Updated October 18th, 2021 Don’t forget to refresh the page!
This page provides a brief sketch of the basic cognitive architecture of the MCF (see also the short introductory YouTube presentations and also the 2017 introductory book on language and cognition (‘Book 3’) mentioned in the publications page). The last part describes the way in which language is embedded into the system as a whole.
GENERAL ASSUMPTIONS
What is a theoretical ‘framework’?
First we should explain what exactly a framework is as opposed to with related concepts. It is sometimes difficult to work out exactly what people mean when they talk about theories, hypotheses, models and frameworks. We see ‘frameworks’ as follows. Whereas a theory attempts to fully explain a particular set of phenomena, a theoretical framework, as the name implies, provides a basis for developing a number of theories about a range of phenomena, leaving researchers to provide more detail, including alternative explanations within that overarching framework. In this sense theoretical frameworks are relatively flexible since they allow such alternative theories and hypotheses to be developed.
The MCF is provides an open-ended psychological model of the mind (not the brain) which requires extensive elaboration by researchers working in different fields of research. It is also a biological model in the sense that it is intended to facilitate brain-mind mapping in accordance with current and future research findings. It might be viewed as a template for developing the human ‘psychome’, detailing features that, taken as a whole, distinguish us from other species.
Theory and metatheory
In another sense frameworks have a basic body of theoretical assumptions all of which are relatively fixed in nature and which define that particular framework.. This can be thought of as the framework’s metatheory. Take linguistics for example. One such assumption might be that humans possess a special language ability which is different from what many think of as ‘general cognitive’ ability. This biologically endowed ability allows them to develop grammars of any language they are regularly exposed to. These grammars how ever varied they maybe in other respects always conform to certain innate principles and constraints. Armed with this ability, young, cognitively immature children can master any language system without instruction. If your theory does not assume this basic assumptions, it cannot be part of that particular theoretical framework. There are, however, many different ideas around about what those innate principles are and how they play out in the building of different grammars: these examples of implementations of the framework rather than of the basic assumptions. There are other frameworks which have different basic assumptions about language altogether (see the Comparisons page).
The scope of a metatheory
Frameworks can vary in scope or, put another way, can be more or less ‘global’ or more or less ‘local’. The Chomskyan linguistic framework mentioned above provides researchers with a metatheory that covers the structures of grammars, that is phonology plus (morpho)syntax. Many aspects of language are not covered, including much of semantics, pragmatics and online language processing. The MCF is a much more comprehensive, overarching framework covering all of language, and cognition in general. More wide-ranging frameworks may of course incorporate more ‘localised’ frameworks such as the one just mentioned as long as their basic assumptions are compatible with the basic assumptions of the ‘mother’ framework).
What appears in various publications devoted to this framework and in the pages on this website includes both a) the unchangeable (metatheoretical) assumptions of the framework as well as b) a number of additional elaborations which may be therefore viewed as one of a possible number of alternative explanations for the various phenomena discussed. These serve as illustrations of how the framework works but they are also proposed theoretical extensions which are not not part of the framework as such but have been built applying the framework.
Since the description of the architecture that follows will look at first glance as though it is purely about processing, it is important from the outset to stress the fact that is about both online processing and representations, how they formed, what principles control the way they are formed and how they develop over time.
THE ARCHITECTURE
Modular Cognition Framework mechanisms.
Cognitive processing in MCF is modular in line with mainstream psychological and neuroscientific thinking. In other words, it is composed of many specialised processing systems (or ‘subsystems’) like those dealing with linguistic structure or those dealing with smell and vision. These systems cooperate via their interfaces to form coalitions which are assembled on line in order to deal with a multitude of different tasks that the organism is faced with. For those primarily interested in different types of knowledge representation and less in what happens in real-time performance, these should not be regarded as just ‘processing units’ but can be equally thought of as functionally specialised representational systems. One major feature of the MCF is that it handles the architecture of both the representational systems in the mind AND of what happens to them in real time as part of a single integrated explanation.
Cognitive processing does not operate as a straightforward sequences of events going in only one direction: the flow goes backwards and forwards as each system, separately and also in collaboration with one another, in order to form an optimal response to whatever happens to be the current set of tasks. This parallel and incremental mode of operation is in general agreement with many current approaches within cognitive science.
In its basic design, as indicated above the MCF is closest to the architecture that has been proposed by Ray Jackendoff for the language faculty (e.g. Jackendoff 2002). This architecture has been extrapolated in various ways in order to account for perceptual, motor and affective systems as well as the thorny issue of consciousness. In short, mental processing typically works with different systems operating at the same time in parallel to establish the best fit with the current state. The ‘current state’ is in constant flux dealing with internal changes including those triggered by the sensory experience of what is going on in the immediate environment of the organism.
As processing occurs, there is also, as a result, development in the system. The various processing units adapt in various ways albeit constrained by the particular principles under which they operate so that, for example, changes in the visual environment will not automatically cause, say, the auditory system or the syntactic system to adapt. A system (processing unit) only changes as in response to activity that is triggered during its own attempts to handle some task and not necessarily because something happens elsewhere in the system or in the outside world, i.e. in the organism’s immediate environment.
Functionally specialised systems
Each system in the mind’s interactive network of systems has its own unique function. This unique status is based on the fact that each runs according to principles not shared with any other systems. What the precise nature of each set of these principles is remains an empirical question which is up to researchers working in the relevant area of cognitive science to resolve as best they can. However all systems have a very basic architecture in common: they are each composed of a processor and a store. One way of looking at this is by taking a processing perspective so that a given system is a processing unit and the store functions as a memory although actually some elements in the store may be there from birth with all the rest being acquired via life experience. These ‘elements’ are simple or complex association of structures generally known as representations. Visual representations, for example, are structures located in the visual system and written in the specific code determined by the principles enshrined in the visual processor. Stores are linked across by ‘interfaces.’ These are represented by the two-way arrows in the example below, in Fig. 1, taken from language cognition, allowing associations to be formed between representations that are written in different codes. ‘Association’ in this case does not imply ‘combination’. Precisely because these representations are written in mutually incompatible codes they cannot be combined in the same way that representations in one store can be combined but they can be associated and hence co-activated during online processing. In this way, for example, a visual representations can be associated with a gustatory (taste) representation such that the sight of something, a fruit say, can activate an associated taste. In the same way an auditory (sound) representation sound can, as result of the chain of associations displayed in Fig.1, coactivate a meaning in the conceptual system.
Fig. 1. Example of 3 systems involved in language cognition with connecting interfaces
(LEFT CLICK ON IMAGE FOR FULL SIZE & BETTER QUALITY)
The stores in these ‘processing units’ (systems) are linked to other (one or more) stores by means of interfaces. Fig. 1 only shows single and double interfaces but the conceptual system (for example) is directly interfaced with many more systems than is shown here, including the auditory system. This means that Fig 1. is incomplete since there is actually an interface (not shown) directly linking the first and last system in that chain of systems. Multiple interfaces represent the norm.
The Outer Ring (“Perceptual Portal”)
The interconnected cognitive systems that make up the mind belong to one of two ‘rings’. The outer ring forms a ‘perceptual portal’ opening out onto the outer world. It is fed by data from each of five senses. These five form the physical portal to the immediate environment with inner ring of cognitive systems receiving sensory input from them from which the create and active representations. Each of these types of perceptual representation are formed according to the principles that govern the system question. This means that for example, auditory representations formed form auditory/acoustic input via the ears will be constructed differently from visual representations formed from visual input via the eyes. The neural equivalents in the brain will look very different but they will also likewise display distinguishing characteristics that make them identifiable as either ‘visual’ or ‘auditory’. The figure below shows the outer ring with its five systems:
click image to enlarge
Fig. 2. The perceptual portal
The Inner Ring (“Deeper Processing”)
Representations in the outer ring have in themselves no meaning and no value, do not, for example, influence the way the body moves or how the mind uses space to understand or navigate though the three dimensional outside world. All these feature, and more, are only possible when they are associated and coactivate representations in the systems on the inner ring. The figure below shows the inner ring with its six systems:
click image to enlarge
Fig. 3. Deeper processing.
Interfaces
The function of an interface (between different stores, see Fig. 1 above) is to marks associations between representations in different stores. Thse associations may then be acivated in working memory (see next section). ‘Association’ is done by indexing structures in adjacent systems. For example an auditory structure and a phonological structure are associated by virtue of having an identical index. They may already have been associated or, if not, the index is assigned to each of them during processing. An index could be thought of as a numbered tag or specific ‘address’. Chains of structure created during processing consist of items with the same tag (i.e. index) linked together.
Note that there is no traffic of information through from one module to another. Items from one system cannot be processed by another; as already mentioned, they also cannot be transformed (‘translated’) and incorporated into items in the next processing unit. The modular nature of the processing units prevents this. The function of an interface is just to put these structural items in registration with one another and not to convert one type of structure into another. This is impossible because each system has its own unique code. This render information contained in structures (representaions) bolonging to one system/store incompatible with anything in another store. In this way networks of association (called ‘schemas’) arise and co-indexed items are created in parallel to cope with the particular task(s) in hand. Fig. 4 (below) shows all the interfaces where this co-indexing may take place. This allows for an enromous number of possible schemas (patterns of association across stores) to be formed and activated online.
The ‘Full’ Picture
The figure below shows the almost complete set of systems on the two rings with interconnecting interfaces to allow many different types of single and multiple associations.
click on image or Figure title for a larger size
Fig. 4. The complete system: 2 rings combined and their interfaces
Interface connections within the OUTER ring marked in BLUE
Interface connections within the INNER ring marked in RED
Interface connections crossing BETWEEN rings marked in BLACK
Representations, networks and schemas
To reiterate what was mentioned above, a representation is a structure that can be more or less complex and can be both something that is housed within the memory store of one module and encoded in the same ay like a visual representation or a syntactic representation for example . Alternatively, it can be a schema composed of associated structures in different stores that are connected via the interfaces between the participating modules so that they can be activated together.
Complex representations that involve connections between different modules form representational networks, .e. schemas. These can occur spontaneously to cope with constantly changing circumstances. meet. However, if these schemas become well established by virtue of frequent co-activation in a regularly occurring type of situation, they can become representational schemas that have become readily accessible and thereby facilitating rapid responses to regularly occurring situations.
A linguistic example: An example of a schema of three representations would be a word which is composed of a phonological representation associated with a syntactic representation and finally a conceptual representation (giving it its meaning), in other words a word can be defined as a schema of association between three independent systems such that, when a given phonological structure is activated then the interfaces involved co-activate in parallel the other structures with which it has become associated. This can operate in the other direction with first the meaning (conceptual structure) being activated: this then triggers the same set of co-activated structures, i.e. the word which, if it becomes readily accessible might be thought of as a ‘mini-schema’. However, schemas normally involve many nodes with multiple connections. Schemas may compared to the complete network shown in Figure 4. Schemas are also networks but, unlike the network displayed in Figure 4, they are malleable and flexible because they can be created and modified with experience.
,
Working memories
Working memory (WM) is not a separate memory system. It is a state that representations can be in, specifically a particular current state of activation. When not activated, they are at a specific ‘resting level.’ An activated representation i.e. a representation ‘currently in working memory’ can be visualised using the metaphor of height rising towards the upper area of a memory store (the lighter-shaded ‘blackboards’ in Fig 2. above) occupied temporarily by items (representations) in the store that have risen up (been activated). This area is where the relevant processor does its work assembling the active items (representations) according to its own particular rules or principles. Another possible metaphor, not used here, would be to depict activation as varying intensities of light. Note that to be in a working memory state a representation can be at varying levels of activation. In other words, WM is not a single fixed, predetermined level (or ‘degree of intensity’).
During online processing, representations will compete with one another in a given memory store for participation in the current task selection to provide what will turn out as the mind’s best possible fit for the current situation.
Conscious experience occurs when representations are activated to a very high level. Most WM operations are conducted at lower, subconscious levels. Notable in this context is WM in the two linguistic systems where (PS and SS) representations always operate efficiently at subconcious levels and so never feature on conscious experience. Knowledge about language structure (grammar, vocabulary, pronunciation, etc) resides in the conceptual system so thinking actively about language involves appropriately high activation levels in conceptual working memory while PS and SS representations remain hidden from awareness.
Activation and resting levels
Resting levels of representations in a store (still using the height metaphor) will vary. A representation (typically a recently formed one) with a low resting level is handicapped when it becomes activated and has to compete with rival candidates for participation in some current processing task. Activation of a given item in a memory store has the effect of increasing its current resting level and makes it that much more accessible over time as it is regularly and frequently activated and therefore more competitive.
Attrition/’Forgetting’
Attrition like acquisition is growth but in the opposite direction: long periods whereby an item in a memory store is not activated may lower its resting level and make it less competitive during processing. Very long periods of inactivation will render it virtually inaccessible so that, for example, it will be hard to tell if, say, a word or a grammatical construction has been lost completely or is simply available ‘in principle’ but so rarely used that in most situations the language user will fail to ‘find’ it and therefore try alternative solutions.
LANGUAGE IN THE MCF
The MCF accords language a major role in the mind as a whole. The use of language activates multiples systems and not just those that specialise in linguistic structure. In conjunction with the conceptual system it accounts for the quantum leap in cognitive ability that homo sapiens has undergone. What follows is a particular implementation of the MCF, in other words it uses MCF architecture to develop a particular account of how linguistics structure is handled as part of a more general count of how language operates in the mind as a whole.
MCF publications and presentations in the Publications page provide many examples of the way the MCF is applied to explaining the role of language in the mind as a whole. The architecture should appear familiar to those who know Ray Jackendoff’s ideas, which have provided its original inspiration. Not all aspects of the MCF are exactly in line with Jackendoff’s ideas on language cognition and cognition in general but there are many similarities. The architecture has been elaborated and extended and focus is very much on on-like processing also with respect to how cognitive representations are created and changed over the lifetime. What is described below and elsewhere on this site is the functional architecture of the mind with special reference to the role of language (see the Mind vs Brain page for how this relates to neurological issue
Human language processing is mediated by a core language system (syntax plus phonology). Sounds (AS) and meanings (CS) can still be associated without this mediation but the result is very different. In the figure below, showing the route beginning with an AS, i.e. an activated representation of a sound pattern that originated in the (acoustic) environment and is now stored in the auditory system.
The bottom route marked by dotted lines (AS<>VS<>CS) is shared by humans and other mammals, for example. The top ‘humans only’ route that goes via the core language system (AS<>PS<>SS<>CS) is specific to humans.
This figure shows how some animals can associate given sounds and images with a given meaning and yet still be unable to create anything like human language. It also shows how we humans can process a word that we hear both simply as ‘a sound with a meaning’ via the dotted, bottom route and, at the same time, as a word made up of phonological and syntactic structure (PS and SS). Note that if we do not actually know or recognise the word, attempted linguistic processing will still kick in automatically anyway even though it might not result in successful processing.
Imagine that someone says “sunflower”. Fig 5 shows the meaning (CS) SUNFLOWER at the end of each processing route. A dog, for example, might be able to learn and respond appropriately to the sound of the word “sunflower” via the lower route which is also shared by humans. However only the human will be able to use the upper route where sound and meaning are given additional linguistic structure (PS and SS). In this way, the dog can still behave in a limited way in a ‘language-like’ manner but only the human will be able to incorporate the word “sunflower” into a vast number of larger linguistic structures and longer utterances. For the human, “sunflower” is what we call a ‘word’ but for the dog it is just a ‘sound’.
(left click on image for better quality)
Fig. 5 Associating sound and meaning: alternative routes
Note that, for convenience only, the auditory structure in this figure has been represented using simplified phonetic script enclosed in square brackets. This is because this is the standard way in which phoneticians represent what in MCF terms would be called ‘auditory structure that happens to have linguistic (in this case phonological) structural associations’. Real auditory code would make no distinction between this and the representation of any non-linguistic sounds like the ringing of a bell or a water splash.
What about decoding a strange language?
When listening to a speech in a strange language, we can certainly arrive at an auditory structure (AS) but manage at the same time only fragmentary phonological structure (PS) and, indeed, no syntactic structure or conceptual structure (CS) at all. If a word or phrase in a strange language is uttered in isolation, we may still be able to associate some kind of meaning with it by using the general context in which it is uttered. Since linguistic processing operates automatically (and subconsciously) we (our minds, that is) will always try to make linguistic sense of the word even though our system has not fully worked out its linguistic structure. With repeated processing in the right circumstances we may gradually develop the appropriate linguistic knowledge
Handling more than one language system
It is important to note that the same two linguistic systems system’ (above) serve any number of languages known at any level of proficiency to the individual. All associated related structures will be coactivated in any act of language processing. So how do we keep our activated languages apart when engaged in production or comprehension (or thinking)? It comes down to a question of interconnections, i.e. different networks of representations for different languages but within and between exactly same set of systems and of the relative levels of current activation which determine the winners and losers in the competition that is a constant feature of online processing.
There are no identifying ‘language tags’ on particular linguistic structures to get them selected appropriately. Language identity is established elsewhere. A very simple example of this is as follows: the links between certain types of sound – say ‘Japanese sounding utterances’ – and/or certain types of visual input such as Japanese kanji script, Japanese objects, etc are associated with conceptual structures with the meaning (CS) JAPANESE’ can boost the activation levels of all the relevant linguistic representations built up during previous exposure to Japanese: this will make them more likely to overcome rival candidates and raising the likelihood of producing a consistently appropriate result. Additional help in these circumstances is provided by an association from the affective systems whereby the CS JAPANESE acquires a strong positive value.
Note, in passing, that ‘spreading activation’ might suggest free-for-all, unconstrained networks of interacting elements as in a classical connectionist view of the mind. It is therefore important to remember MCF has a modular architecture which involves constraints as regards what precisely can interact with what as well as the rules of engagement, that is, the specific ways in which interaction can take place.
The complete system (mind map)
Below is a complete system including the core language system (the two modules shown within the dark grey oval shape on the left). The stores are linked by interfaces: this means in effect that any structure (representation) within the linked stores may in principle be or become associated with structures in the other store(s) and then will be co-activated on line. Stores are represented here as boxes. This figure represents is the complete system of modules by which an infinitely large number of processing schemas can be created and adapted to solve a myriad of different kinds of tasks every second of the day or night:
(left click on image for better quality)
Fig. 7. The MCF (stores, their respective processors and connecting interfaces between the stores)
[Note that a further, non-perceptual spatial system was added in July 2018]
In the perceptual group of systems (pale boxes and circles in the figure), only five ‘senses’ have been chosen for the display not to complicate it still further. Most accounts nowadays assume more than the traditional five. Note that the interface arrow linking phonology with the visual system is crucial for sign language but also for the interpreting of written text. Speaking is the primary mode of language production; the production of written text (‘writing’), however, requires a direct link via the auditory system so only an indirect link with phonology is in principle necessary. You read a word or a (Chinese) character and the primary link is the sound of the spoken word(s): these are represented in auditory code and the relevant auditory structures are directly associated with phonological structures in the phonology store.
Further Questions about Language
So where IS language?
There are many ways to answer this question. Language is to be found practically everywhere. On signs, billboards, webpages, TV screens, the pages of novels. It also comes to us in the form of sound waves, in speech and also in the signs people make when communicating in sign language. We can even ‘feel’ language when it is written in Braille. Despite all these traces of language, the source of language is not to be found in any of these places. Robinson Crusoe found a single footprint in the sand which told him he was not the only human being on the island: the human being, Man Friday, was of course already elsewhere. In the same way, the sounds and signs we call language are traces of language and evidence of the existence of something that is actually somewhere else.
Language(s) in the mind.
The best answer to the question posed above is as follows: language is to be found in the head, in the brain or, in the context of the Framework, in the mind of every individual human being that uses it. The remarkable thing is, that even though technological advances have allowed us, to a certain extent, to locate what parts of the brain and what types of activity have to do specifically with language, we still cannot point to anywhere and say ‘that’s exactly where language, in all its manifestations, is located or make statements like ‘that’s exactly where my words are stored’. All we know is that is language is to be found somewhere there in a number of different places and that, in most people, certain key areas in the left hemisphere are involved. And there’s more. With every new individual, every new child, whatever language is being spoken around it is being recreated anew in that child’s head. In this way the language can be both preserved for the next generation re-emerging in successive minds/brains: it can also evolve as changes are gradually introduced with each successive generation.
Many more specific questions to do with language cognition:
- How does ability in a particular language develop in an individual mind?
- How does language ability decline?
- How does more than one language come to share space, remain quite separate or constantly compete in the minds of individuals at different stages in their lives?
- How much of what we learn and use can we actually become consciously aware of?
- How does human language differ from the way other animals communicate?
- What makes language cognition different from other types of cognition?
Note that none of these questions relate to how you should teach language, a topic that is outside the scope of the framework.
The modular perspective.
In line with many approaches within cognitive science, the MCF adopts a modular perspective, seeing the mind as an ongoing collaboration between various specialist systems in the mind coping with all the different tasks life asks us to perform at any given moment. There are many types of modularity, some of them markedly different from the well known type proposed by Jerry Fodor in the 1980s. One of the specialisations the central component of our linguistic ability. In the MCF, the central component is actually composed of two separate systems (as indicated earlier) and neither of them have to do with meaning. Although language ability covers much more than what lies within this central component, it is the special aspects of this ability handled by the two linguistic systems that marks us out as different from even our closest and most intelligent fellow primates with who we share almost all of our DNA. Although the focus of the MCF is on all the various mental systems that are associated with the use and development (growth) of language, this modular perspective on human linguistic ability has many implications for the way the mind works in general.
What is a linguistic representation?
It is crucial for this enterprise that the framework is compatible with research findings in a range of different disciplines. Cognition, including language cognition is studied in many different ways, with a variety of theoretical viewpoints, research methods and technical terms. For example, within linguistics (in its broadest sense), it is convenient to study the representational system and the on-line processing system quite separately. One might imagine, a collaboration between two sets of researchers each working separately with their own theories, concepts and and technical terms. One group of researchers develop (their version of) the various abstract linguistic categories and the principled relationships between them without any thought at all for how they might be used millisecond by millisecond in real time. The second group, who do work on real time language processing but do not bother with complex theories of language structure, take these categories and relationships unchanged and simply slot them in their particular real time processing architecture. It would be surprising if this would work without serious attempts to adapt and integrate the research findings of the two groups. In line with the the approach developed by Ray Jackendoff, the MCF adopts a position whereby theories about representations and theories about processing can be developed alongside one another, i.e.within the same framework. A linguistic ‘representation’ is consequently a structure defined both in terms of its linguistic properties but also in in terms of how it operates in real-time processing. Theoretical linguistics, psycholinguistics and bi-/multilingual research provide sources for the MCF as an integrated approach to language cognition. The MCF builds on and interconnects selected theoretical claims and findings in these different research fields about issues such as emotions, attention, consciousness and the nature of human linguistic ability.
Processing and storage.
It is impossible to give an account of the human mind without ascribing a central role for language. The importance of language is not always acknowledged in the psychology literature, perhaps because it is not regularly seen as reflecting a special kind of cognition. Nevertheless, language is a crucial defining of what makes us human. The linguist, John Lyons used the term homo loquens (‘speaking man’) to underscore this point. Researchers across a variety of domains within cognitive science, especially linguists, psychologists and neuroscientists have their own particular perspective on human language. For example, we can attempt to describe and explain how and where the physical brain stores and processes language(s). The approach presented on this website is at another, albeit related level of description: it is about the functional architecture of the mind in psycholinguistic terms but it is also very much about processing and storage.
Language acquisition as “the lingering effect of processing”.
Part of the MCF account, as first developed in Truscott & Sharwood Smith 2004 , involves the claim that the acquired (and attrited) linguistic structures are natural by-products of on-line processing: ‘acquisitional mechanisms’ are therefore embodied in the parser (Acquisition by Processing Theory APT).
APT works across the board, i,e, for all types of cognition apart from language. Processing concepts (‘activation levels’, ‘competition’, etc) as exploited by other non-modular, domain-specific approaches are realigned to explain how new representational systems emerge within some form of generative linguistic perspective. The basic idea of APT is shared by other, quite different approaches. However, in the Modular Cognition Framework and therefore in MOGUL project publications and presentations, it operates within strict constraints.
Language attrition as “the lingering effect of absence of processing”?
The framework account of acquisition expressed by the APT may also be used to help in current explanations of language attrition (popularly thought of as ‘loss’ or ‘forgetting’). The decreased exposure to and use of any language will reduce the accessibility of the relevant representations. This alone would affect language performance but attrition seldom happens in a linguistic vacuum: the loss of accessibility also makes more likely the possibility of representations and representational networks associated with other languages known to the individual outcompeting the attrited language. It will also be influenced by the shifting values that the individual places on the languages concerned. This will eventually have outwardly observable effects on performance in the attrited language. In other wordsm a delcine in frequency of use cannot be the sole explanation for what happens when languages are ‘attrited’. The effects may be very complex in character and are the focus of investigators specialising in language attrition research.
MCF-related publications.
The first detailed account is to be found what was also the very first publication introducing the framework, i.e. Truscott and Sharwood Smith (2004) in Bilingualism: Language and Cognition 7, aspects of which are taken up and developed in numerous publications and presentations listed elsewhere on this site). A much more extensive treatment in book form appeared in early 2014 (The multilingual mind: a modular processing perspective. Cambridge: Cambridge University Press). This laid the groundwork for the framework that would come to be called the MCF. Other books followed, the first in line being Truscott. Consciousness and second language learning in 2015, then Sharwood Smith, M. Introducing language and cognition in 2017 and, in 2019, Truscott & Sharwood Smith The internal context of language processing (see the publications page full more details)..
Comparisons and reinterpretations?
Since the MCF is a theoretical framework and not a theory a such , it is difficult to compare it straightforwardly to anything except another framework or ‘research programme’ with a similarly wide scope. It has been set up against a background of many different theories and hypotheses concerning language in the mind, and the mind in general. What follows, then, is more a comparison with some other approaches, some specific, some general. relating to topics that have been covering thus far, i.e. directly or indirectly related to language cognition such as language acquisition and attrition and multilingualism generally. It is useful to look at a number of these approaches from an MCF perspective both ones that are very specific in their focus of interest as well as more wide-ranging ones. It will help to situate the framework and its current implementations within a wider theoretical context and indicate what is shares and what is not shared with approaches to cognition and language in particular. This include short references to the Competition Model, (Radical) Connectionism, Dynamic(al) Systems Theory and Emergentism, all of which are related to one another, Biolinguistics and the Minimalist Program (see, for example: Sharwood Smith 2019). It will also include various recent hypotheses advanced in the language acquisition literature within a generative linguistic perspective such as the Bottleneck Hypothesis, the Feature Reassembly Hypothesis, the Interface Hypothesis and the Interpretability Hypothesis as well as VanPatten’s Input Processing theory. Also worth mentioning is the relationship of the MCF to earlier theories about how second language acquisition takes place in order to situate in the framework various familiar notions such as explicit versus implicit knowledge and the role of consciousness in linguistic development. Much more detailed discussion can be found in the various publications listed in the MCF bibliography. Note that the MCF in its MOGUL guise was not initially conceived as a contribution to language instruction although it certainly does have implications for guided language learning as also noted by others (see in particular publications by Melinda Whong, e.g. Whong 2011).
The Modular Cognition Framework, Biolinguistics and the Minimalist Program (Chomsky 2004, Jackendoff 2011)
Biolinguistics is an interdisciplinary area viewing language as part of the study of human biology. Although its definition does not require adherence to the tenets of the Minimalist Program, at the moment many of the researchers who regard themselves as biolinguists do adopt this perspective when studying language, its acquisition and its evolution. There is good reason to suppose, however, that the MCF is completely compatible with the biolinguistic enterprise as are Jackendoff’s views on the language faculty which have influenced research in the MOGUL project. As stated elsewhere, one major advantage of the Jackendoff approach is to facilitate an approach where representational issues and processing issues can be discussed within a common framework. An orthodox Chomskyan approach as currently implemented by those working within the Minimalist Program views language from a strictly representational angle at level of abstraction that has its own advantages but requires compatible implementations in real time to be devised that cannot straightforwardly incorporate formalisms that bear no relation to real time processes. This remains true even though metaphors of space and time (such as movement, checking etc.) are part and parcel of the minimalist terminological repertoire.
The Modular Cognition Framework & the Competition Model (Bates & MacWhinney 1985)
The term competition is sometimes identified with highly non-modular processing theories hypothesizing unrestricted information flow. Brian MacWhinney’s Competition Model (as developed with Elizabeth Bates) is an example of this. The MCF, however, is highly modular so the treatment of competition is constrained by the particular module (processing unit) and its unique properties. This means that, say, conceptual representations will compete with other conceptual representations in conceptual working memory but never with representations outside the conceptual system, which they may or may be associated with. This model should not be confused with the one described below.
The Modular Cognition Framework & Felix’s Competition Model (Felix 1987, 1988)
Sascha Felix developed his own ideas about competition but within a modular, generative linguistic perspective and assuming that older language learners still have access to the same innate principles that guided their first language acquisition. ‘Competition’ here means a competition between the innate language system constrained by universal grammar on the one hand and general problem-solving mechanisms that come to maturity later in life on the other hand. His idea was that this competition characterises development in older (adolescent adult) language learners especially those undergoing formal instruction. Second language acquisition performance data would expect to show the effects of this competition and would therefore not always be like the data obtained from first language acquirers. Again this use of the term ‘competition’ that is not the same as in the MCF although Felix’s ideas could be reformulated quite easily within an MCF perspective.
The Modular Cognition Framework & Connectionism
‘Connectionism’ is a term that can be interpreted in different ways. The most common interpretation is a radical connectionism of the sort pioneered by McClelland and Rumelhart (1981) which does away with symbolic representations and assumes unrestricted information flow within a system the properties of which are supposed to mimic those of (biological) neural networks. Viewed from an MCF perspective, this strictly non-modular approach is neither what brain research suggests nor is it how psychological function can be best explained. In radical connectionism, the networks are composed of simple units. Each unit has a particular activation potential. When a unit is activated it triggers activation in units that are connected to it within the network. This interactive process is called spreading activation. Learning amounts to changes in the system which can be accounted for by changes in activation patterns or connection strength and without recourse to symbolic representations. To the extent that the MCF is based on networks and also involves spreading activation, it can be called ‘connectionist’ in some sense but not in the sense described above and as already mentioned above, in the first section on the Competition Model. This is because, unlike radical connectionist approaches, the MCF does deal with symbolic representations (see the glossary entry for ‘representation’) and secondly because the MCF is a modular system with interfaces between systems that restrict the way in which activation is allowed to spread.
The Modular Cognition Framework and Emergentism.
This is a term that describes the development of complex systems. New patterns emerge from old ones. The new patterns are different from the patterns from which they have emerged. The process also results in a greater number of patterns in total. At first glance, this might seem an appropriate characterisation of growth in the MCF. Could the MCF be called, in any sense, ’emergentist’? ( see this response) Just as there are different kinds of connectionism, there are also different kinds of emergentism. Closest to MOGUL is perhaps the type of emergentism proposed by William O’Grady (O’Grady 2008). Compared with MOGUL’s APT (‘Acquisition by Processing Theory’: see the entry for APT in the glossary), O’Grady’s version of an acquisition device involves the interaction of (innate) general cognitive principles interacting with the linguistic data supplied by the environment so that mental grammars emerge without the need to posit special grammatical principles unique to language acquisition . The claim behind APT also means that there is no need for a special acquisition device for language and that grammars emerge as a result of on-line processing. Crucially however, the framework architecture assumes language-specific principles that shape and constrain the emerging linguistic system whereas O’Grady rejects this idea along with other emergentists like MacWhinney (see above) who, unlike O’Grady, also reject symbolic representational accounts. In the MCF, acts of linguistic processing always, and to a greater of lesser degree impact on the processing units involved. These processing units have their own operating principles (special phonological principles and special morphosyntactic principles) and so the phonological and syntactic shape of ’emerging’ grammatical systems are not exclusively determined by general cognitive principles or by any other kind of general principle (see this publicatiion for further discussion) . Put another way, the MCF assumes some version of a ‘language faculty’.
The Modular Cognition Framework and Dynamical Systems Theory.
There is no doubt that the human mind is a complex dynamic system. That much is surely uncontroversial. Whether Dynamical Systems Theory (DST) is a way of capturing any aspects of that complexity remains to be seen because its current implementations to our knowledge do not seem sufficiently developed to do this fruitfully: in any case it is still possible to say that it has promise in eventually contributing to comprehensive view of how the mind works and it may also shed light on how the human mind develops in evolutionary time. The Modular Cognition Framework and its current instantiation for explaining language cognition (MOGUL) places the emphasis on the modular nature of cognitive architecture. This imposes a requirement on some form of DST to define and explain dynamical effects in specified places within an otherwise stable, fixed architecture rather than let DST work as an explanation of all aspects of cognition. Thus, for example, we expect the resting levels of activation of many or all cognitive structures to be always to some degree active and continually changing rather than literally resting and we also expect what we call activated structures, i.e. particular types of structures the working memory of any one of the minds modular systems to similarly have fluctuating levels during online processing in response to ever-changing internal and external circumstances. What we do not expect is alterations in the number and basic functioning of the minds modular systems, e.g.. those managing vision, affect, syntax, meaning and so forth. Nor do we expect fluctuation in the nature of the interfaces by which they collaborate via their respective interfaces. The basic mechanisms stay the same. In other words, important aspects of cognition are eminently non-dynamical: any notions of self-organisation and emergence and therefore constrained by this fixed aspect of MCF architecture.
(to be continued)
Mike Sharwood Smith, Edinburgh (Scotland) & John Truscott, Hsinchu (Taiwan)