Stephen van Vlack
Sookmyung Women`s University
Graduate School of TESOL
Second Language Learning Theories
Spring 2007
Week 5 - Answers
M&M, Chapter 4: Cognitive approaches to second language learning
1. What is the basis of the cognitive approaches to language and language acquisition?
Cognitive approaches to second language acquisition and language acquisition in general try to explain the actual cognitive processes underlying language use and, by a certain amount of default depending on the model, acquisition. In their most basic sense they're trying to explain what actually happens in the brain as people are exposed to, use, and learn (new) linguistic information. For language acquisition, then, cognitive approaches will focus specifically on the cognitive mechanisms devoted to or involved in learning language, but the assumption here, in stark contrast to the generative approaches, is that language learning shares many similarities with general learning and cognition and at some point the two must interact. Most
It is important to note that different so-called cognitive approaches (the information processing approaches and connectionist approaches) vary greatly in how they see this unfolding, particularly as they disagree on the actual role or scope of cognitive processes in the use and acquisition of language. In addition, they also disagree on the cognitive models that they use to argue their positions.
All this means that cognitive approaches make use of cognitive models of the brain. The study of cognition or cognitive science, a new powerhouse science only since the 1980s, seeks to explain how the brain works by examining systems in the brain, including what might seem to be not particularly `thought-based` systems like perception, emotion, movement, etc.. It should be understood though that any study of systems also requires studying basic brain anatomy and functioning as the systems of the brain are certainly limited by such anatomical regularities. It must also be acknowledged by us as SLA researchers that language is just one of the many systems of the brain no matter how much some researcher may go on about the distinctly human aspects of language. The other widely accepted systems are - perception, attention, consciousness, emotion, memory, movement, and cognition. A moment`s thought will reveal that language would indeed seem to be reliant, at the very least, on memory and, not surprisingly, the information processing models, for example, make strong reference to models of memory as regards language learning. It is however, becoming increasingly acknowledged by brain researchers that all the systems mentioned above are interconnected, including language. Until now, however, only the connectionist models have fully acknowledged this fact and embraced it in their models of language use and language acquisition. This goes a long way in showing how far behind real brain research SLA models are. For a relatively simple introduction to cognitive science see: Banich, M. (2004). Cognitive neuroscience and cognitive psychology (2nd Edition). Boston: Houghton Mifflin.
2. What are the main processing approaches and how do they differ from one another?
It is important to recognize the position that information processing approaches take to the overall question of language and acquisition of a subsequent language before discussing their particular intricacies. Information processing models were designed primarily to be complementary to existing models of language, and for the most part that meant universal grammar. If UG covers the i-language aspect of language then the information processing models seek to cover e-language, but in a way which meshes well with UG approaches. In essence, information processing models are designed to try to explain how language is used within a UG framework for the most part. This is a valid point because certainly there are cognitive processes which control how we actually go about not only producing language but also recognizing language and this is not handled at all in UG models of language with their focus on competence (i-language). At some point as well it must be acknowledged that these processing concerns must also relate to acquisition and we have the variability of subsequent language learning to argue strongly for this possibility. At this point we can see that information processing models are developed from the idea of input. The big question is, what happens to input? Each of these models seeks to try to understand how input is used in actual brain functioning and how that might affect the acquisition process.
McLaughlin`s Information Processing Model
So simple = Short-Term-Memory (STM), Long-Term Memory (LTM) - Controlled -vs- Automatic followed by Restructuring.
While restructuring is certainly a part of our general cognitive mechanisms in how we treat information in the brain, there is no specific support for a clear and definitive distinction between controlled and automatic processing as relates to memory storage.
Anderson`s ACT Model
Working memory and Declarative -vs- Procedural in three stages - Cognitive, Associative, and Autonomous
Here the very important idea of working memory was introduced but is not fully explained. It is treated as the same as short-term memory, which it is not in most modern theoretical accounts of memory (Miyake and Shah, 1999). As for the other types of memory mentioned they work better and are more widely accepted than those mentioned by McLaughlin, but the model is still a vast oversimplification as no distinction is made between episodic and semantic memory as specific kinds of declarative memory, which is now the common practice in memory theory (Eichenbaum, 2002).
Processability theory
How language is perceived and used is organized into stages based on the precessing required. In each stage a different hypothesis is used which sanctions or allows another to be used after that until the processing process becomes complete. These different hypotheses are housed in a special part of the brain. On top of this there is the idea that the salience of structure will make the noticing of hypotheses from that structure easier or difficult.
Perceptual Saliency
This is the simple idea that certain structures are more salient than others. That means they draw our attention and are easier to figure out because they seem to fit in better with our cognitive structure than other. They are, simply put, intuitively valid. To make this more scientific, formal, and useful specific operating principles were posited which represented potential salient properties.
One weakness of all these information processing models which does not necessarily exist in the connectionist models is that they all try to regularize the cognitive processes among learners. While people share cognitive systems and properties, current research is showing that no two people go about language processing in the same way. Once more,, the same bit of language might be processed differently on two separate occasions. Attention, for example, and how people use it, is largely determined by a person`s individual experience and it is reflected in other cognitive systems. It is hard to make fast and sure rules or overgeneralizations about how all people do this.
3. How do processing approaches relate to UG?
This is an interesting question because processing approaches were essentially developed as a way of complementing the theory of universal grammar. Universal grammar focuses exclusively on i-language or competence. In the 1980s it became apparent that some sort of theory of e-language or performance or language use was necessary. Therefore, these processing approaches were developed to try to answer questions of language learning as it related to structural elements outside of i-language by focussing on generalized ways in which people process linguistic input. The processing model here is somewhat vague in the distinction between dealing with input and the output that learners produce. In effect, they handle both without commenting overtly on their relationship. This is interesting because in trying to link processing to actual usage and trying to come up with a theory of usage these processing theories actually begin to intrude onto the realm of competence. It seems to me that the two different models, that is processing models and universal grammar, try as we may do not really fit very well together. The stronger the processing model the more it is going to intrude into the traditional realm of UG. At this point then UG begins to unravel. It's a problem, a real problem for SLA theories make in this mold in general.
Also, it is important to remember that UG is not simply a property theory as Mitchell and Myles (2004) claim. One of the most interesting elements of UG as a theory of language is that it was, from the very first also and primarily a theory of acquisition. In its initial stages of development UG was actually more a theory of acquisition than of properties. Acquisition was used to pinpoint the types of properties that were thought to exist, but it took a lot of time for the specific properties to emerge. As the properties emerged the transitional states became much less vague (remember the highly vague notion of the LAD as a black box) more specific. In a UG model e-language cannot account for acquisition. They are separate things.
4. What are some of the basic ideas underlying the connectionist view of language and language learning?
Connectionist models come from a very different point of view than processing models. They do not believe in universal grammar at all or even the concept of language put forth in generative linguistics and therefore are not trying to form theories which complement a universal or modular/innate basis. In the connectionist models there is no distinction between competence and performance. For them language usage defines a speaker`s knowledge of their language(s). There is no clear-cut distinction or different systems working for either competence or performance and to claim so is just stupid in this view. It is stupid because not only does it create a system which is overly complex and counterintuitive, but it also ignores a huge amount of actual data which we have about the language acquisition process. Another key idea which the connectionist models sees differently from many of the processing models is modularity. In the connectionist models there is no such thing as a strict application of modularity. Systems in the brain are variant, yes, but they are much more similar than they are different and they can and do affect each other. In fact, following recent brain research we can even say that different cognitive systems, including language are reliant on each other (Damasio, 1994). Part of the problem with a strict application of the modularity hypothesis is the systems can be totally different. This seems really strange the more you know about the brain.
At least currently, the connectionist model seems to be the most cognitively sensitive model. This means that the connectionist plays much closer attention to the rapidly developing field of cognitive science and tries to incorporate new discoveries into the model quickly. This, combined with the fact that connectionism model comes with very different tradition then information processing models, means that the connectionist model tends to see things in a different way than information processing models. While information processing models are busy trying to account for elements of cognition under certain UG constraints, the connectionist models have been developed separately from UG and, to a certain extent, in conflict with UG. For this reason there is little or no overlap between the connection is model of perception and production (e-language) and the UG model of i-language. At this point it should be known that the connectionist model is not merely a transition model either. Connectionist architecture has always been locked a the hip a with a state model of language called alternatively cognitive linguistics or usage-based models of language. These state models could not exist without the underlying architectural design of connectionism and connectionism without such a theory of the language state would be vacuous. This is even more obvious when we consider that in such models there is no distinction between competence and performance. In this way connectionism is part of the state of language and the state of language is also part of language usage. The two are the same. Obviously, we have just barely burst the bubble on these connectionist models, but they are models which need to be watched more in the future because they are very modern and very flexible and will draw increased attention from researchers as certain aspects of the models are refined and made less vague.
5. What effects have cognitive approaches to SLA had on the field?
Cognitive approaches have shifted the focus back to very basic questions about language and language learning. They have allowed us to bring the brain back into the picture of language and language acquisition, a notion Chomsky has only done lip service to. It should be remembered that UG is a model of language which has nothing to do with the actual structure and workings of the brain. Chomsky's justification for this is twofold. He claims on the one hand that, due to the strict modularity he envisions in the brain, there's no need to create a model of language based on what we know about the brain because everything is potentially different within the brain. He also claims that we simply don't know enough about the brain to create a model using brain structure. This of course was said back in the 1950's '60s and '70s, when indeed very little was known about the working of the brain, but is now no longer true. In this respect the development of cognitive approaches has been inevitable. As we know more about the brain then it would seem obvious that such knowledge would have to be incorporated into models of language, which is obviously and without a doubt an important system of the brain. The fact that the currently most popular model of language simply ignores the brain is a problem. Thus we really don't want to see a Chomskyan model as a cognitive model. In most respects it is in fact an anti-cognitive model. The overall effect and is that we can no longer create models of language which in no way reflect aspects of cognitive architecture. We no longer have this luxury. Therefore, classical behaviorism is really and truly dead and UG-oriented researchers are busy scrambling to try to find neurological support for the model. This is just the beginning of a lengthy process, for as we learn more about the brain, models will need to change again and again. Keep your eye on these developing trends.
Johnson, Chapter 4: Information processing models
6. What are the main similarities between Krashen`s and Long`s hypotheses?
The main similarities between the two models or hypotheses is the role of input and the fact that input is necessary for learning. At this point the two hypotheses differ for Krashen does not believe that interaction is a necessary aspect of language acquisition while Long sees this as a central component. It is unclear for Krashen's hypothesis where and how learning actually occurs based on the input. The input must be comprehensible but no discussion or claims are made as to where this comprehensibility comes from. There are two possibilities, it comes from the learners internal state or it comes from the input itself. In looking at this, it seems to Long that there certainly is no guarantee that input coming in from the outside world is ever going to be comprehensible. Certainly people do work to make input comprehensible to others, but how well people are able to do this will vary considerably from person-to-person and also depending on who is being spoken to in which kind of surrounding context. It seems to Long that comprehensibility must be primarily an internal situation. Thus, some input is comprehensible because learners render them comprehensible by using what they already know and making th necessary associations. This means that learners must somehow mediate with the input. This is especially true when we see how faulty input often is. Mediation must occur for input to be effective.
It is interesting to mention at this point that these information processing models are all either direct descendants of Krashen's highly vague ideas or have certainly been strongly affected by it (as in the case of possibility and learnability). They have taken up on the idea of input and broken it down into different areas which seem to be part of our generalize cognitive systems, such as attention, for example. Thus we find many of these different models share a lot of underlying similarities but differ on how they think learners react with or deal with this flow of input.
7. How does VanPatten`s information processing model work and what effects has it had on SLA?
VanPatten`s model is all about attention and the conflict between form and meaning in the reality of limited attentional resources. Meaning generally wins. This is stated as the result of three basic processing principles. A major distinction, following Krashen, is made between input and intake in this model. It does not seem to have anything particularly interesting or compelling in it. It is a more generalized theory for sure. It is a more refined theory, but it is still too simple and to constrained. Yes, attentional limitations are a cognitive reality, but what is being found out at this point is that even though attention is a particularly focused on one area that unfocused area does not go unperceived or unprocessed, particularly if there is some underlying meaning to it. This is called consciousness and is paradoxically one of the most highly studied and least understood areas of human cognition. In this model attention need not be focused on to the meaning overtly in order for it to be processed - yikes. The problem I have with a lot of these models is that they try to make a system which is rather individualistic into one which is highly regular and unvaried. I'm not sure this is really possible based on what we know that the brain right now.
8. How does Gass`s compare with those of VanPatten and Krashen? What is there in it which is particularly telling or innovative?
There are 5 stages - apperceived input, comprehended input, intake, integration, and output
This model is interesting and would seem to have some promise in that it does reflect certain cognitive realities about both perception and storage, but there's one main problem. The system is still based on the idea of modularity and has to be because if it does not follow modularity then it cannot be used to support UG. If we somehow managed to prove that modularity is not a cognitive reality then universal grammar as a model of state will cease to exist. Because this model relies on modularity it does not allow different elements to affect each other. Yes, we have our two different kinds of input, apperceived input and comprehended input and we know that the latter is then part of intake, but at some point they must affect each other in a rather important way. Things really do seem to affect each other in the brain. What actually is taken as apperceived and what is taken as comprehended will vary depending on the past experiences with different inputs so in this way it's hard to have a strictly modular system that really works in describing output and reflects a cognitive reality.
It seems to work a little bit too much like a machine designed to evaluate and distinguish input. It's seems that there isn't enough variability built into the system or ways for the learner to develop in different and again variant ways.
References
Damasio, A. (1994). Descarte`s error. New York: Penguin Putnam.
Eichenbaum, H. (2002). The cognitive neuroscience of memory. Oxford: OUP.
Miyake, A. and P. Shah (Eds.). (1999). Models of working memory. Cambridge: CUP.