Bilingualism: Language and Cognition 1 (1998) 1-20 Q 1998 Cambridge University Press 1

MANFRED PIENEMANN Developmental dynamics in Australian National University L1 and L2 acquisition: Processability Theory and generative entrenchment*

This paper has two major objectives: (I) to summarise Processability Theory, a processing-oriented approach to explaining language development and (2) to utilise this theory in the comparison of development in LI and L2 acquisition. Proponents of the Fundamental Difference Hypothesis (between L1 and L2) assume that LI development can be explained with reference to Universal Grammar (UG) which, in their view, is inaccessible to L2 learners. Instead, they claim that a second language develops on the basis of language processing strategies. I will show that the fundamentally different developmental paths inherent in first and second language acquisition can both be explained on the basis of the same language processing mechanics (as specified in Processability Theory). I will demonstrate that the developmental differences between LI and L2 are caused by the qualitatively different early structural hypotheses which propagate through the acquisition process. The concept of “propagationof structural features” will be viewed as “generative entrenchment,” a logical-mathematicalconcept, which has proved to be highly productive in examining other kinds of developmentalprocesses.

spell out the sequence in which language processing A brief sketch of Processability Theory routines develop in the learner, we can delineate those grammars that are processable at different The wider context points of development. Learnability is defined as a purely logico-mathema- The architecture of human language processing tical problem (e.g. Berwick & Weinberg, 1984). Such therefore forms the basis for Processability Theory. a perspective ignores the fact that this problem has to In this perspective the language processor is seen with be solved, not by an unconstrained computational Kaplan and Bresnan (1 982) as the computational device, but by a mind that operates within human routines that operate on (but are separate from) the psychological constraints. native speaker’s linguistic knowledge. Processability In this section I will sketch out a theory which Theory primarily deals with the nature of those adds to learnability theory the perspective of proces- computational routines and the sequence in which sability, that is, “Processability Theory” (Pienemann, they become available to the learner. It will be argued in press). In my view, the logico-mathematical hy- that language acquisition incorporates as one essen- pothesis space in which the learner operates is further tial component the gradual acquisition of those very constrained by the architecture of human language computational routines. In other words, the task of processing. Structural options that may be formally acquiring a language includes the acquisition of the possible will be produced by the language learner procedural skills needed for the processing of the only if those processing resources are available that language. It follows from this that the sequence in are needed to carry out, within the given minimal which the target language (TL)’ unfolds in the time frame, those computations required for the learner is determined by the sequence in which pro- processing of the structure in question. Once we can cessing routines develop that are needed to handle the TL’s components. * I would like to thank Avery Andrews, Bruno Di Biase, Gisela In the rationalist tradition, learnability analyses Hakansson, Malcolm Johnston, Jiirgen Meisel, Bonnie Schwartz and four anonymous reviewers for their invaluable input into have in the past been based on four components that this paper. Any remaining errors are mine. I gratefully acknowl- edge the financial assistance from the Australian Research council which I received for the research presented in this paper. I The term “target language” is used here to cover L1 and L2.

Address for correspondence Linguistics Program, Department of Modern European Languages, Australian National University, ACT 0200, Australia E-mail: [email protected]

http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 2 Manfred Pienemann

must be specified in any learnability theory (e.g. edge. Instead, it is the objective of Processability Wexler and Culicover, 1980; Pinker, 1979): Theory to determine the sequence in which proce- dural skills develop in the learner. (1) the target grammar, (2) the data input to the learner, (3) the learning device that must acquire that Key psychologicalfactors in language processing grammar, and It is the aim of this paper to hypothesise, on the (4) the initial state. basis of the general architecture of the language The idea behind this is that a learnability theory must processor, a universal hierarchy of processing re- specify how a learner develops from an initial state to sources which can be related to the requirements of the target grammar by means of the available input the specific procedural skills needed to process the and the given learning device.2 TL. In this way, predictions can be made for The rationale for assuming these components is language development which can be tested empiri- rooted in the way in which learnability theory has cally. Before I describe the architecture of the hy- been formulated in response to the “logical problem” pothesised hierarchy of processability and the model in language acquisition (cf. Wexler, 1982). The of language production on which it is based, it will logical problem basically describes the following be useful to sketch out briefly a number of key paradox: children acquire in a relatively short period psychological factors in language Frocessing in order of time and on the basis of limited linguistic input the to characterise the processing environment within basic principles of their native language, although it which the learning of language takes place. The is assumed that many of these principles cannot be framework that follows from this will then be used inferred from the observations made by the learner. to establish a hierarchy of processing prerequisites. It has been noted by several rationalist researchers The formal adopted for the description of the (e.g. Felix, i984, 1991; Clahsen, 1992; Gregg, 1996) processing hierarchy will be shown to reflect key that besides linguistic knowledge, a theory of lan- properties of language processing. guage acquisition must also explain what causes the The view on language production followed in this development of the TL to follow a describable route. paper is largely that described by Levelt (1989), This explanatory issue has been referred to it as the which overlaps to some extent with the computa- “developmental problem” (Felix, 1984). tional model of Kempen and Hoenkamp (1987) My fundamental point is that recourse needs to be which emulates much of Merrill Garrett’s work (e.g. made to key psychological aspects of human lan- Garrett, 1976, 1980, 1982) on which the corre- guage processing in order to account for the develop- sponding section of Levelt’s model is based. The mental problem, because describable developmental basic premises of that view are the following: routes result, at least in part, from the architecture of (1) processing components are relatively autono- the human language processor. For linguistic hypoth- mous specialists which operate largely automa- eses to be transformed into executable procedural tically; knowledge (i.e. a certain processing skill), the pro- (2) processing is incremental; cessor needs to have the capacity of processing those (3) the output of the processor is linear, while it hypotheses. may not be mapped onto the underlying In other words, Processability Theory focuses meaning in a linear way; solely on the developmental problem as an explana- (4) grammatical processing has access to a gram- tory issue; it is not designed to contribute anything to matical memory store. the question of the innate or learnt origin of linguistic knowledge or the inferential processes by which Below I will briefly describe each of these premises linguistic input is converted into linguistic knowl- and highlight some of the empirical research sup- porting them. ’ It has oeen noted that the potential components of a theory of learnability interact: the stronger the first, the weaker the second Processing components are relatively autonomous and vice versa or, in the words of Bates, MacWhinney, and Smith (1982): specialists which operate largely automatically . . . As Pinker (1979) and Braine (1978) both noted, the The appeal of this proposition is that it can account Wexler and Culicover conclusion is not the only one that can be reached with learnability analysis. The strength of their (W for the speed with which language is processed. & C’S)fourth parameter (i.e. innate hypotheses) is required Levelt shows that adopting the assumption that only because ofthe value they have assigned to the other three. processing is neither autonomous nor automatic (P. 15) leads to serious problems. If the processing compo-

http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 Developmental dynamics in LI and L2 3

nents were not autonomous, exchange of information Incremental processing therefore necessitates the between all processing components (or levels of use of storage facilities to allow for non-linearity in processing) would have to be co-ordinated by a the matching of underlying meaning onto surface central control and processing would have to be form. This has important implications for the con- executed in a serial manner which would lead to cepts which will be developed below in relation to the extremely slow lock-step style processing. However, acquisition of language. We will therefore briefly not only has it been shown that language processing look at the interrelation between non-linearity and is much faster than would be predicted by such a memory. model, but empirical evidence also shows that dif- ferent processing components exchange information The output of the processor is linear, while it may not in a parallel (i.e. non-serial) manner (Levelt, 1989; be mapped onto the underlying meaning in a linear way Engelkamp and Zimmer, 1983; Sridhar, 1988). In addition, a central control would imply that the One case of non-linearity is the relationship between operation of the processing components is con- the natural sequence of events and the order of clauses. sciously attended to, while empirical studies have As Levelt (1983) points out, propositions do not neces- shown that grammatical information is normally not sarily have to be produced in language in the natural attended to and can only be memorised if attention is order of events. Consider the following example: focused on it (Bock, 1978; Kintsch, 1974; Garman, (1) Before the man rode off, he mounted his horse. 1990). Autonomous specialist processing components In this example the event described in the second can further be characterised as processing devices clause happens before the one described in the first which are able to accept and pass on only informa- clause. In order to produce such sentences then, the tion of a highly specific nature, for instance, only speaker has to store one proposition in memory. information which is concerning NPs. The advantage There are similar linearisation problems (Levelt, of such task-specificity is a gain in processing speed, 1983) which operate at the morpho-syntactic level. since non-attended processes can be executed in Such cases involves the storage of grammatical in- parallel. Thus the notion of task-specificity is in line formation. One such example is subject-verb agree- with the observation that grammatical information ment. Consider the following example: can only be memorised if it is attended to. Below it (2) She gives him a book. will become clear that the notion of automatic un- attended processing is closely related to the nature of The insertion of the verbal agreement marker cru- processing resources. cially hinges on the storage of grammatical informa- tion which is created before the verb is produced, namely person and number marking of the subject. Processing is incremental Note that the nature of the information held in This premise basically says that “. . . the next pro- memory is not the same in this example as it was in cessor can start working on the still-incomplete the above example. In the previous example it was output of the current processor . . . ” (Levelt, 1989, propositional content which had to be stored, while 24). The idea is that the surface lexico-grammatical in this example storage for grammatical information form is gradually being constructed while concepka- is needed. lisation is still ongoing. This feature was highlighted for the comprehension system in a study by Marslen- Gammatical processing has access to a grammatical Wilson and Tyler (1980) which demonstrated that in memory store on-line processing semantic representations are being constructed by the comprehension system before Both types of information need to be held only grammatical structures have been “worked off.” temporarily until incorporated into the generation of Incremental processing is a feature of human the message. However, the difference between the language processing. One important implication of two types of information is this: grammatical infor- this feature for the structure of processing algorithms mation is highly specific, and (conscious or uncon- is that in order to be in line with human processing scious) attention to it is not necessary; one does not they must be able to cope with non-linear linguistic need to be aware of or control the fact that the form without much “look ahead” (Levelt, 1989). In information concerning “person” and “number” other words every processing component can “see” matches between the lexical entries of the verb and only a small section of the current processing event the grammatical subject. In fact, it is possible to rather than having the complete event displayed. attend to only a small number of such processes.

http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 4 Manfred Pienemann

With the normal speed of language generation trates some of the key processes involved in the Working Memory would otherwise get “clogged up”. generation of the example sentence “A child gives a On the other hand, attention must be focused on the cat to the mother”. First of all, the concepts under- propositional content, because it reflects the intended lying this sentence are produced in the Conceptua- conceptualisation the speaker wants to express. liser. For the purpose of this paper I will ignore the Working memory is the resource for temporary internal structure of this component of language attentive processes which include conceptualising and generation, except for several features of the output monitoring (Baddeley, 1990; Broadbent, 1975). It has produced by the Conceptualiser. a highly limited capacity and is therefore not suitable In the example chosen in Figure 1, the conceptual to process great amounts of grammatical information material produced first activates the lemma CHILD at high speed. Levelt (1989) and other authors (e.g. in the lexicon. The lemma contains the category Engelkamp, 1974) assume that grammatical informa- information N which calls the categorial procedure tion is held temporarily in the grammatical memory NP. This procedure can build the phrasal category in store which is highly task-specific and in which which N is head, i.e. NP. The categorial procedure specialised grammatical processors can deposit infor- inspects the conceptual material of the current itera- mation of a specific nature. In Kempen and Hoen- tion (the material currently being processed) for kamp’s (1987) Incremental Procedural Grammar;the possible complements and specifiers and provides locus of the grammatical buffer is the specialised values for diacritic features, including those from the procedures which process NPs, VPs, and so on. head of phrase. I will assume that the first referent is One can see that the grammatical memory store is marked “- accessible”. This ensures that the branch a requirement that arises from the automatic (i.e. Det is attached to NP, the lemma for “A” is inattentive) nature of grammatical processing: gram- activated, and that the lexeme “a” is inserted. Func- matical processors handle highly specific information torisation Rules instigate the activation of free gram- which the gqammatical memory store can hold tem- matical morphemes and the insertion of bound porarily. Empirical evidence for the different nature grammatical morphemes. of the processing of propositional content and gram- The above attachment of Det to the NP-node matical information comes, among other things, illustrates a key feature of the language production from the study of aphasia. Cooper and Zurif (1983), process, which is crucial in the context of language for instance, showed that in Broca’s and, to a lesser acquisition. The selection of the lemma for “A” extent, in Wernicke’s aphasia lexical retrieval and partly depends on the value of a diacritic feature semantic representation are functional while gram- (“singular”) of the head being checked against that matical information cannot be processed. This is true of the targeted lemma. The value of the diacritic for production as well as for comprehension. feature is ‘stored’ by the categorial procedure until it As with other motor and cognitive skills, auto- is checked against that of the modifier. matic processes in language production utilise what Our production process has proceeded to the is known as “procedural knowledge” or “procedural point where the structure of a phrase has been memory”, which is contrasted with “declarative created and the associated lemmata are activated. knowledge/memory”. The latter “ . . . concerns What is missing to make this the beginning of a everything that can be represented at a conscious continuous and fluent utterance is establishing a level, and which groups together what Tulving relation between the phrase and the rest of the (1990) called ‘episodic’ memory, and Penfield termed intended message. This is accomplished by assigning ‘experiential’ memory (Penfield and Roberts, 1959)” a grammatical function to the newly created phrase. (Paradis, 1994, 395). There is ample empirical evi- In fact, it is the categorial procedure itself that dence from studies of amnesia and aphasia for the chooses its functional destination. This highlights the dissociation of procedural and declarative memory active nature of syntactic procedures. (Paradis, 1994), based on the patients’ loss of ability Possible functional destinations are defined in a to perform or to learn to perform certain tasks set of so-called Appointment Rules which are also which can be defined according to the procedural- language-specific. The default for NP procedures is declarative distinction. “subject of S.” However, this does not quite solve the problem of allowing the tree created so far to grow Incremental language generation into a sentence and to make the production of the sentence continuous. What is missing is the attach- The process cf incremental language generation as ment of the NP to a higher node. In the above envisaged by Levelt (1989) and Kempen and Hoen- example NPSubjcalls the procedure S which accepts kamp (1987) is exemplified in Figure 1, which illus- the calling NP as its subject and stores the diacritic

http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 Developmental dynamics in Ll and L2 5

Conceptualher

give (actor: child) (beneficiary: mother) (object cat) I II EVENT

I 1 i Gramma@ encoder

1emma:CHlLD conceptual specs: “CHILD” j syntacticcategory: N diacritic parameters: singular ...

lemma: A conceptual specs: ”A Syntacticcatepry: Det diaaiticparameters: singular/ ... I

Figure 1. Incremental language generation.

features deposited in the NP, namely the values for the assembly of phrasal subprocedures. I assume that “person” and “number”. for non-configurational languages grammatical roles The outcome of all of this is depicted by a tree can be specified directly from the semantic roles structure in Figure 1. While this structure is produced specified in the conceptual structure. And different and the associated lemmata are activated, the next procedures are assumed in the above model for the conceptual fragment would have been processed in processing of matrix and subordinate clauses. parallel and the output of the Formulator would To recapitulate, in the incremental process of have been delivered to the Articulator. This means language generation, the following processing pre- that new conceptualisation occurs while the concep- requisites are activated - among other things - in the tual structure of the previous iteration is being pro- following sequence: duced. The whole process then moves on from iteration to iteration. This is what Kempen and (1) the lemma, Hoenkamp (1987) and Levelt (1989) mean by incre- (2) the category procedure (lexical category of the mental production. lemma), In the above summary of the process of gramma- (3) the phrasal procedure (instigated by the cate- tical encoding one aspect was left aside, namely word gory of the head), order. The definition of the acceptable set of word (4) the S-procedure and the target language word order constellations for configurational languages is order rules, carried out by Word Order Rules, which co-ordinate (5) the subordinate clause procedure - if applicable.

http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 6 Manfred Pienemann

monolingual speaker, and the bilingual speaker may A hierarchy of processing resources master the two (or more) systems to differing It is important to note that this account of gramma- degrees. tical processing and memory applies only to mature Given the focus of Processability Theory on the users of language, not to language learners. While Formulator, the key assumption from De Bot’s work even beginning second language learners can make for the present context is that in all cases where the recourse to the same general cognitive resources as L2 is not closely related to the L1, different (lan- mature native language users, they have to create guage-specific) procedures have to be assumed. Based language-specific processing routines. For L 1 learners on our previous discussion, the following language- there are obviously no pre-existing procedures which specific processing devices are the least L2 learners can be transferred. L1 learners therefore have to have to construct to acquire the L2 grammar: develop all specific L1 procedures. Below I will gen- 0 word order rules, erate hypotheses as to how language-specific proces- 0 syntactic procedures and their specific stores, sing routines develop, given the general architecture 0 diacritic features in the lexicon, of the language processor. 0 the lexical category of lemmata, In this context it is important to ensure that 0 functorisation rules. Levelt’s model can, in principle, account for language processing in bilinguals, since second language acqui- Obviously, word order rules are language-specific sition will lead to a bilingual language processor. De and there is no a priori way of knowing for the Bot (1992) adapted Levelt’s model to language pro- language learner how closely related L1 and L2 are. duction in bilinguals. Based on work by Paradis Learners therefore have to be equipped to bridge (1987) he shows that information about the specific maximal typological gaps in their L2 acquisition. language to be used is present in each part of the Diacritic features of lemmata contain items such as preverbal message, and this subsequently informs the “tense”, “number”, “gender”, “case” etc. Again it is selection of language-specific lexical items and of the obvious that the list of diacritic features varies from language-specific routines in the Formulator. language to language. Drawing from Paradis’s (1987) research, De Bot Similarly, syntactic procedures that build consti- concludes that “. . . the speaker who speaks two tuent structures and store temporarily specific gram- closely related languages will for most part use the matical information such as diacritic features are not same procedural and lexical knowledge when the same across languages. Given that diacritic fea- speaking either of the two languages, while in the tures are language-specific and that these are stored case of languages which are not related an appeal is in syntactic procedures, L1 procedures are not made to much more language-specific knowledge.” equipped to handle the specific storage task required (De Bot, 1992, 9). De Bot further shows that Para- by the L2. dis’s (1987) “Subset hypothesis’’ about the bilingual The lexical category of lemmata may also vary lexicon is in line with the overall design of Levelt’s from language to language. Again, the language model. According to the subset hypothesis, the bilin- learner is only fit to acquire any of the world’s gual lexicon is a single storage system in which links languages if he or she tests the lexical category for between elements are enforced through continued every new lexical item. use. This has the effect that links will be stronger The reader will recall that Functorisation Rules between elements from one language. However, in instigate the activation of free and bound gramma- the case of bilingual communities with a tendency for tical morphemes. And the same is true for gramma- code-switching, links between elements from different tical morphemes as what was said about word order languages may be similar to those in a monolingual rules: these are language-specific and therefore have lexicon. to be acquired with the L2. De Bot (1992) demonstrates that the extended version of Levelt’s model accounts for the majority Exchange of grammatical information of the additional requirements which have to be met by a language production model in a bilingual In other words, the L2 learner is initially unable to context. These include the following requirements. deposit information into syntactic procedures, The two language systems concerned may be used because (1) the lexicon is not fully annotated, and, quite separately from each other or in varying more importantly, (2) because even if the L1 annota- degrees of mixes (code-switching). The two systems tion was transferred, the syntactic procedures have may influence each other. Neither system will neces- not specialised to hold the specific L2 syntactic sarily slow down in speech rate in comparison with a information. For this reason one can predict that the

http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 Bilingualism: Language and Cognition 1 (1998) 21 -22 0 1998 Cambridge University Press 21

PEER COMMENTARIES ELLEN BIALYSTOK What’s in a process? Department of Psychology, York University, 4700 Keele Street, Explaining development in Toronto, Ontario, Canada M3J 1 P3. E-mail: [email protected] language acquisition

The tension between considering language acquisition to be this way, it offers a sensible resolution to the apparently a process of learning or a process of development is the intractable dichotomy between learning and development visible surface of a theoretical iceberg. The difference and the implausible wedge it places between first and between learning and development is neither simple nor second language acquisition. Nonetheless, the use of the self-evident, yet the implications that follow from a com- term “process” and the nature of supporting evidence are mitment to one or the other are considerable. Development problematic. is usually construed as an unfolding, the impassive evolu- What does Pienemann mean by “process?’ Apart from tion of “being.” Children develop physically, cognitively, acknowledging a theoretical lineage to Levelt (1989) and emotionally almost in spite of themselves; they may not others, he offers no direct definition. He does, however, develop to their full potential, and they may require provide concrete examples of language processing devices: massive environmental support, but even in cases of benign word order rules, syntactic procedures and their specific neglect, children will normally become taller, smarter, and stores, diacritic features in the lexicon, and so on. Each of more stable. Learning implies a measure of control and these language-specific processes, he argues, needs to be responsibility. It depends on experience and on the avail- acquired in order to process (learn) the relevant linguistic ability of resources: conceptual, social, intellectual, and so structure. But how does one distinguish between the on. process and its attendant structure? Indeed, the so-called The distinction between formal and functional linguistic processes appear to be slightly abstract formulations of theories rests partly on the extent to which theorists linguistic structures, in other words, knowledge. What is consider that language acquisition is a process of learning needed here is a different conception of process, one that (functional) or a process of development (formal). Some- truly stands outside the specifics of linguistic knowledge. If times they coexist. Noting the fundamentally different it could be shown that a hierarchy of cognitive processes kinds of experience involved in first and second language influences the order in which linguistic structures can be acquisition, some researchers have explained them using learned, then that would indeed contribute to the reunifica- different presumptions about learning and development tion of language acquisition, not only consolidating the (e.g. Clahsen, 1990; Meisel, 1991). Specifically, first lan- acquisition of a first and second language but also inte- guage acquisition is left to follow the preordained paths grating language acquisition into broader cognitive laid down by universal grammar (hence, development), but domains. The nature of the definition of process adopted in second language acquisition is characterized as an exercise this theory could not meet those goals. in problem-solving and active participation by the language The problem of evidence is a circularity in the argument. learner (hence, learning). The claim is that the production of certain structures is Pienemann’s intention is to resolve the problem inherent constrained by the availability of the necessary processing in proposing two epistemologically different kinds of de- resources; the corollary is that structures not supported by scription to explain two phenomena that have obvious such resources will not be produced. The evidence, intrinsic similarity. His proposal is a framework he calls however, comes only from observed production. Hence, the “Processability Theory”, named for its assertion that con- claim that “Once we can spell out the sequence in which straints derived from human cognitive processing must be language processing routines develop in the learner, we can included in formal algorithms for language acquisition to delineate those grammars that are processable at different account for the grammars that language learners are able points in development” is fallacious. The “sequence of to handle at various stages of development. The important language processing routines” is derived from the “gram- claim is that this system can equally explain (and predict) mars . . . at different points in development” so cannot be the acquisition of both a first and second language. used as evidence for their predicted occurrence. Furthermore, it obviates the need to distinguish between Finally, a note on generative entrenchment. This is an considering language acquisition as either learning or devel- interesting idea: a structure set up in one stage affects all opment, since the two become interconnected. subsequent developments. The idea is reminiscent of the The theory is an attempt at reconciliation: it acknowl- notion of historical contingency that Stephen Jay Gould edges the role of formal linguistic knowledge (even allowing (1989) offers as an alternative, indeed, as an antidote, to for its possible innateness), yet emphasizes the need to the more usual logical forms of prediction. Events occur include a cognitive system that operates on that knowledge because they follow other events, and altering any incident in ways commensurate with known processing resources. In in the chain would change everything. Language acquisi-

http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 22 Ellen Bialystok

tion is exactly like this. Knowledge of language accrues by obvious differences in detail, were the same. He concludes: building on what we already know, not only about lan- “That, together with the strategic wisdom of taking two guage, but also about meanings, communication, structure. mysteries as one . . . commends the conclusion that they Second language acquisition necessarily has different his- are equivalent” (p. 236). torical contingencies than first language acquisition, not the least of which is that the learner already knows a language. The contingencies may be different, but the process must References surely be the same. Clahsen, H. (1990). The comparative study of first and second Processability theory is a contribution to the tradition of language development. Studies in Second Language Acquisi- seeking explanations that supersede specific events. We tion, 12, 135-53. need to be able to look past apparent surface differences to Gould, S. J. (1989). Wonderful life. New York: W.W. Norton. firid underlying commonality. Acquisition of language, Levelt, W. J. M. (1989). Speaking: From intention to articulation. including first and subsequent systems, rests on both the Cambridge, MA: MIT Press. Macnamara, J. (1982). Names for things. Cambridge, MA: MIT mechanisms of development that guarantee growth and the Press. processes of cognition that define competence. Pienemann Meisel, J. M. (1991). Principles of universal grammar and strategies may not have the processes right yet, but the attempt to of language use: On some similarities and diffecences between identify them is a notable start. John Macnamara (1982) first and second language acquisition. In L. Eubank (ed.), ended his important book on language acquisition with the Point-counterpoint: Universal grammar in the second language, conclusion that the minds of children and adults, despite pp. 231-76. Amsterdam: John Benjamins.

http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 Bilingualism: Language and Cognition 1 (1998) 23-24 01998 Cambridge University Press 23

On Processability Theory and SUSANNE E. CARROLL Applied Linguistics, Institute for English C? American Studies, second language acquisition University,of Potsdam, PF60 15 53,14415 Potsdam, Germany. E-nd: carroll@rz. uni-potsdam.de

Responding to an invitation to comment on Pienemann’s learning conditions are. Therefore, it does not add to either paper leaves me on the horns of a dilemma: I think the linguistics or psychology’s contribution to learnability or approach is novel and shows how future second language feasibility. Moreover, Pienemann grants that Processibility acquisition (SLA) research can be done. However, it is Theory will not account for the contents of the learner’s difficult to judge from the paper what the exact contribu- representations. Even allowing for severe constraints on the tion of the theory is to be. form of encoding systems provided by Universal Grammar, How can one disagree that Pienemann’s approach to we still need an account of how speech units are perceived, explaining SLA, which demands a commitment to an how categories are formed on the basis of cues from either explicit theory of grammar while focusing on the role of a system “lower down” in the processing system or a processing in describing the developmental stages of L2 system “higher up”, how generalisations are formed, and speech production, is a good path to follow? Isn’t it how exceptions to generalisations are observed and obvious? No! Processibility Theory research marks a sig- encoded. In my view, this is where most of the action in nificant departure from the standard literature on SLA, SLA takes place (see Carroll (submitted)). Finally, the which is only occasionally concerned with describing relationship between the hierarchy of processing resources mental grammars, and almost never concerned with the and the stages of development observed in the ZISA study details of how learners learn them. While the field abounds is not a matter of logic. Pienemann, like many others, with interesting, rich, and occasionally important empirical assumes that one learns “words/lemmas” before one can studies, including the ZISA studies in which Processibility learn the procedures which combine “wordsllemmas” in Theory is rooted, there are few attempts to write even production. A theory of grammar, however, is not the best partial grammars of a learner’s , and even source of models for a theory of linguistic memory. What fewer attempts to answer the question of how those gram- gets stored in linguistic memory may, or may not, corre- mars could have been learned from the input the learner spond to the units grammarians call “words,” and what has got. Little use is made of contemporary cognitive gets put to use in production will consist of all sorts of science research and SLA’s contribution to it has been to units, including affixes, morphosyntactic words, phonolo- date nil. Pienemann, in contrast, knows that we need clear gical words, idioms, and lexicalised “sentence stems” models of what grammatical knowledge consists of, com- (Pawley & Hodgetts Syder, 1983). We know that learners bined with equally clear models of processing mechanisms. can isolate long strings, presumably on the basis of their He ensures that his own work is rooted in a formal theory phonetic properties, which means that those strings can be of grammar, Lexical Functional Grammar (LFG), and stored and later analysed in the lexicon. Phrasal units need Kempen and Hoenkamp’s procedural account of speech not, therefore, at least in principle, result in production generation. solely from the combination of “words” into phrases by It is odd, however, that nothing in Processibility Theory phrasal procedures. hinges on adherence to LFG. This result is surprising, given Pienemann’s account of development begins only once the stated focus on learnability. That learnability theory learners have somehow encoded L2 stimuli in linguistic will not tell us how humans learn language is well-known in terms. Is it therefore a reasonable model of bilingual formal treatments of grammars, and serves as the starting language production? Perhaps not, for nothing in it ex- point of debates on feasibility (Chomsky, 1965; Bresnan, plains which production procedures will be transferred 1978; Pinker, 1984; Carroll, 1989). Learnability studies from the L1 to L2 speech production (or when, or why), define grammars which are learnable in principle. Feasi- nor how learners come to suppress transferred L1 produc- bility studies define grammars which are learnable in tion procedures, which must occur if L2 production is to be principle under conditions resembling those that humans fast, accurate and effortless (Green, 1986). It is clear to me actually face: limited exposure to stimuli, time constraints that a model designed for explaining production in knowl- on learning, reduced or noisy transmission conditions, edgeable and fluent monolingual native speakers will not limitations on attention and memory, and so on. Since carry over unamended to SLA. Processibility Theory is also feasibility is not a statement of the factors which determine not an account of skills development per se, since there is how a given individual learns some linguistic phenomenon more to be said on that topic than that learners develop at a given moment, on the basis of given stimuli, provided new processing procedures and “automatise” them (Pawley in a given environment, it stands to reason that defining a & Hodgetts Syder, 1983). What Pienemann has done is to feasible grammar is just one step in developing a theory of apply a particular formal model of production to a well- language learning. Pienemann’s paper is not about the established set of developmental facts. I want to emphasise nature of grammars. It also makes no independent contri- that this effort is anything but trivial. On the contrary, bution to spelling out empirically what psychologically real working out the details is difficult, and understanding the

http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 24 Susanne E. Carroll

limitations of one’s own model is hardly straightforward. References We can see this work as an effort to take seriously the most sophisticated empirical and theoretical findings of cognitive Bresnan, J. W. (1978). A realistic transformational grammar. In science research and to test these models against robust M. Halle, J. Bresnan & G. A. Miller (eds.), Linguistic theory data from developing bilinguals. As my commentary and psychological reality, pp. 1-59. Cambridge, MA: MIT should make clear, such models force us to be explicit Press. about language, learning, parsing, memory, and produc- Carroll, S. (1989). Language acquisition studies and a feasible tion. Explicitness allows us to formulate questions which theory of grammar. The Canadian Journal of Linguistics, 34 (4), 399-418. could not otherwise be asked. The Processibility Theory Carroll, S. (submitted). Znput and evidence: The raw material of exhibits these virtues. Even if the current version leaves second language acquisition. MS: Universitat Potsdam (1996). many issues unresolved, it will push us to clarify our Chomsky, N. (1965). Aspects of the theory of grammar. Cambridge, thinking about how L2 learners perceive, encode, parse and MA: MIT Press. understand L2 speech, and how they produce, monitor, Green, D. W. (1986). Control, activation, and resource: A frame- and self-correct it. Somewhere in this equation, if all goes work and a model for the control of speech in bilinguals. well, we will also find “language learning”. Brain and Language, 27,210-23. Pawley, A., & Hodgetts Syder, F. (1983). Two puzzles for linguistic theory: Nativelike selection and nativelike fluency. In J. C. Richards & R. W. Schmidt (eds.), Language and communica- tion, pp. 191-225. London: Longman. Pinker, S. (1984). Language learnability and language development, Cambridge, MA: Harvard University Press.

http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51 http://journals.cambridge.org Downloaded: 30 Aug 2011 IP address: 192.87.79.51