SLA Topics Vivian Cook
Requirements for a multilingual model of language production
Very rough draft April 2001: non-attributable!
This paper sketches a model of language production that accepts that knowing two languages is the normal condition of mankind rather than an unusual exception. It outlines seven requirements that any model of production has to meet. The first six are general in that any model has to meet them and are independent of actual instantiations such as the one developed here. The seventh requirement is more theory dependent and reflect linguistic, more specifically, minimalist, assumptions about principles and parameters, and the priority of the lexicon. For convenience the paper will talk about two languages and use the terms 'L2 user', 'L2 learning' etc; 'two' here simply indicates the plural, i.e. more than one language. It will also use the word 'multilingualism' rather than 'bilingualism' again because it does not tie down the number of languages.
1. The multi-competence requirement: multilingualism is the default case, not monolingualism
The foundation of the approach is the multi-competence assumption that it is normal for human beings to use more than one language. Linguistics has classically taken it for granted that the ideal user of language speaks only one language; ‘We exclude, for example, a speech community of uniform speakers, each of whom speaks a mixture of Russian and French (say, an idealised version of the nineteenth-century Russian aristocracy)’ (Chomsky, 1986, p.17). Chomsky's sole justification seems to be a variation on the theme of the linguistic competence of the idealised native speaker. Idealised language knowledge should be the object of study rather than the muddy waters of bilingualism – chemists study the purest form of water they can get rather than dipping a bucket into the Hudson River (Chomsky, 1986).
Within the past decade this dominance of the monolingual has been challenged. Partly this reflects the sheer numbers of people who speak two languages; statistically monolinguals may be in a minority world-wide; an average human being uses and knows another language. If linguistics is concerned with what the average person knows about language rather than with what the ideal person knows, it has to start by assuming the speaker knows more than one language, to a greater or lesser extent. The monolingual user of one language is the minimal case rather than the norm. It is a matter of whether the idealisation simplifies or distorts; if most people know the grammars of two languages, the model of language has to allow for this possibility from the outset rather than taking the single monolingual grammar as the norm. As Stenson (1990, p.194) puts it, 'Any grammatical theory that purports to account for human linguistic competence must also be able to account for bilingual competence and the associated performance.'
Partly the challenge to monolingualism reflects the growing realisation that people who speak two or more languages are actually different people from those who speak only one. L2 users process language differently and they have different knowledge of both their first and their second languages (Cook, 1997 etc). An L2 user is not just a monolingual with another language added on, a house with an extension added at the back, but a different kind of person, a house with all its internal walls rearranged, neatly summarised in the title of Grosjean (1985): 'The bilingual as a competent but specific speaker-hearer'. Indeed Chomsky's own argument about 'purity' in itself concedes this point by assuming it is unsafe to look at the L1 of bilinguals as it may have been contaminated by the L2.
Models of speech production therefore need to recognise that the normal average person uses more than one language. This needs to be built in to the system right from the beginning rather than adding on features to a monolingual model. As de Bot (1992, p.2) puts it, 'one could argue that the basic model should be concerned with bilingualism, with an option to have a unilingual version'. A frequent route in SLA research is to take Levelt's 1989 model as the starting point and then to see what needs to be done to include another language, as de Bot (1992) does himself. A proper approach is to redesign Levelt's original model for monolinguals as well as bilinguals, since even the monolingual potentially can acquire another language. The L2 model is not a refinement of an L1 model; the L1 model is a special minimal case of the multilingual model in which in principle anybody can use more than one language, and in practice often does.
The first requirement is therefore that multilingualism should be the default case for production, not monolingualism; any model of speech production must potentially handle the ordinary human condition of using more than one language. De Bot (1992) suggested that a bilingual model should 'be able to cope with a potentially unlimited number of languages'; we argue here that this must be true of any model. The problem is how this multilingual system can be restricted to a single language to deal with the monolingual rather than how a monolingual system can be expanded to deal with multilinguals. Elements of models such as Levelt can be used if they would apply regardless of how many languages were being processing. But everything that is present in the multilingual model must be potentially present in the restricted monolingual model.
2. The multi-competence cognitive requirement: speakers of different languages have different cognitions
Most models of language production see the process as starting from the speaker's cognitive knowledge — the set of concepts which underpins everything they say. We can call this process by Levelt's term 'conceptualisation': the speaker has a speech impulse that causes the activation of concepts and their incorporation into a conceptual message. We will assume that a separate aspect of the mind called 'cognition' stores such concepts. For convenience this will be divided into cognitive aspects and pragmatic/social aspects, partly as these are claimed to be stored primarily in different hemispheres of the brain (Code, 1987; Paradis, 2000). The questions for a multilingual model of language production to ask are whether cognition is in fact separate from language and whether cognition varies from speakers of one language to those of another. Inevitably this brings up the vexed problem of linguistic relativity: to what extent is the way we see the world shaped by the language we speak?
Anna Wierzbicka' list of semantic primitives common to all human beings includes substantives such as I, and YOU, determiners such as this and THE OTHER and Time concepts such as WHEN, BEFORE and AFTER. These are part of the speakers' conceptual knowledge, whichever language they speak. The universality of the concept YOU is presumably grounded in the human ability to recognise participants in a scene. Yet languages such as French use the well-known distinction between singular T pronouns to show social closeness and plural V pronouns to show social distance; languages such as English do not. French mountain climbers are said to change from vous to tu above a certain altitude when the common danger brings them closer together; English climbers have no equivalent way of showing their comradeship. A language-specific system for social recognition is grafted on to the universal second person pronoun. Hence even in terms of semantic primitives there are different instantiations in different languages.
The problem with establishing the difference between universal and language-specific is the chicken-and-egg conundrum of the Whorf-Sapir hypothesis. We all possess a universal perception of saltiness in the things that we taste. O’Mahoney and Muhiudeen (1977) tested the ability of speakers of Bahasa Malaysia and English to distinguish solutions with different amounts of salt. Malaysians were able to make much finer distinctions than English speakers. Bahasa recognises far more degrees of saltiness than English. So has the speaker's conceptual world of taste been influenced by their language? Or has their language been adapted to suit the flavours of their cuisine? Lucy (199X) has shown that speakers of languages such as Yucatec Maya are dominated by the feel of things; their language emphasises tactile properties through a complex set of XXX that are interpreted by Western linguists as a determiner system rather than as a crucial conceptual system. Speakers of XXX react tactilely in all sorts of ways; take them to a restaurant and they will feel the texture of the table-cloth and the table legs. Their way of thinking is complemented by the structures of their language. Whether we see their language as influenced by their thought or vice versa, nevertheless speakers of different languages have different concepts. Indeed Levelt recognises this in his discussion of X and Y.
A person who knows a language then possesses one set of concepts available to all human beings through shared experience or biology, another set built up of specific cultural and learning experiences, hard as it may be to separate the two. Language may restrict the pool of concepts potentially available to all human beings; language may enable these primitive concepts to be developed, refined and extended in language-specific ways. Their origin is not relevant at the moment; the point is, not that language influences thought or that thought influences language, but that speakers have different thoughts according to the language they speak. Levelt calls the product of conceptualisation the 'pre-verbal message' (Levelt, 1989) and 'lexical concept' (Levelt et al, 1999); these labels already incorporate language elements and so undermine the generality of the models by making the conceptualisation stage specific to a single language. Hence we will use the term 'conceptual message', recognising that concepts may or may not be language-specific.
So the mental world of monolinguals contains universal concepts and concepts linked to their own language. This makes it difficult for them to see the world in a different fashion because they are not aware which aspects of their thinking are universal, which specific to a particular language. I see an arm and a hand; a Greek speaker sees a single object heri. Monolinguals cannot tell which parts of their thinking are based on their culture, which are part of everybody's; it would not have occurred to me that that arms and hands were not separate objects to all human beings. People used to a sound-based writing system cannot conceive that a meaning-based character system is just as natural for its users. Even books by linguists often assert the superiority of their own writing system. Our concept of writing is dependent on the type of writing system that we have been exposed to first.
A second language may then have different language-specific concepts from a first language. Going from English to Japanese means incorporating a social concept of respect. The use of Japanese pronouns depends upon the speaker and listener appreciating their relative social status. The first person singular ‘I’, watakushi is used to a social superior, watashi is more neutral, boku is for men under 21, ore is used to close friends, and chin may be used only by the Emperor.
L2 users, however, have acquired a second set of language-specific cognitions, to a greater or lesser extent, potentially contrasting with their L1 set. The L2 user potentially has three types of element in the mind: universal elements, L1-specific elements, and L2-specific elements, related in some complex way, say the universal concept of YOU, the English concept of first, second and third persons in pronouns, and the Japanese system of respect. One question is the extent to which these cognitions become an overall single set or remain in separate compartments labelled L1 and L2. About the only relevant L2 research on this is Caskey-Sirmons and Hickson (1977), who found that monolingual speakers of Korean use the Korean word for 'blue' (paran sekj) to mean something greener and less purple than Koreans who know English. Their colour concepts form a single system in which there is no L1 value, no L2 value, but a combined value.
So far as social perceptions are concerned, a more radical possibility can be developed from the theory of cognition put forward by Michael Tomasello (2000). This treats language as arising from the uniquely human ability to see the point of view of others, already seen in babies' abilities to follow people's gaze. The basis of language is social cognition of others. But a monolingual has in a sense a single other to model: all the people whom he or she encounters use the same world-view and the same language. A person who speaks multiple languages has at least dual others – a stereoscopic vision of the world from two or more perspectives. As we know from many studies of young children, this indeed has an effect on their cognitive development, enabling them to be more flexible in their thinking (XXX), learn reading more easily (XXX) and (Bialystok). The point is not so much that 2+2=4 as 2+2=5; it is not just the addition of a different world perspective to the person's scope as the realisation that many perspectives are possible. Monolinguals therefore are not only restricted to a single world-view, but they also have more idea that other world-views are possible. Indeed this has always been seen as one of the main educational advantages of language teaching, for example XXX
The second requirement is then that speakers of different languages have different cognitions; the model must accommodate the differences between language-independent and language-specific cognitions. The conceptualisation stage of the production process is not independent of language but already incorporate some aspects of language, as indeed occurs in Levelt (1989). The L2 user potentially has two sets of language-specific cognitions. Taken in conjunction with the first requirement, this means that any model of production must allow for language-independent cognitions, language-specific specific cognitions in potentially more than one language, and the relationships or conflicts between them, a possibility not allowed for in contemporary models of production. To start with monolingual models, in Levelt et al (1999) production starts from 'conceptual preparation in terms of lexical concepts'; concepts are specific to the lexicon of a single language (Levelt, 1989, 1030); the model does not allow at this stage for speakers of more than one language. The bilingual models make no such concession either; Roelofs (XXXX) sees conceptualisation as mapping communicative intention onto a message, with part of the illocutionary intention being the choice of language; Green (1998, 71) proposes 'a conceptualiser that is independent of language'. de Bot (1993, 194) assumes that 'the conceptual structure is not language-specific' and utilises Levelt's distinction between two levels of the conceptualiser – macro-planning (elaboration of communicative goals/ intentions) and micro-planning ('selecting the information whose expression may realise the communicative goals' (Levelt 89, p.5) – to argue that the second stage of micro-planning involves a choice between languages. The problem, as we shall see below, is that this choice is not made once and for all but in code-switching mode from word to word or phrase to phrase.
We can represent this requirement in the diagram of conceptualisation in Fig 1. This shows the speech impulse leading to a conceptual message built up from social and cognitive concepts both language-independent and language-specific for at least one language A, or two languages A and B, or indeed as many languages as the person knows. We will assume that there is no language choice at this stage; some conceptual messages may then get blocked because they can be expressed with the greatest of difficulty in the language that the person wants to communicate in.
3. The code-switching requirement; the model must allow the speaker to choose one language or the other at every stage of production.
L2 users can opt to use one language or another, ranging from long-term decisions about which language to use in bringing up their children to moment-to-moment decisions about which language to use for a particular word. Given that multi-competence is the norm, this choice has to be built-in to the basic model of speech production. Indeed it has been argued that the choice of language for the bilingual is the same choice as that between styles and registers for the monolingual (Paradis, XXX), even if this may seem a pale shadow. Lambert (1990, p.204) posed the problem 'How is it that the bilingual is able to 'gate out' or set aside a whole integrated linguistic system while functioning with a second one, and a moment later, if the situation calls for it, switch the process, activating the previously inactive system and setting aside the previously active one?' Language choice might be a matter of turning one language one on or of turning one language off (Green, XXX).
In Grosjean's monolingual mode of language, the L2 user indeed opts for one language or the other to speakers of the appropriate language; in the bilingual mode, however, L2 users can choose to have both languages on line at essentially the same time in circumstances where they know the listener also commands both languages. Code-switching may take place at almost any level in the sentence between utterances, phrases or words, according to complex sociolinguistic, structural and psycholinguistic rules. In the bilingual mode the choice of language is not made once and for all but modified from moment to moment. As well as an overall decision about language, there are local decisions about language at many different decision points during the sentence. To account for code-switching a multilingual model of speech production cannot then allow just a single decision point, as say in de Bot's point between the macro-planner and the micro-planner, but must allow a series of decision points at every stage of production. These decisions may indeed be related; only an early decision to use the bilingual mode permits later switches between language. Even within the bilingual mode, there may be a overall decision to prefer to use one language as the main syntactic skeleton of the sentence from which later decisions cascade, as in the Matrix Model (XXXX).
However it may be actually incorporated, the model must reflect the overall possibility of choosing or gating out one language rather than the other, which we can call a macro language decider. This includes not only the choice of which language but also the choice of monolingual or bilingual mode within a language; as well as choosing language A or B the user decides whether to stay solely within that language or use it as the matrix language in code-switching. The reasons for this choice depend on the speaker's evaluation of such factors in the situation as the languages commanded by the listener. But there are also local decisions about language in what cam be called the micro language decider. This covers the choice of language at different levels in the sentence, again dependent on the speaker's evaluation of the situation through the various rules of code-switching. There might in a sense be a different micro-decider at each level. We see below, however, that a single micro-decider might form part of a theory of lexical selection based on aspects of the Minimalist Program.
4. the developmental requirement; the model must be usable at all stages of acquisition.
Barring the small percentage with language disabilities, L1 users are taken to arrive at more or less the same knowledge of their native language. Linguistic theory takes as its cornerstone the linguistic competence of the ideal native speaker in the homogenous community (Chomsky, 1965). This has always been paradoxical in that linguists are concerned with the 'language' of an 'average' neutral native speaker, and with the 'grammar' in a single individual mind; 'The grammar in a person's mind/brain is real ... The language (whatever that may be) is not' (Chomsky, 1982, p.5). Linguistics is not only concerned with the linguistic competence of the adult but also with the linguistic competence of a person at any stage; the two-year-old has a competence as appropriate to their stage as the twenty-year-old; hence the attempt in the 1960s to write grammars for children's grammars (REF). One of the perpetual disputes in linguistics is whether interim competences have the same nature as final competence. By definition linguistic competence is whatever the person knows, as true of the two-year-old as of the sixty-year-old. Whether the form of competence itself changes is then a moot point, bringing up the perennial divide between maturational and non-maturational. Given the insight that there is no single final state of L2 competence, we are then committed to a view that has to accommodate many levels of knowledge within the same model.
Most SLA research claims that final linguistic competence is rarely if ever achieved in an L2. L2 users vary in their final states, L1 users do not; L1 learners are successful, L2 learners are not: 'Very few L2 learners appear to be fully successful in the way that native speakers are’ (Towell & Hawkins, 1994, p.14). This lack of success is based on the assumption that the only acceptable final state is that of the monolingual native speaker: as put explicitly by Felix (1987, p.140), ‘children generally achieve full competence (in any language they are exposed to) whereas adults usually fail to become native speakers’. The assumption underlying this paper is, however, the multi-competence view that L2 users are L2 users in their own right, not failures at achieving monolingual L1 competence (Cook, XXX); their success should not be measured against a group of which they can never be members by definition.
There are therefore many levels of L2 competence, ranging from the beginner taking their first steps in a new language to the balanced bilingual at home in both languages. The multilingual production model has to be capable of accommodating all levels and all final states of language, not just the single state of monolingual native competence. Even if production processes change radically during the stages of acquiring, the possibility of different states of knowledge needs to be allowed for. Klein and Perdue (XXXX) for example have suggested a common basic grammar for L2 users with diverse L1s and L2s, using inter alia semantic processing strategies; the Processability Model (Pienemann, 1998) reveals universal stages of syntactic acquisition based on processing requirements. In terms of the lexicon, not only are the sheer numbers of words different for different stages of acquisition but also the completeness of the information for any specific word EG The multilingual production model must handle all of these as well as the highly untypical balanced bilingual with equal command of two languages. When the arbitrary tie to the monolingual native speaker is broken, the model has to be flexible enough to allow for the differing levels of L2 users, particularly when these seem to be shared by many users; there is no single L2 final state for production as there arguably is for monolinguals. The single final state is perhaps then an illusion created by the over-concentration on monolinguals in linguistics.
This developmental requirement insists that the model work for different stages of acquisition. It could also be termed a variable final state requirement if it is conceived solely in terms of the level that L2 users eventually achieve; any of these developmental stages might be a final state so far as a particular learner is concerned. The requirement can be built-in to the model in various ways; the proposal below relates it to a lexically-driven minimalist model. The unavoidable consequence of treating production as inherently multilingual is to see it as being used by people at all levels of language, from children to adults in the L1 and beginners to advanced users in the L2.
5. The orthographic requirement; the model must allow for the effects of literacy and for written production in literate language users.
This requirement reflects on the one hand changes in the whole production system due to the possession of literacy, on the other to the processes used specifically for writing language. many L2 users are not literate but also write in both languages. Due to the bias in linguistics towards the spoken language, production models have typically confined themselves to speech. Yet this ignores crucial aspects of typical speakers. A literate person is no longer the same as a non-literate person, just as a literate society differs from a non-literate one (REF). The ability to read has influenced the speaker's knowledge of phonology. It is well-established that peoples' estimates of the number of sounds in a word is influenced by their knowledge of the number of letters (XXX), for example ridge being believed to have more sounds than rage. It has been claimed that the linguistics concept of the word is an artefact of the invention of word spaces in sound-based scripts (Saenger, 1997) and the concept of the phoneme the projection of an alphabetic writing system onto speech (XXX).
Using the same type of argument as for requirement 1, the average L2 user is probably literate. Hence they have a production system that has been influenced by the ideas of writing. Their knowledge of language is not the same as it was before they learnt to read and write. While it may be useful to keep the default for the multilingual language production model as spoken language, the representation of knowledge in the mind has to allow for the changes that literacy brings, in particular to the representation of lexical entries. This requirement is made here simply to make certain that the alternative form of production through writing is allowed for in the model
6. the compensatory strategies requirement; the model must allow speakers to alter their production when it is not succeeding by utilising alternative strategies
With this requirement we start moving into the area of 'lexical selection' in the terms of Levelt (1999) or the 'formulator' (Levelt, 1989). After the conceptual message has been constructed, it has to be fleshed out with vocabulary and with grammatical structure. We will recognise here a stage that can be called the Lexical Item Selector. The concepts in the conceptual message need then to be matched against the lexicon in the speaker's mind.
This requirement disposes of the case when the matching does not work. It is common enough in a second language to feel that one is groping for a word: a Dutch looking for the verb iron creates ironize (Poulisse, 1990); a German student trying to say pottery produces when you make a container. Poulisse (XXX) has shown, however, that this is not unique to the L2. In situations where we do not know or can't remember the L1 vocabulary to express our message we resort to the same types of strategy in the L1: Where's that pump-thing for spreading filler?. Extreme forms can be seen in some aphasias: it was the where you make all the food, you make it all up today the keep it ’till the next day with the French, you know (Gardner, 1974, ??). Poulisse (1990; 1993) has linked this to the Levelt model.
It is then necessary for the model to provide the safety valve of compensatory strategies. While in a sense this is necessary at all stages, the model has to make it explicitly available for failures in lexical selection. It is particularly necessary in the multilingual model since the link between the conceptual message and lexicon may break-down not just in the sense that the learner cannot find the right words to express the concept but may not find the right words in the right language selected by the Macro Decider. Hence the conceptual compensatory strategy feeds back the failure into the conceptual message which is rejigged so that a concept can be found that can be expressed.
7. the lexical priority requirement; lexical entries project on to different levels of syntax
The underlying assumptions behind this requirement are that an adequate notion of linguistic structure has to be incorporated into a model of production, not ignored apart from a box labelled 'morphological encoding' (Levelt, 1999) or 'grammatical encoding' (Levelt, 1989). Inter alia this means that the model cannot be concerned solely with the production of words; the issues about more than language only start to emerge when we see several words combined within a lexical structure.
This lexical priority requirement bundles together several issues concerning the integration of grammatical structure with the lexicon. This section tries to illustrate how different properties can be handled within the framework of the Minimalist Programme (MP) (Chomsky, 1995). They can nevertheless be phrased in terms of other theories.
The starting point is the lexical entry, which represents all the information about a particular vocabulary item, ranging from argument structure to spelling. The MP in effect extends the syntactic elements in the lexical entry beyond argument structures to include inflectional as well as derivational morphology. A speaker knows a word dog with a lemma containing not only meaning components, [+canine, +animal], basic level and grammatical information, Noun, Countable, ... but also a specification of its actual form showing its plural form +/z/<s>, as well as the actual formal representations /‰ / and <dog>. The MP is lexically-driven in that the lexical entries project their meaning and syntactic features on to the sentence; they prescribe how the lexical entries may or may not be used in sentences.
We need to introduce two of the main hypotheses of the MP, presented in more detail in Cook (1995).
A. the lexical parameterisation hypothesis; parameters are part of the lexicon.
Since 1981 or so, Chomskyan theory has dealt with variation between languages in terms of a choice of settings for a parameter. The parameter-setting forms part of the lexical entry for that particular word and so shows what happens when the word is put into the syntax. The entry for a reflexive such as himself or zibun sets the parameter value for the language involved whether English or Japanese and thus permits the reflexive to be used in different types of structure in the sentence. The parameters that account for much of language variation form part of particular lexical entries, not of the syntax itself.
B. the functional parameterisation hypothesis: parameters are attached to functional phrases in the lexicon
Since 1986 syntactic analysis has been dominated by functional phrases—inflection phrases, agreement phrases, and the like. Each grammatical element has a phrase of its own; there’s agreement in sentences so you need an Agreement Phrase; there are Determiners so you need a Determiner Phrase, and so on. The lexical parameterisation hypothesis insists that these functional phrases are stored in the lexicon along with their heads, whether grammatical words or inflections. The lexicon thus has entries for functional phrases containing their parameter settings, as well as for the content words of the language. Learning the lexicon means learning the heads and parameter settings for each functional phrase. Thus the property of movement is assigned a parameter value in the entry for the Complement phrase head that so that in English the CP is +movement, in Japanese -movement. The principles of syntax are not learnt; only the lexicon. It is thus an extended system with both lexical and functional entries.
Inflections such as present tense ‘-s’ in English are therefore now part of the lexicon. The inflectional or lexical heads of functional phrases appear in the lexicon and are then projected in the syntax and 'checked' to see if the sentence works. are represented in the same way, for example that has a lemma including Complement Phrase, +wh movement, and a representation ¶ ¾ t//<that>. The syntax is lexically driven.
We can now try to see how this fits into the selection and generation components of a production model. To sum up, the conceptualiser produces a conceptual message which is then passed to a Macro Language Decider. The message now needs to be put into the language or languages that have been chosen. This involves consulting the lexicon, made up of lexical entries that give all the necessary information to construct sentences. The multi-competence requirement requires a lexicon that may include more than one language. If the lexicon consists of lexical entries, this raises the perennial issue, going back to at least Weinreich (1953), of whether there is one lexicon for each language the speaker knows or whether there is a single lexicon for both languages, or various other combinations. The assumption here mirrors that for conceptual organisation: undoubtedly some lexical entries belong to one language or another, others are shared by both to a greater or lesser extent. The assumption is that there must be a parameter for language in each entry, which ideally is set as + or – or ±.
In a monolingual mode the speaker sticks to entries matching the language chosen; if matching is unsuccessful within that set of entries, compensatory strategies are invoked. In a bilingual mode of production, the code-switching requirement means that the speaker has to be able to choose which language each element in the conceptual message is matched in the lexicon; if the Matrix model is correct, this applies to the entries for content words rather than to the heads of functional phrase. We therefore need a Micro Language Decider intervening between the Lexical Selector and the Lexicon, which will doubtless have many rules that restrict the choice of language.
Because a user has an entry for dog does not meet that all its specifications are present. The lexical entry varies at different stages of development; L2 users may know the word dog but may not have acquired the meaning seen in the song Who let the dogs out? or know the grammatical structures involved when dog is a verb, as in He dogged my footsteps. To meet the developmental requirement the lexical entry has to vary from, say minimal knowledge that dog is a Noun meaning 'canine quadruped' up to maximal knowledge of all its grammatical possibilities and forms.
The Lexical Item Selector therefore tries to match an element in the message with lexical entries with parameter settings marked for a particular language or neutral between languages. If it succeeds in finding a match, it produces a lexical package containing the lexical entries matching the conceptual message, incorporating all the grammatical and formal elements of the entries. If it fails to find a match, it turns to compensatory strategies for alternative ways of expressing the message, according to Poulisse (1993, p.180), using either substitution, with or without morphological changes, or reconceptualisation 'a change in the preverbal message involving more than a single chunk'. These different aspects are represented in figure 2.
Figure 2 Lexical Selection in Speech production
We now need to turn to the actual generation of the sentence, in the sense of assembling the elements rather than the earlier Chomskyan sense of describing them formally. The assumption is that there are two universal language components of the system that incorporation the one hand the unvarying principles of syntax, on the other the universal; aspects of phonology. In other words these are the unvarying characteristics of human language in the mind, the principles that remain the same regardless of the parameter settings acquired in the lexicon. These two components represent the computational system for language independent of any specific language.
The process of generation consists of fitting the information supplied in the lexical packages into the computational system. The properties of the packages dictate what the actual sentence structure will be, in syntax whether null subject or not, whether movement in questions or not etc, in phonology the permissible syllable structures, distinctive features, etc.
The developmental requirement requires different levels of syntax. In the Processability Model there is a progression from Stage 1 at which the L2 user has access to lemmas of individual content words to stage 2 at which grammatical functions are available; then the learner acquires the ability to project the entries successively onto phrases, sentences and subordinate clauses. This progression is captured in the multilingual production model as different aspects of the lexical entry. The generation component projects the aspects of the lexical entry that are available to the speaker at a particular moment on to the sentence, in ways that are too complex to deal with here.
The rest is silence! Sorry will get back to it one day