothing to say about how language is used and little to say about how it is processed.  Language teachers must look elsewhere for ideas about communicative competence, pragmatic competence, or listening and speaking 'skills'.  UG is concerned with core areas of language knowledge expressed as principles and parameters, not with numerous areas of syntax that teachers have to deal with every day.  It is unlikely that any overall teaching methodology could be based on UG, if for this reason alone.

UG is concerned by definition with 'obvious' things about language.  Ideas like structure-dependency are built-in to the mind; they are not mentioned in typical grammar-books for a language, because it can be taken for granted that all readers know them.  They are not learnt, so do not need to be taught.  Parameters too are so broad and 'obvious' that little is needed to set them off.  As Chomsky has pointed out, a single sentence such as "John ate an apple" can set the values for the major word order parameters in English (Chomsky, 1988).  It is not surprising that most principles and parameters are not mentioned in syllabuses for language teaching as there is no real need to state knowledge that is automatically part of the mind, if the no-access model is wrong.

Parameter-setting in a second language may be a less simple matter than in a first, partly because of the already existing settings for the L1.  In the indirect access model the learner may be trying to go from a setting that is not a possible setting in the L1 and, like the old saying, "You can't get there from here", the difficulty of resetting a parameter from some positions may be greater than from others.  The learner may also be more restricted in processing capacity in the second language and so unable to cope adequately with demands of processing the L2.  As we have seen, L1 learning is believed to need only positive evidence, while L2 learning may utilise a variety of other types of evidence. 

One consequence for teaching is that concentrated examples of sentences showing the effects of a particular parameter could be helpful.  In first language acquisition for instance Cromer (1987) showed that exposure to ten sentences with "easy/eager to please" constructions every three months was enough to teach children the difference between these two constructions.  One perspective on language teaching is to see it as provision of concentrated language examples in this way and to deliberately provide evidence appropriate for setting particular parameters; English learners of French for instance might be given a range of sentences using the non-opaque position for negative "pas", adverbial "souvent", and quantifier "tous". 

A further consequence for the language input provided to students for language teaching is the claim by Morgan (1986) that 'bracketting' is necessary for acquisition of some structures.  Suppose a learner does not know whether a language is Subject Object Verb or Object Subject Verb; given sentences such as "The woman the man likes", how does the learner discover that it means that the woman likes the man (SOV) rather than the man likes the woman (OSV)?  Only, according to Morgan, if something in the input 'brackets' the structure of the sentence so that the learner effectively hears "The man [the woman likes]" with the Subject clearly separated from the Verb Phrase by some form of signal, shown here as square brackets.  Morgan demonstrates that certain aspects of syntax may be unlearnable if the input does not have clear clues to its phrase structure.  This may be done phonologically by the intonation pattern or by the length of vowel; the division of the sentence into two tone-groups separating Subject from the rest is one way of exploiting an existing resource in English to achieve this end.  Again L2 teaching could deliberately exploit this process by exaggerating such 'bracketting' intonation or increasing its frequency.  If L2 learning needs different types of evidence, a systematic presentation of the evidence necessary for setting a parameter could be useful, whether in the form of concentrated doses of particular sentences or of amplified bracketting.  Once the step has been taken to admit other forms of evidence in L2 learning, the door is open to facilitating the process via 'super-evidence'.  In a parameter-setting model the chief role of teaching is to provide language evidence that can trigger the setting of parameters in the learners' minds, by whatever means.

The other main implication curiously enough concerns vocabulary.  Current UG theory minimises the acquisition of syntax, maximises the acquisition of vocabulary items with lexical entries for their privileges of occurrence and so on.  Teaching again may be most effective when it builds up this mental dictionary in the student's mind.  L2 learners need to spend comparatively little effort on core grammatical structure of the type covered by UG, since it results from the setting of a handful of parameters.  They do, however, need to acquire an immense amount of detail about how individual words are used in grammatical structures.  This represents vocabulary learning of a particular type, not only learning the dictionary meaning of words or pronunciation, but also learning how they behave in sentences.  It is not just a matter of the L2 learner of English learning the syntax, function, and meaning of "Cats like milk", it is learning that in English the Verb "like" needs to be followed by a grammatical object and preceded by an animate subject.  In other words what can be called a syntactic view of vocabulary.  Some idea of the scale and importance of this task can be found in work by Gross (1990) who found 12,000 'simple' verbs in French of which no two could be used in exactly the same way in sentences.  Vocabulary may indeed be due for a revival in language teaching, not just through the traditional technique of frequency of usage, but through ideas of syntactic specification of lexical items inter alia. 

Above all the UG view is important for language teaching because of its view of language as knowledge in the mind.  Cook (1990) drew some implications for the classroom of the distinction between I-language and E-language approaches to linguistics introduced by Chomsky (1986).  External (E) language is concerned with language as a social reality - with people's relations with each other; the main function of language is communication; children learn language by working out the regularities in a sample of speech and by relating to other people.  In this view, language teaching means providing sufficient data for students to work out regularities, and opportunities for them to relate to each other so that they will learn to communicate with other people.  Internal (I) language is concerned with language as psychological reality in the mind of a single speaker.  It stresses the many functions of language, both inside the mind and out; children acquire language by applying the internal structure of their minds to the speech they hear.  In this view language teaching is enabling the student to construct knowledge in his or her mind from language evidence, which can then be used for any purpose the learner likes.  Recent developments in language teaching are beginning to balance the 'communicative' approaches concerned more with E-language with goals that are I-language in orientation; for example the National Criteria for Modern Language Teaching in the UK include 'To develop an awareness of the nature of language and language learning' and 'To develop the pupil's understanding of themselves and their own culture'. 

The concept of multi-competence also has consequences for teaching.  At the level of goals, it suggests that teaching should not produce ersatz native speakers so much as people who can stand between two languages and interpret one to the other - what Byram (1990) calls 'intercultural communicative competence'.  At the level of syllabus, it points up the fact that it is not enough to describe the target the learner is aiming at in terms of what native speakers do when talking to other native speakers; what is needed instead is the description of how fluent L2 users behave using both their L2 and their L1.  Such a syllabus already exists for Languages for International Purposes designed by the Institute of Linguists (1988).  A beginners task may be reading an L2 travel brochure or listening to L2 answerphone messages to get information that can be used in the L1.  An advanced learner might have to research a topic through reading and conducting interviews in order to write an L1 report.  In this international use, the L2 learner is always the mediator between two cultures, never operating solely in one language or the other, but balanced between the two.

Let us summarise some of the overall implications of the UG model for language teaching.  UG is concerned with the core area of language acquisition; its very centrality means that it can be taken for granted and much of it does not need to be taken into account in language teaching, which has other more pressing concerns.  Nevertheless, at the practical level, the UG model suggests more attention is paid by teachers to the nature of the language input offered the students and to the teaching of specifically syntactic aspects of vocabulary acquisition.  At a more general level, the UG model is a reminder of the cognitive nature of language: L2 learning is the creation of language knowledge in the mind as well as the creation of the ability to interact with other people.  

 References

Altenberg, E.P., & Cairns, H.S. (1983).  The effects of phonotactic constraints on lexical processing in bilingual and monolingual  subjects.  JVLVB 22: 174-188.

Beauvillain, C., & Grainger, J. (1987).  Accessing interlexical homographs: some limitations of a language-selective access.  J.  Mem. & Lang. 26: 658-672.

Bley-Vroman, R.W. 1989.  The logical problem of second language learning'.  In Linguistic Perspectives on Second Language Acquisition, ed. by S. Gass and J. Schachter.  Cambridge: Cambridge University Press: 41-68.

Bley-Vroman, R.W., Felix, S., and Ioup, G.L. 1988.  The accessibility of Universal Grammar in adult language learning.  Second Language Research 4 (1): 1-32.

Borer, H., and Wexler, K. 1987.  The maturation of syntax'.  In Parameter Setting, ed. by T. Roeper and E. Williams.  Dordrecht: Reidel: 123-187.

Byram, M. 1990.  Foreign language teaching and young people's perceptions of other cultures.  ELT Documents 132: 76-87.

Chomsky, N. 1981a.  Lectures on Government and Binding.  Dordrecht: Foris.

Chomsky, N. 1981b.  Principles and parameters in syntactic theory.  In Explanations in Linguistics, ed. by N. Hornstein and D. Lightfoot. Longman: 000.

Chomsky, N. 1986.  Knowledge of Language: Its Nature, Origin and Use. New York, Praeger.

Chomsky, N. 1988.  Language and Problems of Knowledge: The Managua Lectures.  Cambridge, Mass: MIT Press.

Chomsky, N. 1989.  Some notes on economy of derivation and representation.  MIT Working Papers in Linguistics 10: 43-74.

Clahsen, H., and Muysken, P. 1986.  The availability of universal grammar to adult and child learners - a study of the acquisition of German word order.  Second Language Research 2 (2): 93-119.

Clahsen, H., and  Muysken, P. 1989.  The UG paradox in L2 acquisition. Second Language Research 5: 1-29.

Cook, V.J. 1986.  Experimental approaches applied to two areas of second language learning research: age and listening-based teaching methods.  In Experimental Approaches to Second Language Learning, ed. by V.J. Cook.  Oxford: Pergamon

Cook, V.J. 1988.  Chomsky's Universal Grammar: An Introduction. Blackwell.

Cook, V.J. 1989.  The relevance of grammar to the applied linguistics of language teaching.  Trinity College Dublin Occasional Papers 22.

Cook, V.J. 1990.  The I-language Approach and Classroom Observation. English Language Teaching Documents 133: 71-80.

Cook, V.J. 1991a.  The poverty-of-the-stimulus argument and multi-competence.  Second Language Research.   7 (2): 103-117.

Cook, V.J. 1991b.  Second Language Learning and Language Teaching. Arnold.

Cook, V.J. (to appear).  Evidence for multi-competence.  Language Learning

de Houwer, A. (1990).  The Acquisition of Two Long Languages from Birth: a Case Study.  Cambridge: Cambridge University Press.

DuPlessis,J., Solin,D.,Travis,L.,and White,L.1987.UG or not UG,that is the quest­ion:a reply to Clahsen and Muysken.Second Language Research 3(1):56-75.

Cromer, R.F. 1987.  Language growth with experience without feedback. Journal of Psycholinguistic Research 16 (3): 223-231.

Gross M. 1990.  Lexique - Grammaire LADL.  Paper given at the AILA Congress, Thessaloniki.  April.

Haegeman, L. 1991.  Introduction to Government and Binding Theory. Blackwell

Hawkins, E. 1984.  Awareness of Language. Cambridge University Press.

Hulk, A. 1987.  L'acquisition du francais et le parametre pro-drop.  In Etudes de linguistique francaise offertes a Robert de Dardel, ed. by B. Kampers-Manhe & Co Vet.  Amsterdam: Editions Rodopi.

Hyams, N. 1986.  Language Acquisition and the Theory of Parameters. Reidel.

Institute of Linguists 1988.  Examinations in Languages for International Communication.  London: Institute of Linguists.

Lenneberg, E. 1967.  Biological Foundations of Language.  New York: Wiley.

Liceras, J.M. 1989.  On some properties of the "pro-drop" parameter: looking for missing subjects in non-native Spanish.  In Linguistics Perspectives on Second Language Acquisition, ed. by S. Gass and J. Schachter.  Cambridge: Cambridge University Press: 109-133.

Long, M.H. 1990.  Maturational constraints on language development.  Studies in Second Language Acquisition 12 (3): 251-285

Marcus, M.P. 1980.  Theory of Syntactic Recognition for Natural Languages.  Cambridge, Massachusetts: MIT Press.

Morgan, J.L. 1986.  From Simple Input to Complex Grammar. MIT Press.

Pollock, J. 1989.  Verb movement, UG, and the structure of IP. Linguistic Inquiry 20: 365-424.

Radford, A. 1986.  Small children's small clauses.  Bangor Research Papers in Linguistics 1: 1-38.

Radford, A. 1990.  Syntactic Theory and the Acquisition of English Syntax. Blackwell

Riley, P. 1985.  Mud and stars: personal constructs, sensitization, and learning.  In Discourse and Learning, ed. by P. Riley.  Harlow: Longman: 000.

Rizzi, L. 1982.  Issues in Italian Syntax.  Dordrecht: Foris.

Rutherford, W.E. 1987.  Second Language Grammar: Learning and Teaching. Longman.

Schachter, J. 1988.  Second Language Acquisition and its relationship to Universal Grammar.  Applied Linguistics 9 (3): 219-235.

Schachter, J. 1989.  Testing a proposed universal.  In Linguistic Perspectives on Second Language Acquisition, ed. by S. Gass and J. Schachter.  Cambridge: Cambridge University Press: 73-88.

Singleton, D. 1989.  Language Acquisition: The Age Factor.  Clevedon: Multilingual Matters.

Sportiche, D. 1981.  Bounding nodes in French.The Linguistic Review 1: 219-246.

Wanner, E., and Maratsos, M. 1978.   An ATN approach to comprehension.  In Linguistic Theory and Psychological Reality, ed. by M. Halle, J. Bresnan, and G.A. Miller.  Cambridge, Massachusetts: MIT Press: 000.

White, L. 1986.  Implications of parametric variation for adult second language acquisition: an investigation of the pro-drop parameter. In Experimental Approaches to Second Language Acquisition, ed. by V.J. Cook. Pergamon:

White, L. 1988.  Island effects in second language acquisition.  In Linguistic Theory in Second Language Acquisition, ed. by S. Flynn,  and W. O'Neil.  Place: Kluwer Academic: 000.

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