The excessively dominating role of language on mind theories

Introduction

The concept of a brain-mind system where specific part of the information collected by the monitoring devices are crunched by ad hoc modules of the mind-brain has been put forwards by a number of important researchers operating in many sectors ranging from philosophy to experimental and theoretical neurobiology and psychology. Such a trend is based on…..as well as relevant experimental discoveries which have put into evidence a multi-scale spatial structure of the brain tissue as well as of many genetic and chemical processes which share the responsibility of actualising such a complex spatial organisation. Generic modularity as on instrument for evolvability has also contributed to the increasing interests forr mind-brain modularity. Other important discoveries on the environmental information accessing only specific area of the brain by independent specific circuits have further given force to the modularity hypothesis.

Theories interpreting these novel evidences about the modular structure of the brain have appeared in order to propose new visions about the local and global mechanisms capable of giving some rational about the coordination of such hypothetical local modular information elaboration and the observed high level psychological performance of mind. In facts, if the decomposition of large complex task into a number of specialised simpler ones looks particularly attractive in view of a quite universal engineering principle stating that it is highly improbable that a machine can perform equally well two or more tasks, the same decomposition ask for the identification of mechanisms responsible for coordinating a lot of partial outputs into sensible global actions. A realistic modular theory of brain-mind must be capable to identify plausible neuro-physiological structure capable of contributing to the coordination of partial outputs.

Evolution enters as well the modularity problem in its own right. Complex human brain is the result of a long complex evolutionary history which, if the modular model is correct, must be somehow responsible for such an evolutionary strategy. As far as anatomical modules are considered, the genetic has been shown to be the responsible for modules appearance [ ]. Unfortunately, as we will see below, functional modules, which correspond to the above defined modules, responsible for specialised sub-computation, do not correspond in many case to anatomical- genetically determined modules[].

Philosophers, analysing the features of mind performance with particular attention to the language, have introduced other, more abstract concepts of modules and have risen the problem on the existence of a language of thought, which we do not discuss at this point, but which inevitably interact with the theories of mind-brain modularity. The situation is, as the reader can understand, rather complex. Each of the alternative theories evolve autonomously by the work of their enthusiasts, and tends to “explain everything” making use of its own basic concepts and debates are carried out apparently without major results.

The dominating role of language: The connection between language and brain computation.

Language uttering can be classified as a large set of specific actions. On the other hand the existence of human languages, by its grammatical and syntactic structure, has deeply influenced our conceptions about our concept of what thought is, and has created a strong a tendency to consider it something different from somatic actions which should respect completely different constraints.

Wittgenstein was compelled, revising the ideas expressed in its “Tractatus” to introduce the well known concept of linguistic game which approximately captures a scenario of possible proto-language appearance in species with significant sound utterance capabilities. Language can well have emerged as a linguistic game, useful for group coordination and its use extension can have imposed those structural features that we now call grammar and syntax under the constraint, between others, of its serial form. We will reconsider the problem in detail. For the moment it is enough to put into evidence how the rigid division between language utterance and other somatic actions can be easily put into serious discussion from many points of view.

Language has appeared very late in human species evolution (some hundreds thousand years ago at best). It seems making sense to ask why and how it appeared, as it is a particular competence which characterises, as far as we know, just humans.

The question looks sensible at first glance, but answering it can be very complex and even impossible due to the lack of experimental data.

One can nevertheless envisage different possible evolutionary scenarios which I do not intend to discuss here, while I prefer to concentrate on some important hypothesis regarding the evolutionary interconnection between language and thought. The phonetic complexity of the human languages evidently need many specific physiological characteristics of the speakers. It is however conceivable that the possibility of proto-languages use can emerge because of the appearance of evolutionary indifferent somatic features, possibly connected with intraspecific competition. Some form of simple communication can have appeared as a competitive advantage in hunting or other activities. These are already investigated scenarios.

The basic question, of great importance for mind- brain theory regards rather the often proposed (explicitly or implicitly) hypothesis that the language appearance was also the consequence of some pre-existing “talent” of ancient human beings for the use of some form of abstract conceptualisation. If this was the case, language can be seen as a tool for expressing such a talent.

An alternative scenario is that where the appearance of the language was the responsible for the access of human beings to abstract conceptualisation.

This is not a chicken and the egg problem due to the late appearance of language of human species and the evidence of the important role played by catastrophic dynamics in evolution. Catastrophic changes are characterised by discontinuous rather than continuous features, and therefore it would not be scandalous to imagine that abstract conceptual thinking evolved rapidly making use of some simple language originally emerged for simple group coordination.

Was the language the occasion for the manifestation of a pre-existing human conceptualisation talent, or rather was it the new tool which made abstract thinking possible?

We are so used to language that our intuition cannot help us very much in answering this question. In any case one can observe our tendency to superpose our point of view when, as an example, we call the capability of a neural network to subdivide the set of inputs into two classes, as a generalisation competence. Actually its generalisation competence is our description of its performance; the network simply performs a computation.

Another simpler way to restate the above question is: does competence to successfully survive in a given environment depends on the capability of a knowledge system to incorporate in its hardware or software some basic features of abstract thinking as categories, classification and simple logical operations?

Brooks’ seminal works suggest this is not the case, and for sure there does not exist any theorem stating that the answer is “yes”. Competence, after all, is just undertaking an approximately correct action in a given environment, and not classifying correctly the environment and deducing the right action, as this last form is the expression of a linguistic formal-logic approach.

We need to take a decision whether intelligence is a medal we give to competent beings or just to those whose competence is the result of the implementation of formal-logic procedures we can express making use of our language; that is intelligent behaviour can exist only by mirroring, possibly by a very complex, hard to detect, procedures, our formal-logic languages production rules.

This is why our interpretation of the emergence of language is so important as, if we consider it just an occasion for the emergence of a pre-existing abstraction talent, we will be lead to conclude that many abstraction processes are the unavoidable feature of our mind. If we, on the contrary, believe that the language was the tool by which the human species has invented abstract processes, any form of abstraction process must necessarily be attributed to mind. Abstraction could be defined in such a case as a projection on some linguistic space and would loose its ontological pretensions. The choice of a scenario where abstraction’s competence has co-evolved with language would lead to the same conclusions.

If abstract thinking is a linguistic competence, it is a linguistic by-product, of paramount importance for the human species, but still a by-product. Logic is, in this perspective, a linguistic product as well, and its rules regards the linguistic world’s description, and therefore do not necessarily capture the essentials of the competence oriented control system machinery of the living beings, independently of their complexity.

Linguistic relativity theory LRT [] has raised this problem since many years and is slowly changing our opinions about the relationship between language and thought. It seems nevertheless that the final step towards the investigation of a more radical form of LRT is in many cases avoided for different reasons. The most important of them has to be identified in the habit of investigating animals’ intelligence by a priori adopting our conceptualisation system as the only reference frame to evaluate it.

What I mean here is that there exist an historical feedback based on the conviction that the Mentalese hypothesis is correct. This feedback has been so strong that any observed behaviour (maybe even that of worms!) has been interpreted , as a matter of fact, in terms of conceptualisation capability. The, evidently not proved, hypothesis underlying this move, was that any competent behaviour could not only be represented or described in terms of abstraction competence, but actually consists of such competence. This has seriously affected investigations on the brain computational nature, which was interpreted as something necessarily mirroring abstract classification underlying the human language. Mentalese hypothesis after all does not favour the ontology economy principle, and it is strongly associated with the idea that brain computation, however implemented, should mirror logical production rules, class theory and many other fundamentals of formal systems which have been discovered during the last centuries.

It seems that language implies a complete different elaboration of input data.

Internal Representations and LOTH

The term “internal representation” is the responsible for many ongoing debates in the theory of mind. In facts the term can be interpreted in many different ways. One could simply call “representation” the totality of changes induced in the brain or in the whole nervous system by the perception of a state of affairs in the external world. Even in this very economic interpretation of the term, many problems can be raised when different time scales are considered. One can refer to short time scales and try to capture the local in time representation, or rather consider longer time scales and discuss the long term effects on the perception of future experiences. In any case one could hope to capture such kind of internal representation by representing it as a change in the brain very complex and still unknown phase state. One can as well then raise the problem about the spatial distribution of induce phase state change and compare sparse memory models with highly localise ones. In any case the discussion would regard the brain as a physical system and, independently on the real possibility of reaching a sensible description of such state changes, it would be based on a scientifically sound concept. An example of such an approach to brain’s toy models is the research on the changes induced in the internal units of a trained feed-forward artificial neural net correctly implementing a yes-no classification of a data set. In most cases the attempt has been that of identifying neurons or groups of neurons whose state can be associated with some basic feature of the specific element of the data set under consideration. Such kind of investigations are evidently scientifically correct as far as they are not influenced by some underlying, not yet proved, implicit assumption. In facts the debate about internal representation regards essentially the internal organisation of the physical changes induced by the observation of an external state of affairs, and is strongly associated with the interpretation of the role played by the language.

Some researchers are convinced that the internal organisation of the state change must be such to incorporate the essentials of occidental languages and, in particular, their underlying logical classification structure. This conviction implies that only those organisations which make available the elements of the basic elements of the linguistic description of a state of affairs are considered as viable alternatives.

An alternative view is of course that where the linguistic description of a state of affairs is considered as a specific competence oriented to inter-individual information exchange. In this last case the linguistic translation of the brain changes induced by some external observation is considered as an independent problem, and the linguistic description does not need being mirrored by the brain’s internal state change’s organisation. In facts the first hypothesis considered above is totally equivalent with the idea that the language appearance was also the consequence of some pre-existing “talent” of ancient human beings for the use of some form of abstract conceptualisation corresponding to occidental languages.

Let us consider the Stanford Encyclopaedia of Philosophy’s definition of Mentalese’s theory:

“ The Language of Thought Hypothesis (LOTH) postulates that thought and thinking take place in a mental language. This language consists of a system of representations that is physically realized in the brain of thinkers and has a combinatorial syntax (and semantics) such that operations on representations are causally sensitive only to the syntactic properties of representations. According to LOTH, thought is, roughly, the tokening of a representation that has a syntactic (constituent) structure with an appropriate semantics. Thinking thus consists in syntactic operations defined over such representations. Most of the arguments for LOTH derive their strength from their ability to explain certain empirical phenomena like productivity and systematicity of thought and thinking.”

LOTH is strictly based on the hypothesis that the structure of thought is linguistic in nature; mental representations are like sentences in a language in that they have a syntactically and semantically regulated structure. Actually the acceptation of LOTH does not implies the belief in the unicity of representational system realized in the brain, as it does not claim that all mental representations are language-like. What LOTH claims is that there is no adequate way alternative representation systems can be involved in judgments about the truth of propositions. Some basic problems of LOTH reside, quite obviously in facts, in the mechanisms of experiences translation into prepositional form, while the last basic claim can be interpreted as a quite tautological one, as we do not have any firm idea of truth, but rather some knowledge about truth conservation in linguistic systems. So, after all, LOTH is, at least in revised Fodor’s version, more the expression of some unavoidable brain processes which must be carried out by the brain, once the experiences have been translated into a linguistic utterance. Without such a preliminary translation, no LOTH is needed except in the case where the language is considered as the emergence of a pre-existing organisation of the pre-linguistic mind. After all many mind scientists are more o less explicitly convinced that the unicity of the human species is the evolution towards that form of abstract pre-linguistic reasoning. But this is a belief rather than a proved fact. There are few doubts that LOTH, again in the revised Fodor’s version, must be the case nowadays but, in my view, this fact does not imply that the belief of a previous human evolution towards abstract pre-linguistic reasoning is correct.

Naming , abstraction and nomadism

Many empirical observations of what we call “primitive languages” suggest that the major difference between them and European languages is their of excess of names for situations of some relevance in their environment. This feature seem to characterise most of the “primitive languages”, being of course “primitivity” a totally relative concept. The process of naming is then a common feature of languages, which characterise them, the “intensity” of such process being a base for classification. The simple process of associating a sound or a sign to a given state of facts can be reasonably emerged from the capability of uttering a significant repertoire of different sounds. In this perspective the number of states of affairs which can be named is approximately proportional to the cardinality of distinguishable sound series. Making use of a finite number of “elementary sounds” , say N, one can invent a large number of names of length M, more precisely N^M, which is normally a very large number. With 10 elementary sounds one can produce 10¹⁰ different nouns of length 10. If we consider also shorter names, that number retains the same, very large order of magnitude. Names can therefore signal a lot of states of affairs. The naming process represents some important cognitive advantage as, for example, recalling a state of affairs independently of its actual presence. If the names are randomly chosen within the potentially available reservoir defined just by N and M, you have no shortcuts to remember them. You need however to collect a reasonable amount ( bits) of information to retrieve one of them, but you must know all of them. If you consider a population living in a stationary persistent environment, the strategy of naming all the relevant states of affairs ( for example all the relevant geographical sites) can be somehow justified. If however the population rapidly moves across different environments, the same strategy turns out being rather inefficient: giving a proper name a to mountain which in a couples of week will have disappeared from our environment is clearly a very bad memory investment. Perpetually forgetting names an learning other ones would most probably exhaust the names reservoir together with the memory of the members. Then another strategy must be identified.

A possibility is keeping the names and using them by some form of analogy between the previous state of affairs and the new one. This will imply a reasonable effort. At the same time however it will represents some important move towards abstract classification. In other words it will imply the existence of some common features in the information collected from the environment. The communication will result statement similar to the following “there is a new states of affairs which is similar to that named “X””, where X is an already stored name of the language. On the other hand if some form of interpersonal information exchange must persist within a nomad population, it hard to see other viable strategies. Once such a strategy is possibly a mature one, more developed form of abstraction can be envisaged. The basic feature of the process remains however that of keeping the same name for states of affairs which share similar features in the collected information, as this move is inevitably associated with a number of possible interesting problems. Consider again a state of affairs corresponding to some geographical location and its particular attributes (good hunting place, on the top of a high hill, for a certain prey species). It is quite evident that the probability of finding somewhere else a completely similar state of affair is quite low.

The adoption of the above identified strategy will exercise a pressure towards the identification of the feature/s the previous state of affairs shares with the new one. This can be obtained by a new social agreement of the use of the name. The hunting of a newly identified kind of animal will pose a quite similar problem. What we suggest here is that nomad populations could well have faced a high pressure towards a more abstract use of nouns due to the non-static nature of their environment. Spatial directions for example must be named following very different logics by populations living in a fixed defined area and the nomad ones.

An elementary discussion about serious constraints on realistic knowledge evolution

There exist many fundamental researches on learning and, in particular on machine learning. We do not intend to discuss them here. We will introduce here a different problem which can be named “knowledge evolution while surviving”. For a real, not virtual, living species the improvement of knowledge poses a serious problem as it must be put into operation while safeguarding previously acquired competences. In other words knowledge must be safeguarded while acquiring new one. This is not problematic within human culture, but it becomes a very serious constraint in machine learning. This is particularly true when the possible evolution of the monitoring system is contemporarily taking places. An increasing amount of information about the environment is made available, which can be of some use for improving the living system behaviour in a spectrum of situations of undefined wideness. Unfortunately the knowledge system does not a priori knows the optimal use of such newly available information. In any case what cannot be done is a complete retuning of the knowledge system on the basis of the enlarged information flux. This choice would in most cases lead to the extinction. A realistic model of knowledge evolution must take into account such a fundamental constraint.

Modularity of the brain

It is a well known fact that the term module is used in many different senses. A quite clear concept of modularity can be obtained by referring directly to the connection topology in a body of neurons. We can, in this case define a module as an ensemble of strongly interconnected elementary computing units showing a lower level of connectivity with the rest of the body. Different efficient measures of such kind of modularity can be introduced. Postponing for the moment a comparison between different proposal, we can clarify what we mean when adopting such a paradigm. We fundamentally consider each module as a unit capable of implementing some complex computation in a state of semi-isolation, receiving its input via some of the connections with the rest of the body and transmitting its outputs again to the rest of the body by making use of some other connections with it. This model can be further extended by introducing further details on the connectivity topology within a body of biological neurons. In any case we more or less explicitly identify each module as a semi-isolated computing device and its level of internal connectivity as the feature that makes of it such a localised computing unit. We can imagine to partition the whole body in a number of coexisting modules, or a series of nested modules. The obvious extension of such a simple conceptual model is in fact that of a module of modules, a module of modules of modules, and so on. Such a model can also be considered as some “consequence” of the complex structure of various spatial scales modularity observed in the human brain. Intuitively we attribute to the wider modules’ module a computational task which can be performed by a network of units with a higher level of expressivity than a single neuron and, more or less explicitly deduce that it can be the responsible for higher level computational tasks. Following commonly accepted theories, this kind of physical modularity is the result of genetic processes with the possible intermediation of adhesive or anti-adhesive substances.

We are nevertheless aware of a number of simulation where ANN, during the training phase, have sown to autonomously organise themselves in a modular way at least for certain classes of tasks and depending on their learning paradigm. In other words simulations show that when the right neural model, learning algorithm, and task are chose, ANN can well autonomously tend to modularisation. This subject has been extensively studied by Edelman, Tononi and Sporns in a series of fundamental papers [] specifically intended to intervene with what they call “a long-standing controversy in neuroscience has set localisationist views of brain function against holistic view. Making use of measurement of new indexes based on information theory, they show how high level performances of ANNs is obtained by contemporary strong level local connectivity in what they call “neuronal groups”, and patchiness in the connectivity among neural groups characterised by prevalent reciprocal connections. It is impossible here to discuss in details all the above cited works and all the details of their model (ETSM). What seems more interesting for the present discussion is, together with the evidence of some form of localised modularisation efficiency, is the presence of dynamic re-entrant interactions between specialized groups of neurons which leads to patterns of short-term correlations. In their simulations, the authors notice how neuronal groups can act as population oscillators which can themselves exert effects on other groups giving rise to functional integration of sets of groups which they interpret as representing conjunctive sets of features, and higher-order objects or concepts.

While original anatomical connectivity patterns determine the functional connectivity patterns can exist; in other words, which areas of the brain can influence each other, and to what extent, is largely (although not exclusively) determined by their mutual interconnectivity.

Additional important factors determining the strengths and patterns of functional interactions are the physiological effectiveness of individual pathways, as well as their termination density and pattern. These findings have suggested that functional connectivity can contribute to shaping anatomical connectivity on a developmental time scale and may thus provide an additional set of developmental mechanisms underlying the mapping between classes of functional dynamics and structural

motifs.

Here the concept of local modular units performing specific functional task is again called into play. The nature of the information processing within each module seems not to correspond with Turing computation, rather some form of simpler significant state tuning capable of participating to a coral resonance with other groups and therefore give rise to more diffused and global brain states.

What we would call a Functional module is identified by neurophysiologists in neural circuits composed of nerve cell aggregates, which are connected by fiber tracts in a highly ordered fashion They are called modular as they carry out information processing independently in separate structures, despite some limited flow of information between the circuits.

Each system is typically composed of parts that are derived from several embryonic divisions and consists of several neural circuits. Besides the sensory systems, multi-sensory regions integrate sensory information, and the motor systems generate body responses to environmental or internal stimuli. The advantage of having specialized functional modules is that the neural architecture of each module can be optimally in order to carry out optimally the type of information processing that is required under environmental pressure in each case.

Finally we must cite the Turing morphogenetic effects in a growing tissue as a mechanism responsible for inducing modularity in the brain. After the original works by Turing [], morphogenetic theory has been improved by different authors. Recently Sandberg [] has discussed in great detail how Turing double diffusion model can be responsible for brain physiological modularisation. A detailed discussion on modularisation in a growing biological tissue is found is Koch and Meinhardt []. Moreover evidence is also reported that cell adhesion molecules provide an adhesive code : It is suggested that embryonic modularity is transformed into functional modularity, in part by translating early-generated positional information into an array of adhesive cues, which regulate the binding of functional neural structures distributed across the embryonic modules. It is suggested that The adhesive may be as basic to brain development Similar the Hebbian synapse mechanism as neurons that glue together, wire together. Both processes allow for adaptability of neural connectivity in response to environmental pressure, but possibly at different time scales []

The complexity of functional circuit concept when applied to the brain

The widely used concept of functional module needs some clarification as it can be interpreted in many different ways. Broadly speaking a functional circuit should in principle resemble to a radio set circuit, as, for example, that transforming high frequency signal into low frequency for speakers sonic output. This is a very specific example, but quite easy to understand. A radio receiver, at least an old one works because it has such functional circuit. We know the input, and the output needed, and the circuit has a very well defined function. In this case the function is understood in terms of the input and the output. The same could be said for the amplifier circuit. We understand perfectly the amplification and the frequency conversion functional circuits in our radio set. Radio sets are however very simple machines with respect to the brain. In particular the same circuits act exactly in the same way independently on what is the radio emission about (this is false in modern radio receivers, but we ignore it). Assume we buy a strange machine X which is at the same time a radio receiver and a coffee maker, and where the heat produced by the circuits of the radio set is used to keep the water of the coffee maker warm. With respect of the coffee maker the functionality of all heat producing circuits have the function of keeping the water warm. One can invent probably better examples. In any case it is quite evident that the concept of functionality of a circuit is quite complex when the machine is not a simple one. Engineering correctly suggests that it’s hardly the case that a machine does two things equally well. This does not imply however that a circuit doing is original task cannot be a resource for doing something else quite decently. This suggests that the concept of functional circuit can become more complex once the machine does many things. This is particularly true if we consider an evolving machine, that is a machine which, for some reason, tends to increase the tasks it can implement. This is of course a rather simplified analysis, but it correctly raises the problem about functional circuit concepts, when applied to evolving complex machinery. Of course, if we knew in advance all the tasks the machine had to implement, we could make a detailed project and, most probably, the best global machine would be the sum of perfect machines, one for each task. But this is another story, which does not capture the real nature of the human brain. First off all because a poorly working extra performance could well correspond to a huge survival advantage. After all the same story of the brain is partially at least made of the appearance of new poorly efficient new tasks.

When analysing a working brain we are in a particular bad situation with respect to functional circuits as we are performing some kind of reversal engineering of a machine which was not conceived without any concept of rational design and whose efficiency has been tested, during its evolution, in environments very different from those we know and where a small difference in efficiency could have made a lot of difference independently of its theoretical absolute value, assuming it could be defined. All this to show how the concept of functional circuit (and module) can be seriously put under discussion making use of realistically applied common sense.

The same concept of function is quite poorly defined. Its definition is subjected to a lot of, generally implicit (and therefore often unconscious as well) constraints. The transform of food into energy can be seen as a function and, at the same time, the absorption of certain chemicals by the internal intestine surface can be considered a function. Both choices are reasonable; it depends on the scale of observation. We are forced to go back to the radio set problem; if you know in advance what function you want to implement, you can solve the problem by transforming it into a series of precise sub-functions and create the sub-functional circuits. But none, except possibly divinities, knew in advance what was the function of the brain. Its our everyday experience the discovery that some detail of human anatomy, which was believed to have a function, actually has some 3 other functions as well. Why these observations should not force us to suspect that the same situation holds for the brain as well? Functional modules are a concept that fit very well the activity of smart engineering.

Modularity of Mind

Modularity of mind is not a theory directly referring to the physical structure of the brain. As an abstract theory it needs to be discussed in greater detail to avoid misunderstandings. What looks more preoccupying, at least to us, in such paradigm is the relevant role played by linguistics and logical computation models in its foundation. N. Chomsky was the first proponent of the Universal Grammar Theory (UGT). He argued that all languages have a common or universal core. The general structure and a number of the substantive features of the grammar of all languages overlap; He suggested that all languages have a particular set of syntactic classes such as nouns and verbs, or subject and object .This fact is considered by Chomsky as the evidence that they are determined by the nature of the mental structures and processes which characterise human beings. Chomsky refers to Formal Universals as the form of the rules which will appear in the grammar. Transformational rules, for example, may be required to handle the syntactic component of the grammar satisfactorily. Following UGT the similarities between languages is considered as the consequence of a mind organisation which emerges with the advent of human languages. Chomsky is extremely clear on this point:

“Language is a mirror of mind in a deep and significant sense. It is a product of human intelligence ... By studying the properties of natural languages, their structure, organization, and use, we may hope to learn something about human nature; something significant, if it is true that human cognitive capacity is the truly distinctive and most remarkable characteristic of the species.” (Chomsky, 1975, p. 4)

This is a strong hypothesis which cannot be considered in any sense the unique possible explanation of Chomsky’s observations on human languages. Languages both in spoken and written forms are submitted to a number of constraints (seriality or, in general some form of agreed ordering of sound and symbols); one can well interpret both grammar and syntax as some form of structure of signs and sounds imposed by such constraints. If this is the case grammar and syntax intervene during the evolution of languages and do not precede them. Grammar and syntax have, in this case nothing to do with the way of functioning of mind, which, by the way, has evolved significantly in the absence of any language. UGT has played a major role in the formulation of the Language of Mind Hypothesis. (LOMH). For Chomsky a module is a body of innate knowledge or of innately cognised propositions.

In his 1975 J. Fodor suggested an other hard hypothesis, strongly connected with Chomsky’s UGT, by hypnotising the existence an innate language that is distinct from all spoken languages and is semantically expressively complete. Such innate language (Mentalese) is interpreted as an inner language that contains all of the conceptual resources necessary for any of the propositions that humans can elicit and understand, think or express, the basis of thought and meaning.  After Fodor the brain is not a general purpose hypothesis testing machine, but rather a series of special purpose modules. Possible modules are those for object shapes recognition, for recognising and interpreting language, another for recognising faces, and so on. Fodor’s Modules domain specific, subconscious, automatic, and informationally encapsulated. They have access to some useful information. Finally, in analogy with Chomsky’s theory, they are innate. The information in the modules may include acquired data, but the existence of the modules is actually innate. Modules quickly identify some salient information out of their input signal, send their analysis to a central processors that draw conclusions on the basis of (potentially) disparate sources of information, including analysed signals from the modules, and (sometimes) raw signals from the transducers that are not fit to be analysed by any module. In Fodor’s vision of mind, as well as in Chomsky’s UGT there does not exists any explicit reference to computation. The two concepts of module can be somehow identified by admitting that Fodor’s structure is responsible for processing the innate bodies of knowledge identified by Chomsky.

The most recent mind theory, which incorporate its own concept of Module, is Evolutionary Psychology (EP) The basic reflection on which EP is funded is a well known basic engineering principle which states that the same machine is rarely capable of solving two different problems equally well.

EP founders concluded by analogy that our minds must consist of a large number of circuits that are functionally specialized as they are capable of solving a very large spectrum of problems equally well. Differently from Chomsky and Fodor EP embodies the different functions into these specialized circuits, which can be somehow interpreted as a mini-computer each dedicated to solving one problem. These specialised circuits are referred as modules in EP. One can notice how EP modules can be easily identified with the physical modules one can observe and define in a body of biological neurons. What EP adds is the functional nature of the computations carried out by such physical modules. EP cannot avoid, with some analogy with Fodor’s model, need for circuits whose design is specialized for integrating the output of all these dedicated mini-computers to produce behavior. The brain is then represented as a collection of dedicated mini-computers whose operations are functionally integrated to produce behaviour.

Recognising that Biological machines are calibrated to the environments in which they evolved, and that they embody information about the properties of worlds inhabited during their evolution, EP suggests that the modules cannot be simple multi-purpose logical machines, rather they must be equipped with crib sheets containing basic information about the particular problems they must solve. Moreover EP does not fixes a specific kind of computation valid for all modules. . Some of them can well embody rational methods, but others may implement different inference procedures that respond not to logical form. Fodor has discussed and criticised EP theory in a well known book [ ].

The situation is, as we have seen rather confused, and shows that the term “module” must be used with a lot of attention.

Some of the problems emerging seem be clearly stated. One of the basic intuition involved more or less explicitly in most of the theories refers on the identification of a strongly interconnected cluster of neural networks as semi-isolated and semi-autonomous devices, implementing their specific computations on more or less specialised inputs and passing their outputs to the rest of the body for further elaboration. This problem is particularly critical for EP but, making use of deeper rational analysis is implicitly involved also in the other theories we briefly sketched above. While a high level of connectivity can easily make available an expressivity similar to Turing machines [], the role of the connections with the rest of the nervous tissue body looks problematic. We can well imagine that some of them function as inputs and make available the strings to be elaborated, while the rest of them act as output passing to the rest of the networks the final results of the computations carried out by the module. These conditions implies a strict organisation of the directionality of the connections with the exterior of the modules as well as their in time tuning. If the above conditions are not satisfied and the module is affected by the exterior while computing, the analogy with Turing Computations, at least in its classical form, fades inevitably away. So EP’s specialised circuits can be identifies with TM after having introduced further strong hypotheses on the dynamics of the interconnections with the rest of the nervous system. Actually the founders of EP consider also the possibility of non logical information extraction and do not seem to insist too much on the TM nature of specialised circuit information crunching.

While Chomsky’s and Fodor’s modules are in many senses abstract units whose embodiment is not specifically identified by the authors, EP modules explicitly refers to brain’s connection topology. Such modules can resemble from some point of views those discovered and discussed by Edelman, Tononi and Sporns. The need of information exchange between modules is in both case a necessary condition for letting the modular elaborations be made available to the rest of the brain in order to allow higher order performance. EP is based on some basic reflections about a machine capability to perform more than one specific task correctly, ETS on experimental evidences and the associated theoretical interpretations. On the other hand The neuro- anatomists describe of functional modules as very complex structures, occupying different portions of embryonic modules.