Generative Oscillation

A Cognitive Model for the Emergence of Language

Research Material for a PhD

DRAFT COPY ONLY

NOT FOR PRINT PUBLICATION

NOT FOR USE OR QUOTATION IN ANY FORM WITHOUT WRITTEN PERMISSION OF THE AUTHOR

Thorold May

University of Melbourne

Generative Oscillation -- A Cognitive Model for the Emergence of Language
© copyright Thorold May 1994, 2004 All Rights Reserved
pre-published by The Plain & Fancy Language Company ACN 1116240S Melbourne, Australia

note 1: This document is about 50,000 words. It was work done towards a PhD in Cognitive Linguistics at the University of Melbourne (part-time) between mid-1990 and 1994. The research was put aside for some time but is now continuing (2004). The original thesis topic was Formulas, Repetition, Substitution & Ellipsis in Discourse Organization: the Limits of Creativity in Language. Research led to the present model.

note 2: The GO model proposes a co-generative view of the emergence of language which has, as far as I know, not been suggestd before in linguistics. Some conventional linguists may find it a bit "off the wall" (one reason for my decision to withdraw from the PhD program), but the paradigm could well be quite productive). My present views (2004) about some elements of this theory have evolved somewhat. For example, recent work on neural networks leads me to a much more nuranced view of "encapsulation" as discussed here. The terminology applying to what the GO model explores is also likely to be modified substantially in the future.

note 3: The early chapters on the GO model are heavily influenced by one particular book: Varela, Thompson & Rosch 1991 The Embodied Mind: Cognitive Science & Human Experience, which sparked many of my own ideas. The final work will obviously require greater referencing depth. Later, more conventionally linguistic chapters are not yet fully integrated with GO, and some are little more than outlines.

The Generative Oscillation Model

Thorold May

Introduction

The Generative Oscillation Model explores a dynamic tension between forces which shape the generation of language, and which ultimately control the evolution of linguistic systems themselves. Some of these forces are found in the internal dynamic of language systems themselves, and some in the coevolutionary accomodation of linguistic systems with external environments. The totality is held to be subject to laws which control the emergence of order in complex dynamic systems : the so-called "order at the edge of chaos" postulated by Complexity Theory <1>.

The crux of enquiry in this thesis is the question of why linguistic meanings are encapsulated <2> semi-permanently at word level but, in general, not in more extended strings of text. It examines the properties of that restricted class of formulaic phrases which have acquired some of the characteristics of permanent encapsulation, and considers what distinguishes such phrases on the one hand from single words, and on the other hand from strings generated uniquely in particular contexts.

Early chapters develop the idea that language is an independent, organic growth, but possessing a propensity to correlate, chameleon-like, with features in other environments (thought, perception). This process is seen to be not inherently representational, hence not strictly constrained by those other environments, but rather as fairly indiscriminate in its matching tendencies (although outcomes are patterned). Processes at this level are taken to be largely autonomic and quite unconscious. The internal dynamic of the system encourages a proliferation of language which obeys phrase structure rules but which has only accidental social functions. It is almost "weed-like"<3>.

The countervailing force to organic growth in language is taken to be the conscious manipulation which can come with topic, goal and social context. These forces move intermittently to harness language into deliberately representational patterns. Language at this level is patterned by many hierarchies of normative rules. Such rules reflect not only conventional relationships between the systems of language and other systems, but also mimic contingent and intrinsic properties of those other systems. Thus a formal context will be mimicked by formal language, and so on.

There are properties inherent in the organic growth of language,particularly repetition, which encourage extended formulaic structure. Repetition, by enhancing neuronal connections, is a vehicle for the encapsulation of a certain kind of meaning: that where the calibration of relationships is autonomic and almost instantaneous. However, there are also constraints on the extent of encapsulated constituents in a single process, mainly to do with limitations on reliability or ease of recollection.

Words formularize a set of relationships at a single address (ie. by a single label). Each word meaning finds some optimum level of specificity which correlates with discriminated phenomena in perception and recalled detail in memory. As a consequence, some general properties of words are determined by organic constraints within the cognitive system itself.

Not only organic, but representational pressure may exist for the development of phrasal formulas. That is, the demands of social communication also encourage in certain instances the formularization of whole phrases. Language which requires some social calibration to be interpreted is a common source of such formulas. Compliments are a good example; (how can we know the worth of a compliment unless the phrase which carries it has a familiar social value?).

Also, repetition has a representational as well as an organic role. In social terms, partial repetition can initiate conversational strategies which are more adventurous than formulaic phrases. It reaffirms a topic of discussion, thus laying the basis for interpreting fresh propositions about ideas and perceptions, as opposed to encapsulated formulas.

The outcome of competing pressures in the organic and representational roles of language is, at both the generative and performance levels, not a smooth, uninterrupted flow of language. It is a stream of semi-encapsulated units which appear in utterance as intonation units and which the GO model refers to in their processing stages as g-vortices. G-vortices are close relatives to the perceptual frames well known in experimental psychology <4>.

The GO model posits two types (or at least two extremes) of cognitive processing activity. One is autonomic and almost instantaneous. It applies to the kind of encapsulated relationships found in formulaic structures, and at certain threshold conditions, in g-vortices. The second kind of processing is more linear and deliberative. It applies to the organization of g-vortices, or ultimately of intonation units, into extended utterance.

The lumpiness of linguistic performance over time, stemming directly from the cycle of tension and resolution just discussed, gives rise to the notion of generative oscillation.

Chapter 1 The Emergence of Language in a Neural Network

Glossary

[Because the GO model expresses a new paradigm, new terms have often been the shortest route to minimizing ambiguity and short-circuiting preconceptions. The glossary at the beginning of each chapter is intended to prime the reader for what follows. There is also a combined glossary at the end of the dissertation. Of course, many older expressions with historical baggage are also used. Where possible their intended meaning is indicated in the glossary as well. Note however that the glossary is indicative rather than difinitive. Concepts are often expanded in subsequent chapters.]

action [1] in the GO model refers to patterned activity in the brain or elsewhere which may or may not have correlates in other "observable" behavior. Thus "thought" is a form of action in this analysis. Action is governed by a shifting hierarchy of habits, and it is the interplay of actions within this immensely complex hierachy which conveys a sense of mind as the resources in the organism shift first to one action sequence, then another. This paradigm is familiar in Buddhist philosophies.

Buddhist philosophies [1] Many references are made in this study to concepts from Buddhist philosophies. There is no intent here to promote Buddhism as a theology (nor is the writer Buddhist). It is just that many of the most sytematic, lucid, and (as it happens) long-lasting insights into the nature of human cognition have come out of this very old tradition. It has more to contribute in this field than the Aristotelian notions which have dominated Western thinking.

coevolutionary relationship [1] A nervous system cannot change without changing its environment. An environment cannot change without affecting dependant nervous systems. There is not a priori an ordered world external to the individual, nor a priori an ordered mind imposing logic on a chaotic world. Rather, a nervous system together with its percepual aparatus grows and emerges in concert with an environment. It is the relationship between them which has structure and continuity. This is a view familiar to ecologists.

free will [1] is really the unfettered (and often unplanned) exercise of inclinations to action (habits) within a person, without regard to influences from and likely reactions from the wider environment. In a sense, nothing is less free than "free will". The Buddhist influence on this characterization is acknowledged.

GO model [1] is a hypothetical inner human environment within which the systems of language might operate. This environment is moderately elaborate, but in the end is no more than a working construct. It enables a dialogue about that which cannot be observed directly. Its real function in this study is to facilitate the understanding of complex prcesses, such as how repetition impacts on creative activity in language generation.

habit [1] effectively a neuronal highway leading to action, is not merely the easiest response pattern; it is the only one which will be followed in the absence of coutervailing habit (possibly at a different level of organization). The notion of habit here is borrowed from Buddhist philosophy.

idea [1] as an artifact to be located in a "mind", is a cultural construct. In the GO model, following the Buddhist concept, "idea" is a specific subset of inclinations to action. Mind comprises the general set of such inclinations to action (see below). Historically, there are many perceptions of "idea". For example, before the seventeenth century in Europe, "idea" was often a property of godhead, not of the individual;(Boulton 1991).

inclinations to action [1] arise from habitual associations, ultimately associations or pathways among neurones. In a single lifeform, these associations emerge from the way in which experience reacts with inherited biological design. They are not necessarily beneficial to the organism. The balance of inclinations to action must be adequate however to preserve the functional sufficiency of the organism.

language [1] is regarded in the GO model as an intermediate system between an inner ecology and an outer environment. However, the ecology of the inner human domain is considered to be a subsystem and a continuum of the outer environmental systems. This is not intended to imply a crudely mechanistic view. Mind, language and the environment external to the person are all subject to the same natural laws. Mind/body dualism which implies a metaphysical component is rejected.

language generation [1] though amenable to some conscious control, seems, as it were, to be programmed to operate (spoken or unspoken), regardless of any immediate survival need. It is not so much an act of representation as of habitual pattern-making. This pattern making is apt to take as its momentary mould any set of salient perceptions close to hand, a phenomenon recognised in the extreme as "idle chatter". Perceptions can be drawn from outer senses like sight or hearing, or inner senses of memory, emotion and so on;(the notion of inner senses is borrowed from the skandha or "aggregates" of the Buddhist abhidharma (commentaries)).

mind [1] is taken in the GO model to be a code label for the organised totality of established inclinations to action in an individual; (see action).

mind/body dualism [1] The unanalyzed cultural orthodoxy in almost all modern scientific behaviour is that a human being exists in two domains : the mind and the body. This behaviour contrasts with stated orthodox belief and constitutes one of the great hypocrisies of the age.

mindfulness/awareness [1] [ref. Varela et al. 1991:21] "derives from the Buddhist method of examining experience call mindfulness meditation... Mindfulness means that the mind is present in embodied everyday experience; mindfulness techniques are designed to lead the mind back from its theories and preoccupations, back from the abstract attitude, to the situation of one's experience itself.. We believe that the Buddhist doctrines of no-self and of nondualism that grew out of this method have a significant contribution to make in the dialogue with cognitive science."

self [1] as named by an individual is his possessive identification of an exclusive mind-body.

thought [1] is a form of action in the GO model.

The Emergence of Language in a Neural Network

The chapter has begun by stating a number of premises in the form of a glossary, some of which may at first seem unsupported by familiar canons of linguistics, philosophy or general science. However care is taken to develop linguistic argument within a wider context of philosophical thought. The main body of the thesis later will in fact show that most of the premises have respectable origins, and synthesize to express a view of linguistics which is viable and productive.

1.1 Historical perspectives on language and mind

This thesis is a study of discourse which will attempt to marry empirical observation with a particular philosophy of mind. No philosophy arises in a vacuum, and in this instance it will be necessary to spend some time to establish a context for the approach adopted.

Assumptions which have governed the study of language are very nearly those which have governed the orthodox medical view of human beings in each age, a conjunction which can sometimes help us to guess at linguistic scholarship even where records no longer exist <5>. It is clear that there have been many paradigm shifts in these medical and linguistic views.

The nature of linguistic phenomena, and the mechanisms proposed to explain them, were essentially inseparable from beliefs about the nature of life, mind and intelligence until the ascendance of behaviourist and structuralist methodologies in the twentieth century. These latter traditions eschewed the idiosyncratic speculation that often passed for research in the nineteenth century and earlier. They pursued a useful fiction (with varying rigour) which was also a discipline. This fiction was that what occurred subjectively - that is, beyond observable verification - was no business of science. A very great deal of organised knowledge has derived from this empirical positivism.

Gradually an understanding has also arisen however that to interpret the relative significance of objective observations, systematic recourse must be had to rational, hence subjective, evaluations. Some scholars have gone further to argue that models built from rational insights must anticipate balanced enquiry in the public domain. It is necessary, in other words, to decide rationally which questions need to be answered. The objective answers which are found, it is realized, cannot be separated from the subjective questions in which they are sourced.

Most recently, particularly in some areas of cognitive science, the paradigm focus has come full circle back to the human mind, but the overall intellectual objectives are shifting subtly. Traditionally,

"..Western philosophy has been more concerned with the rational understanding of life and mind than with the relevance of a pragmatic method for transforming human experience." [Varela et al 1991:218].

Now there is at least a tentative assertion that the ecology of cognition can yield something to systematic enquiry. Further, the enquiry itself may influence what cognition can achieve, (and an echo with quantum mechanics here may not be accidental).

This paradigm shift has been only slightly acknowledged in the practice of mainstream linguistics, but if history is any guide then it is bound to become a major preoccupation. The dissertation which follows is a very preliminary foray into the kind of refocussing in linguistics that might be necessary where the study of mind is an integral part of the study of our greater environment.

Rousseau's eighteenth century view that language originally imitated "cries of nature" is scarcely taken seriously today. However the belief that language is a symbolic representation of nature is very nearly universal. It has accompanied an almost instinctive popular feeling that mind and body are different realms, differently governed. Much more will be said on this topic later.

Cognitive representation is often defended vigorously as functional <6>. Johnson-Laird (1993:xi)proposes that

"thinking normally leads from one mental representation of the world to another in order to prepare the individual for the demands of life."

The sense of a purposive guiding spirit is very strong in this kind of statement, but its ontology forms no part of the explanation. Rather, there is an appeal to apparently logical process:

"..a causal chain leads from an object, A, to a pattern of energy impinging on an organism's sensory organs; there devices convert the energy into nerve impulses, and from them the brain constructs a further pattern, A1, in an internal symbolic notation that can be used to control action. The representation will be useful to the extent that it covaries with the state of the world ad makes explicit those aspects of it that are relevant to action. The organism can avoid an obstruction, A, by virtue of the representation A1."

Johnson-Laird acknowledges that there have been critics of the representational approach (Husserl 1929; Edelman 1987; Gibson 1966) but responds somewhat impatiently that:

"[xiii] ..sooner or later a functionalist must abandon the endless rearrangement of leaves in the philisophical album and enter the psychological laboratory to test a theory of thinking based on mental representations."

So real men must push buttons. Some other hands-on artificial intelligence researchers have arrived at very different conclusions. Thus Varela quotes the robotics specialist Rodney Brooks from the AI laboratory at MIT:

[Varela et al. 1991:208] "In this paper I ....argue for a different approach to creating Artificial Intelligence ... We have been following this approach and have built a series of autonomous mobile robots. We have reached an unexpected conclusion (C) and have a rather radical hypothesis (H):

C: When we examine very simple level intelligence we find that explicit representations and models of the world simply get in the way. It turns out to be better to use the world as its own model.

H: Representation is the wrong unit of abstraction in building the bulkiest part of intelligent systems."

Many results from the representational/symbolic program can in fact be utilised by researchers from any tradition. The danger of the symbolic game, as I will try to argue in later chapters, lies in the assertion of rigid cause and effect between a system and its environment. It is one thing to observe correlations, perhaps very strong ones, between, say, a worldly event and a piece of language. It is quite another thing to say that the language can only be explained as a representation of the worldly event, and (if you are a Johnson-Laird) go on to write a computational computer program to give this connection the appearance of logical necessity.

In pragmatics, sociolinguistics, discourse analysis, and the whole general area of so-called applied linguistics (basically language teaching), reference to "language functions" is very common. Early, influential lists of language functions were compiled by Roman Jakobson (1960) and Dell Hymes (1962). There have been many since. All of these derived from an implicitly representational view of language which also asserted that human speech behaviour was by definition purposive and survival oriented. In short it was (and is) a neo-Darwinian philosophy. I cannot think of a major second or foreign language learning curriculum in the last thirty years which has not adopted the same approach.

Well, how purposive is language? The answer turns very much on who is asking the question and why. If challenged, most people most of the time can give a socially acceptable reason for having uttered something. In fact this is part of being a competent speaker of the language. The machinery of English itself (for example) comes extremely well equipped with a repertoire of purposive constructions, deductive operators, and so on. It is therefore perfectly reasonable for language teachers, sociolinguists and others to talk about language functions. However there is another, more biological level at which questions of purposive behaviour can be raised.

An argument will be developed in this thesis that both our speech and our post-hoc rationalizations of it are an inescapable part of being the biological systems that we are. We cannot, as it were, avoid using language once we have acquired it, though it might be restricted at times to private thought. If language does grow unbidden in our minds as surely as hair grows on our eyebrows, then the purely derivative, "functional" explanations of general linguistics are going to have rather restricted explanatory power.

1.2 Cognitive Science since 1942

A mechanistic attempt to dispense with the mind/body divide has been implicit in much cognitive research since about 1942. Coinciding with the rise to prominence of digital computers there has been a view of the mind as essentially "software in the computer". Varela (1991:8) documents the cognitivist worldview:

"Cognitivism consists in the hypothesis that cognition - human cognition included - is the manipulation of symbols after the fashion of digital computers. In other words, cognition is mental representation: the mind is thought to operate by manipulating symbols that represent features of the world or represent the world as being a certain way. According to this cognitivist hypothesis, the study of cognition qua mental representation provides the proper domain of cognitive science, a domain held to be independent of neurobiology at one end and sociology and anthropology at the other."

There are ironies in this mechanistic approach, not least that the insistence on representation creates two worlds quite reminiscent of the original mind/body divide. The cognitivist view also became enmeshed with a fairly extreme version of representational linguistics, emanating particularly from the M.I.T. school:

[Varela et al 1991:40] "1956 was clearly the year that gave birth to cognitivism. During this year, in two meetings held at Cambridge and Dartmouth, new voices ( such as those of Herbert Simon, Noam Chomsky, Marvin Minsky, and John McCarthy) put forth ideas that were to become the major guidelines for modern cognitive science.. The central intuition behind cognitivism is that intelligence - human intelligence included - so resembles computation in its essential characteristics that cognition can actually be defined as computations of symbolic representations."

Artificial intelligence (AI) research, which was the natural home of cognitivism, has had a long gestation of high promise and low performance. The relative lack of practical success in the most ambitious areas of AI has led to a re-evaluation of the original cognitivist position. Again, it is explained very well by Varela (1991:8) <7>:

"In the past few years.. several alternative approaches to cognition have appeared. These approaches diverge from cognitivism along two basic lines of dissent:(1) a critique of symbol processing as the appropriate vehicle for representations, and (2) a critique of the adequacy of the notion of representation as the Archimedes point for cognitive science.

"The first alternative, which we call emergence .. is typically referred to as connectionism. This name is derived from the idea that many cognitive tasks (such as vision and memory) seem to be handled best by systems made up of many simple components, which, when connected by the appropriate rules, give rise to global behaviour corresponding to the desired task. Symbolic processing, however, is localised. Operations on symbols can be specified using only the physical forms of the symbols, not their meaning. Of course, it is this feature of symbols that enables one to build a physical device to manipulate them. The disadvantage is that the loss of any part of the symbols or the rules for their manipulation results in a serious malfunction. Connectionist models generally trade localized, symbolic processing for distributed operations (ones that extend over an entire network of components) and so result in the emergence of global properties resilient to local malfunction. For a connectionist a representation consists in the correspondence between such an emergent global state and properties of the world; it is not a function of particular symbols.

"The second alternative ... is born froma deeper dissatisfaction than the connectionist search for alternatives to symbolic processing. It questions the centrality of the notion that cognition is fundamentally representation. Behind this notion stand three fundamental assumuptions. The first is that we inhabit a world with particular properties, such a length, colour, movement, sound etc. The second is that we pick up or recover these properties by internally representing them. 'the third is that there is a separate subjective "we" who does these things.These three assumptions amount to a stong, often tacit and unquestioned, commitment to realism or objectivism/ subjectivism about the way the world is, what we are, and how we come to know the world."

Varela, Thomson & Rosch (1991) are concerned to establish that a non-representational view of mind is a viable starting point for scientific activity:

[1991:9] "We propose as a name the term enactive to emphasize the growing conviction that cognition is not the representation of a pregiven world by a pregiven mind but is rather the enactment of a world and a mind on the basis of a history of the variety of action that a being in the world performs. The enactive approach takes seriously then, the philosophical critique of the idea that the mind is a mirror of nature but goes further by addressing this issue from within the heartland of science."

Perhaps the the most potent consequence of an enactive approach to science is that a unidirectional paradigm of cause-and-effect is no longer adequate. Phenomena co-arise. The effect, as it were, shapes the cause, and vice versa. As Varela et al. put it

[1991:172] "the enactive approach consists of two points: (1) perception consists in perceptually guided action and (2) cognitive structures emerge from the recurrent sensorimotor patterns that enable action to be perceptually guided...[173].. for the representationist, the point of departure for understanding perception is the information-processing problem of recovering pregiven properties of the world. In contrast, the point of departure for the enactive approach is the study of how the perceiver can guide his actions in his local situation. Since these local situations constantly change as a result of the perceivers's activity, the reference point for understanding perception is no longer a pregiven, perceiver-independent world, but rather the sensorimotor structure of the perceiver (the way in which the nervous system links sensory and motor surfaces.)"

1.3 Coevolution in natural systems

I accept the basic philosophical premises adopted by Varela et. al., that is, the critique of the representational view of cogntition, and the proposal of an "enactive" alternative. Giving that acceptance a clear rationale however, especially in the context of linguistics, and exploring its consequences will take considerable argument.

The analysis in this thesis takes another, rather difficult step beyond the notions of enactive cognition. I propose that the forms of language, non-representational in their essence, can be consciously put to representational purposes in a social context. From this possibility emerges an underlying tension between language as an organic growth and language as a representational instrument. Why does this occur? No doubt the consciousness (if we wish to call it that), which commands a representational manipulation of language at one level, is itself mutually engaged with the environment: "God's radio receiver" of the 17th century (a popular view of mind at that time) is seen to be a self-modifying transmitter too. However, one effect of its action is to create a sub-domain of language use, a sort of dependent system, which is amenable to social control.

In choosing to examine any natural system it is not sensible to assume that it evolved selectively to survive in a static environment. Let us take some easily understood examples:

(i) Plants, animals, and the ratio of unstable atmospheric gases each use have quite clearly coevolved with the earth's atmosphere; (oxygen in the atmosphere has increased from about one part in a million to one part in five in the presence of life forms. Over four billion years the sun's luminosity has increased by 50% yet the earth's temperature has remained constant within a few degrees, also as a result of atmospheric interaction with life forms).

(ii) There is evidence that the trichromatic ultraviolet sensitivity of bee vision and the colour in flowers coevolved (Lythgoe, J 1979).

(iii) Human language, human society and human biology seem to have also coevolved to an extent that make discussion of "nature Vs nurture" at species level quite sterile. Nature is not a constant.

1.4 Biological brains, the environment; emergent worlds

Biological brains and all systems with which they are associated are in a coevolutionary relationship. A nervous system cannot change without changing its environment. An environment cannot change without affecting dependant nervous systems. There is not a priori an ordered world external to the individual, nor a priori an ordered mind imposing logic on a chaotic world. Rather, a nervous system together with its percepual aparatus grows and emerges in concert with an enviroment. It is the relationship between them which has structure and continuity. This view will be familiar to ecologists, but sits uneasily with the traditions of analytic science.

1.5 Repetition, sensitization

Repetition, even accidental repetition within a nervous system sets up a kind of strengthened neural connection that multiplies the weighted possibility of even further perceived repetition along the same lines. That is, the system becomes sensitized to a certain pattern of stimulus, which in turn increases the likelihood that that stimulus will be responded to in an otherwise random bombardment of stimulii. The aural discrimination of phonemes in speech is an excellent example of such sensitization. Comparable, but unfamiliar sounds of similar duration are not discriminated (not "heard").

1.6 Inclinations to action, habit

Inclinations to action arise from habitual associations, ultimately associations or pathways among neurones (this is essentially a Buddhist description). In a single lifeform, these associations emerge from the way in which experience reacts with inherited biological design. They are not necessarily beneficial to the organism. The balance of inclinations to action must be adequate however to preserve the functional sufficiency of the organism. Habit, effectively a neuronal highway leading to action, is not merely the easiest response pattern; it is the only one which will be followed in the absence of coutervailing habit (possibly at a different level of organization). Thus, the smoker will automatically reach for a cigarette unless competing inclinations, for example a generalized but potent inclination to self-preservation, intervene.

1.7 Mind, idea, self

"Mind" is taken in this discussion to be a code label for the organised totality of established inclinations to action in an individual. An "idea" would be a specific subset of such inclination to action. Action here refers to patterned activity in the brain or elsewhere which may or may not have correlates in other "observable" behavior. Thus "thought" is a form of action in this analysis; (once again, a Buddhist notion). An "inclination to action" in such a paradigm could be rephrased as "a tendency to form certain associations or patterns". Action is governed by a shifting hierarchy of habits, and it is the interplay of actions within this immensely complex hierachy which conveys a sense of mind as the organism shifts resources first to one action sequence, then another. In such a context it makes little sense logically to talk of a mind-body dualism: both are ways of viewing a single emergent macrosystem.

Similarly, "idea" as an artifact to be located in a "mind", is a cultural construct, and by no means universal. For example, before the seventeenth century in Europe, "idea" was often a property of godhead, not of the individual; (Boulton 1991). A human being in this schema was a sort of radio transmitter and largely an automatom for the will of God (or occasionally for evil spirits). Such a belief is still widespread. The consequences of competing views on this matter are profound for any society.

The mystical sects of most religions have always expressed humans as captive agents of another intelligence. The possible levels of captivity are also almost exactly represented in designs for computer networking: from the utterly enslaved dumb terminal, to quasi-autonomous machines dependent upon a central server, to exchange networks in which independent machines are facilitated by a universal network protocol. The copyright on an "idea" in each of these paradigms clearly varies, and with that goes a whole range of views on ultimate personal responsibility.

No matter how carefully constructed the arguments are on the topic of mind/body however, personal beliefs about it remain tenaciously resistant to change. In both Eastern and Western philosophies it is commonly taken that the naming of "self" by an individual is his possessive identification of an exclusive, persistent mind/body. Thus any threat to that individual's hierarchy of habits is a threat to self and will be resisted fiercely.

The disjunction between private belief and intellectual argument is sharply marked on questions of human self-identity. In a way, the intellectual argument was won long ago:

[Varela et al 1991:59] ".. all of the reflective traditions in human history - philosophy, science, psychoanalysis, religion, meditation - have challenged the naive sense of self. No tradition has ever claimed to discover an independent, fixed, or unitary self within the world of experience. Let us give the voice for this to David Hume's famous passage: "for my part, when I enter most intimately into what I call myself, I always stumble on some particular expression or other, of heat or cold, light or shade, love or hatred, pain or pleasure. I never catch myself at any time without a perception, and never can observe anything but perception.""

[60] "We believe that many non-Western (even contemplative) traditions, and all Western traditons, deal with this contradiction [the elusive self] by turning away from it, refusing to confront it, a withdrawal that can take one of two forms. The usual way is simply to ignore it .."

This dilemma is confronted directly in Buddhist philosophy, where the struggle to accept the perceived non-reality of self is a major part of Buddhist practice. Drawing on Buddhist traditions, Varela, Thompson & Rosch mount a quite persuasive argument that by learning to discipline the human mind in a principled way which involves subduing illusions of selfhood, one can in fact analyse many of mind's important properties. This author is not equipped, at least at present, to properly test such a possibility as a linguist. However the proposal is so intriguing that it needs to be recorded:

[1991:23,24] "What relevance does [mindfulness] have for cognitive science? We believe that if cognitive science is to include human experience, it must have some method for exploring and knowing what human experience is."

[24] "To get a sense of what mindfulness meditation is, one must first realize the extent to which people are normally not mindful. Usually one notices the tendency of the mind to wander only when one is attempting to accomplish some mental task and the wandering interferes. Or perhaps one realizes that one has just finished an anticipated pleasurable activity without noticing it. In fact, body and mind are seldom closely coordinated. In the Buddhist sense we are not present.

"How can this mind become an instrument for knowing itself? How can the flightiness, the nonpresence of mind be worked with? Traditionally texts talk about two stages of practice: calming or taming the mind (Sanskrit: shamatha) and the development of insight (Sanskrit: vipashyana). Shamatha, when used as a separate practice, is in fact a concentration technique for learning to hold .. the mind to a single object. Such concentration could eventually lead to states of blissful absorbtion; although such states were assiduously catalogued within Buddhist psychology, they were not generally recommended. The purpose of calming the mind in Buddhism is not to become absorbed but to enable the mind to be present with itself long enough to gain insight into its own nature and functioning. ..."

My own present analysis of "self" incorporates but also reinterprets the Buddhist/enactive view. I accept that searching in the domain of present sensation, my "self" is not to be found. As Kant, Derrida and the Buddhists have equally found for themselves, my interior present seems to be a shifting pastiche of ideas and sensations without a circus master.

Yet I do recall a definite self. To misquote Rene Descartes, "I remember, therefore I am". I have a vivid memory of walking down a dirt country road thirty-five years ago as the last rays of the sun turned the wheatfields to deep gold. The many parts of that memory have a unity and an "I" whom I somehow still possess is at the centre of it. To you, my reader, my memory is a mere collection of words on a page. Where is my self in the present therefore? Not yet made perhaps, but in the process of becoming. As these words hit the computer screen they are already past, and they have become a committed part of my past. The power of the meditation discipline (if it has the power claimed for it) will, I think, be a power to restrain the past,to govern habit and inclination, so that other realities have a chance to synthesise.

1.8 Free will, Buddhist philosophy, cognitive science, remaking the self to re-examine nature

In a sense, nothing is less free than "free will". Free will is really the unfettered (and often unplanned) exercise of inclinations to action (habits) within a person, without regard to influences from and likely reactions from the wider environment. The claim to free will is therefore frequently as much a claim to uncritical inertia as the action of an habitual devotee "submitting to the will of God". To innovate, to create, requires the exercise of an overriding order of will.

The relationship of creativity to human will is especially interesting. It is a paradox of all kinds of creativity that its most skilled exponents are those who understand and control the existing rules best, rules which are normally an embedded part of their own habitual behaviour. A true Picasso is a master-draftsman first, a Fawkner or Joyce has infinitely finer control of the language than the untutored reader who thinks that language is being massacred. The most devastating intellectual is the one who has mastered, then surpassed existing knowledge. This generalization extends to everyday experience. The most humble quip or insult depends upon control of a contrast with the normative pattern.

Some philosophies, and notably Buddhist philosophy, have recognized this paradox of will for at least two millenia. Mahayana Buddhism (and later devlopments like Zen) have developed quite practical techniques for disciplining and ordering the community of habits (especially habits of thought) within an individual. These procedures are said to effectively remake the person, as well as the emergent realities which flow from the coevolution of a person and their environment. (In Buddhist parlance, the existence of a "self" is actally denied, and hence ceases to be an obstacle to change. An acceptance of this lays the foundation for proper development).

Western religions, philosophies and consequently, Western science have in some cases reached a point of recognizing the elusiveness of self;( Kant, Neitzsche, Sartre, Derrida, cognitives scientists like Minsky and linguists like Jackendoff). The reactions have been largely coffee-house intellectual (nihilism, existentialism) or a complete rejection of unhealthy introspection (American pragmatism). Prior to some recent developments in cognitive science (e.g. Varela, Thompson & Rosch 1991), the West has not generally taken seriously the pragmatic possibility of reshaping inner and outer environments to arrive at fresh insights into the character of language, of being, or of nature itself. There has been a failure to exploit the codependence of nervous systems and their environment,

1.9 Some consequences of mind-body dualism

Normal physical action by a person is readily understood as interaction with a larger environment. To chop wood is not to "represent" that larger environment, but to participate as a part of it. In so doing both the wood chopper and the larger environment are changed in certain ways.

There is a powerfully ingrained cultural habit in the European ethic which presumes that to think "about" some aspect of the larger environment is not to interact and grow as a part of it, as in chopping wood, but to "represent" it symbolically. That is to say, dreams are free. The universe is thus divided into a "real" or "objective" world external to the person, and an "imaginary" or "subjective" representation of the external world inside a person's head.

The consequences of this dualism are as profound and intractable as those contradictions which faced Ptolemaic astronomers who put the planet earth at the centre of the universe. This dissertation, like the Buddhist philosophy mentioned above, proposes a paradigm shift. It proposes to de-centre language from its status as a system of exclusively symbolic representations in human "minds"; (this is not to deny that symbol creation and manipulation remains one option). Language is no longer taken to be a mere mirror of an objective external reality. With that loss comes a surrender of privilege.

The privilege of the dualistic human is a claim to separateness of "self" from the laws and actions of external nature. Thoughts are private and timeless; only the act of speech makes them accountable by mapping them to a real world. By exempting a person from the discipline of reality by an act of will, the privileged self exercises great power, or at least, an illusion of power. There is an old saying that power corrupts, and perhaps this is what the Buddhists are about in dethroning self as a prelude to inner growth.

1.10 Non-representational patterns

What then is a design for language which is not representational? The partial answer is that it has a certain similarity to the action of chopping wood, and to the ability to chop wood. That action seems to be volitional, and the ability is learned. Another part of the answer is that language has a certain similarity to the action of breathing, and to the ability to breath. That action seems to be autonomic, and that ability is instinctual.

Neither the act of chopping wood nor the act of breathing encapsulate an external world in symbols. Rather they are both actions which show that the human is utterly embedded in the supposedly external world, and grows from it. To prevent such classes of action is to terminate life itself. Is it true then that to prevent those classes of action which we call thought and the generation of language is also to terminate life? We can not function as humans without these things, so it would certainly be the end of humanity. Is our humanity defined by a purely subjective and unreal imagination, a shadow play of external worlds? I think not. Language, thought, the whole apparatus of cognition are phenomena which are of and in the wider world. They are not modules which can be permanently decoupled somehow, or as in the conception of some cognitive scientists [] "downloaded" to another computer.

1.11 Intrinsic design and the applications of language

At this point it is necessary to draw a distinction between the intrinsic nature of language and the purposes to which it can be put. No reasonable person is likely to deny that a proper name like Thor May has a special relationhip to a being whom the wider world identifies as Thor May, nor that a common noun may have a special relationship to a whole class of objects, say, chairs.

The effective argument here is that a chameleon lizard which changes its skin colour to blend with (to represent?) its surroundings remains nevertheless a real, live chameleon lizard, a distinctive part of the habitat whose disappearance would change that habitat forever. And so with language, whose surface patterns change constantly to apparently reflect inner and outer environments of the organism. For all its chameleon character, language remains a distinctive, wholly real phenomenon, and an integral part of the ecology of the human habitat.

1.12 Language defined as a representational system

The problem with defining language as an exlusively symbolic, representational system, is that where representation appears to fail, then either the linguistic model has to be redefined to fit some notion of reality (meet a certain kind of test of explanatory adequacy) or reality has to be reconceived to fit the model.

An example of the former would be the ever more elaborate fixes found in transformational generative grammers from 1965 on. The second strategem is more subtle. We are familiar with the way in which each generation reinterprets history to fit an ideal of a remembered reality. The linguistic equivalent is well expressed by arguments deployed to defend the shifting philosophies of ontology: man possessed by god's spirit, man as a dualistic being of body & mind("self's spirit"), man as a mechanistic automaton (essentially the classical cognitivist view, shawn of finnessing).

In each of these incarnations, the ways in which language represented the world was taken as evidence for the way the world was. For an Augustine (and many a living fundamentalist since) to challenge the biblical word was to challenge god. For a Cartesian rationalist,the link role of language between inner life and an outer environment was ipso facto proof of a mind-body dualism. For a pragmatic behaviourist, coherent expression had to represent a coherent world that the speaker plugged into.

For a cognitive scientist like Minsky, or a linguist like Chomsky, the patterns of language mapped the patterns of some assumed, real world so intimately that an "ideal speaker hearer" was supposed to be somehow more real, more significant than the messy beings who clutter up our coffee breaks.

1.13 Language defined as a non-representational system

The unanalyzed cultural orthodoxy in almost all modern scientific behaviour is that a human being exists in two domains : the mind and the body. This behaviour contrasts with stated orthodox belief and constitutes one of the great hypocrisies of the age.. Philosophers from Neitschze to Sartre may have despaired of capturing the nature of self, yet Sartre says that "we are condemned to belief in the self" (Varela: p.60). A cognitive linguist like Jackendoff mutters that "consciousness is not good for anything" (The Computational Mind, p.26), yet cannot dispose of it:

[Varela et al 1991:69] "In our habitual and unreflective state of course, we impute continuity of consciousness to all our experience - so much so that consciousness always occurs in a "realm", an apparently cohering total environment with its own logic (of agression, poverty, etc.). But this apparent totality and continuity of consciousness masks the discontinuity of momentary consciousness related to one another by cause and effect. .. the flame is passed from one candle to the next without any material basis being passed on. Taking this sequence as a real continuity however, we cling tenaciously to this consciousness and are terrorized by the possibility of its termination in death. Yet when mindfulness/awareness reveals the disunity of this experience - a sight, a sound, a thought, another thought, and so on - it becomes obvious that consciousness as such cannot be taken as that self we so treasure and for which we are now searching... Perhaps the self is an emergent property of the aggregates? .. At this point however, the idea is of no help. Such a self-organizing or synergistic mechanism is not evident in experience. More important, it is not the abstract idea of an emergent self that we cling to so fiercely as our ego; we cling to a "real" ego-self."

The net result of this dualistic orthodoxy is that scientific understanding has progressed by leaps and bounds in the realm of bodies and objects. Scientists have left the realm of consciousness to shamans and astrologers while continuing to act in their daily lives on the basis that another shadowy, parallel universe does in fact govern their fate. They argue persuasively for evolutionary processes in a physical world, yet offer neither a rationale nor a mechanism whereby some comparable evolution may have been enacted in the mental domain.

Scientists are not naturally dishonest people. We must conclude that the compartmentalization of their behaviour has stemmed from a dilemma which has seemed to have no practical solution. The dilemma has been a sense that mind is an active agent but somehow intangible, and therefore not accessible to normal modes of enquiry. At least one other tradition of study has concluded that the impasse faced by Western scholars can be broached:

[Varela et al. 1991:30] "From the standpoint of a mindful, open-ended reflection the mind-body question need not be, What is the ontological relation between body and mind, regardless of anyone's experience? - but rather, what are the relations of body and mind in actual experience (the mindfulness aspect), and how do these relations develop, what forms can they take (the open-ended aspect)? As the Japanese philosopher Yasuo Yuasa remarks, "One starts from the experential assumption that the mind-body modality changes through the training of the mind and body by means of cultivation (shugyo) or training (keiko). Only after assuming this experential ground does one ask what the mind-body relation is". .. We may notice that this viewpoint is resonant with pragmatism, a view in philosophy that is having a modern revival. The body and mind relation is known in terms of what it can do".

This dissertation proceeds from the position that language is indeed a link system between an inner ecology and an outer environment. However, it does not accept that language flows, like the river Styx between a sunny, predictable, living outer world and a Hades of mental images where the devil's laws hold and no man may enter. Nor does it see semantics as a craft which must somehow make its journey between these realms.

The ecology of the inner human domain is considered to be a subsystem and a continuum of the outer environmental systems which we observe and attempt to catalogue in all their complexity. In nature we note that within certain restrictive degrees of freedom, living systems grow wild even to the point of self-destruction (not least the human species). We see a proliferation of wierd lifeforms in numberless sub-variations, and a detritus of evolutionary properties and appendiges which have long ceased to be of value in living systems. Even the human genome, we are now told, is at least 40% historical junk (Varela, p.191).

If the observed world is an exuberant and sometimes irrational jungle of historical relics, and if the inner ecology of human beings is a part and continuum of this world, then it would be extraordinary if it alone were a model of economy and mathematically elegant solutions. This writer has always had trouble in taking seriously the claims for explanatory power made by comprehensive models of computational linguistics and formal semantics. We can expect to find potent regularities in inner human systems, as we do in outer nature, but the factor of random discontinuities can never be discounted.

The metaphor was used earlier of language as a chameleon lizard, changing its appearance to blend now with this environment, now that. The process of language generation, though amenable to some conscious control, seems, as it were, to be programmed to operate (spoken or unspoken), regardless of any immediate survival need. It is not so much an act of representation as of habitual pattern-making. This pattern making is apt to take as its momentary mould any set of salient perceptions close to hand, a phenomenon recognised in the extreme as "idle chatter". Perceptions can be drawn from outer senses like sight or hearing, or inner senses of memory, emotion and so on;(the notion of inner senses is borrowed from the skandha or "aggregates" of the Buddhist abhidharma (commentaries)).

If language is something which does not necessarily exist for a "purpose", but grows like topsy as an organic system between inner and outer human environments, then we can track its structures without expecting them to have an unexceptional relationship to any other system. We can look for relational regularities between language and outer perceptions, or between language and other cognitive systems without being too surprised when a particular interface seems to break the rules, or be barely productive. At such points there is at least the possibility that we have stumbled upon another of nature's accidents.

If language does not represent but merely correlates partially in some characteristic ways with properties of adjacent environments, then we can note the common properties of such correlation, and call it semantics, without expecting that the correlation will never be partial, distorted, or sometimes fabricated from quite disparate sources. These after all are common failures that we learn to live with in the everyday interpretation of language. Our linguistic models cannot exceed in perfection the failures of the linguistic systems themselves.

The most plausible biological explanation of language is that it is an organic growth, engaging with surrounding environments, reacting to all kinds of stimuli with extraordinary sensitivity, yet not bound in any total way to external patterns. Against this we have the simple observation as conscious beings that we do constantly use language in purposeful ways. We deliberately attempt to represent aspect of perceived reality with language, and will agree with a little reflection that our language is predictably influenced by all kinds of social contexts. Our analysis of language must therefore take into account that language is not inherently representational, but that much of the time it is used with a representational intent.

In the next chapter I set out to describe a hypothetical inner human environment within which the systems of language might operate. This environment, termed the Go Model is moderately elaborate, but in the end is no more than a working construct. It enables a dialogue about that which cannot be observed directly.

A prime function of the GO model in this thesis is to facilitate the understanding of how repetition impacts on creative processes in language generation. This in turn should lead to a greater understanding of why linguistic meaning is encapsulated at word level, but only intermittently and unreliably in phrases. Finally, moving from the complete formulaic encapsulation of words, to the partial encapsulation of meaning in phrases, the much looser organization found in extended discourse will be seen as a natural extension of the processes of generative oscillation.

Chapter 2 Artifacts of the GO model

Glossary

-ent [2] = entity. Used as a suffix for a class of GO artifacts.

ard-ent [2] autonomic rule drive; simultaneous processor and the source of filtering within a g-vortex; some analogy with the strict categorial rules of generative grammars. Ard-ent is a collective reference to the preset machinery of phrase structure and other clause-level constraints. As the constituents of an incipient g-vortex meet certain threshold conditions they are zapped for parallel processing by the ard-ent. The ard-ent generates something like a well-formed clause which is then manipulated as a unit by the mord-ent (monitor rule drive) for textual purposes.

assembly plane [2] a conceptual space in the GO model, useful for locating the interaction of model artifacts; no fixed defining properties.

attractor [2] essentially any factor which will influence the the development of a dynamic system.Attractors (dynamic systems attractors) in the GO model are a concept borrowed directly from the "attractors" of mathematical chaos theory. A post (an "attractor") stuck in a river will shape the current flow downstream. More intriguingly, a "post", that is, a constant, stuck into any randomly developing (chaotic) environment will result in the emergence of ordered patterns, but not always the same patterns.

consciousness/awareness [2] Wallace Chafe notes the following as typical properties of awareness: [Chafe (1980b):11] Four properties of consciousness:
a) limited capacity
b) limited duration;
c) consciousness moves in "jerks" or "snapshots", not fluidly;
d) consciousness has a central focus and a periphery: an especially small amount of information is maximally activated.
The properties of consciousness identified by Chafe are precisely those noted by Nagarjuna two thousand years ago and elaborated into the sunyata of Mahayana Buddhism. Further, they match wonderfully the characteristics of perceptual frames described in neuroscience and adapted for the GO model.

g-ripple [2] repeating entity; provides cohesive ties in texts; selected g-ripples may be progenitors of g-vortices.

g-vortex [2] a linguistic mechanism analogous to visual perceptual frames of about 100 milliseconds duration. No specific claim is made for the duration of g-vortices, but an important subtype is said to contain clause-like constituents which are later projected into intonation units.

GO [2] generative oscillation model

HRA [2] harmonic resonance attractor; a compact way of describing the amplification of particular cognitive variables, their sympathetic resonance with elements in memory, and emergence from minor status to controlling influence in particular events. Harmonic resonance attractors acting on elements of language provide a principled way of talking about what language reflects of its co-environments.

mord-ent [2] monitor rule drive; linear processor and the source of filtering between g-vortices; major determinant of textual coherence. "Mord-ent" is a cover-term for the very eclectic variety of rules and decisions which may impact on language that is to be projected into surface strings. Mord-ent operations would normally correlate with the second, sequential type of processing, and the typical output would be a stream of intonation units.

perceptual frame [2] a cognitive unit which classically deals with sensorimotor rhythmicity and parsing. The GO model extends its use to linguistic parsing. The term is borrowed from the literature on neuroscience and psychology where, for example, one of the better known phenomena discussed is "perceptual simultaneity" or "apparent motion"; (refer Varela 1991:72)

plane of conscious imagination [2] a way of giving conscious thought, as opposed to unconscious mental activity, a location and identity in the notional cognitive geography of the GO model.

topic [2] interpreted as a quasi-linguistic entity in the GO model; it enters both into the linguistic sytem and other kinds of thought; provides a frame for matching sequences of linguistic g-vortices with non-linguistic ideas and objects. Topics act as mnemonic attractors (see attractors). Topics also offer a stable framework for matching coherent patterns of idea-against-language while linguistic constituents are shaped into linear phonetic output, or as phonetic input is decoded. Topics thus engage but are not fully defined by linguistic expression. There is always something more, such as background knowledge, to give them substance. They are never quite coextensive for speaker and listener.

waveform filter [2] a notional mechanism by which the ard-ent matches and edits the constituents of an emergent g-vortex. These constituents themselves would be recognized by linguists as collocating elements of a clause, but together are thought of physically in the GO model as a complex standing wave form.

Artifacts of the GO Model

The assembly of concepts dealt with in this chapter will be referred to collectively as the GO model (Generative Oscillation Model). The GO model is a speculative hypothesis about the cognitive organization of language. It makes deliberate use of iconic imagery as an aid to conceptualising linguistic processes, but stakes no strong claims about the relationship of such imagery to actual physiological processes. Consonant with the approach explained in Chapter 1, the artifacts of the GO model are intended to be compatible and interactive with the outer human environment. Later chapters will engage the concepts outlined here as analytic tools, and as instruments for articulating a coherent model of language generation.

Much of the GO model's power derives from a central proposal that the generation of language involves two different kinds (or extremes) of computational processing, which in turn entail radically different constraints and structures. The first kind is a subset of what is technically known as parallel processing, a procedure in which many problems are factored simultaneously but converge to a unified solution. It is proposed that simultaneous processing in language applies to small clause-like packets of information. A whole family of processing designs have in fact been developed with parallel architectures, but the differences need not concern us for the moment. The second kind of processing hypothesized for the GO model is sequential and more applicable to the organisation of extended discourse.

2.1 The parsing of perceptual frames for simultaneous processing

Faced with a perceptual event, any nervous system is limited by certain intrinsic problems. The near-simultaneity of parallel processing is a first major constraint. Even in such an awesome device as the human central nervous system only a finite number of variables can be usefully made salient in a working location for a finite duration. This parsing of mental activity has been discussed at some length by Varela, Thompson & Rosch (1991). Their discussion is worth reprinting:

[p.72.] "Is there any evidence for momentariness in the functioning of the brain? ... An examination of experience with mindfulness/awareness reveals that one's experience is discontinuous - a moment of consciousness arises, appears to dwell for an instant, and then vanishes, to be replaced by the next moment. ... [73] ... There is a literature in neuroscience and psychology that can be referred to as "perceptual framing", which deals with sensorimotor rhythmicity and parsing. One of the best known phenomena in this literature is called "perceptual simultaneity" or "apparent motion". For example, if two lights are shown successively with an interval of less than a period of 0.1 to 0.2 seconds, they will be seen as simultaneous .. if the interval is slightly increased, the flashing lights will apperar to be in rapid motion. If the interval is increased further, the appearance of motion becomes distinctly sequential... .. It is well known that the brain has a periodic rhythm of activity... Since the dominant rhythm for the visual cortex is also about 0.15 seconds, it is natural to assume that there is a relationship between temporal framing and cortical alpha rhythm."

[75.] "Experiments such as these assume that there is a natural parsing in the visual frame and that such framing is at least partially and locally related to the rhythm of one's brain in the range of duration of about 0.1-0.2 seconds at its minimum. ... Everything that falls within a frame will be treated by the subject as if it were within one time span, one "now". .... Furthermore, it has become apparent the neurons in the central nervous system have a rich diversity of electrical properties based on ionic conductances that endow them with autorhythmic oscillatory properties... This entire cooperative activity takes a certain time to start and to culminate. Such oscillations/resonances can be seen as timing sensorimotor coordination (among other possible functional roles)."

[76] "..there is good evidence that the activity of a neuron in the thalamus and the cortex of mammals has a unitary time course of about 100 milliseconds following a burst of presynaptic input." [ TM: note the much coarser analogy from phonology, where speech articulation is tied to the thoracic rhythm of breathing].

The second major constraint on a nervous system is that it seems likely that the operational patterns (ie. rules) for input and output to a given parsed frame will be established well before anything proceeds. Or in the parlance of fractal mathematics (which is probably what is at work here), the relevant "attractors" will be in place to filter probable outcomes. The reason for a an assumption of preset input filters is simply that in a single perceptual frame (or other cognitive equivalent) of maybe 100 milliseconds duration there is simply no time or mechanism for adjustment and learning.

If we make the intuitive leap in linguistics that clause or intonation unit boundaries derive from a mechanism which is analogous with the perceptual frames discussed by Varela et al, then categorial and other constraints that engage within a phrase structure would, in a given frame instance, also operate autonomically (rather than analytically, as with a foreign language student trying to recall a grammatical sequence).

The constrained number of inputs per unit of "frame" time, and the need for preset filtering rules both appear to accord very well with the behaviour of generative grammars with their strict categorial rule structures for outputting or decoding sentential clauses. Note that it is not claimed that phrasal processing "frames" are of 100 milliseconds or so duration. We just don't know. (As it happens, the duration of an unstressed syllable pronounced in English is about 150 milliseconds. That could reflect a gross motor (ie. muscle) limitation, or it could be associated with a lower boundary of synaptic discrimination, as in vision).

2.2 Sequential, decision-directed processing

The second kind of processing envisaged for the GO model has a complimentary but quite different profile to perceptual micro frames with their simultaneous syntheses of new structure. Clearly many elements of language production are planned, manipulated in various ways or corrected in mid-stream. This kind of adjustment can occur at any level from morpheme to text, but in general there is most evidence for some kind of executive control in the macro ranges of organization, notably those above clause level. I have suggested that such executive control could not easily extend in any constant way into perceptual frames (or their equivalents) of tiny duration.

There are various kinds of processing design which might handle the manipulation of extended language strings. Some effective long-term mechanisms for parallel processor learning have been proposed. One of the best known is a mapping of inputs to outputs and back-propagating any error vectors for future reference (Rietman 1988:91). Just what a conscious decision might amount to in this paradigm is less easy to see. Perhaps, as was suggested in the discussion of mind in Chapter 1, it could come from the interplay of a hierarchy of habits. If we conceive of this decision outcome as an "attractor" or filter, then a dynamic array of such filters might be adequate to direct the output of extended text. Conventional digital computers are inherently linear of course, and so well adapted to sequential management. Whatever its normal mode of operation, the human brain can emulate linear processing, as it must with problem solving in formal logic or mathematics.

Although the actual design which implements it is unclear, the second kind of neural processing hypothesized here behaves in an open-ended linear manner, where the outcomes from other operations are organized as they become available, where elements may be held back, sequenced or promoted, and where all sorts of contingent conditions may be bought into play when it seems expedient.

This kind of processing seems inherently more congenial to the often loose management found in the diachronic organization of discourse. It would seem to be open to pragmatic considerations, and well-adapted for applying repair strategies where discourse, or even clause generation, breaks down.

Although I have suggested that the second, linear kind of processing is motivated in the GO model by internal linguistic considerations, something comparable has been found necessary in artificial neural networks which attempt to model human cognition. Let me draw a short quotation from a book by Ed Rietman (1988:10), Experiments in Artificial Neural Networks :

"The human brain provides an existence proof that massive parallel processing with billions of processing elements [100 billion logic units] is possible. Often in connectionist architecture the individual elements do not have much memory associated with them. Long term storage is accomplished by the interconnection of the processors themselves ... The processing units in most connection machines do not follow complex instructions or programs. Each processor is capable of only very simple action such as threshold logic. The entire connection machine is usually controlled by a host computer.." [my italics].

In other words, I have articulated such executive control as the second kind of processing.The presentation of GO assigns this control to a hierarchy of notional executive centres and processes which I will now briefly identify.

Iconic spaces, operators and processes in the GO model

2.3 The Assembly Plane

One iconic invention of the GO model is to propose a cognitive space, a work area called the assembly plane, in which the constituents for various linguistic operations are collected. The assembly plane has no defined properties itself, but normally features at the space where topics are projected into the linguistic code. Its purpose is purely to let us group a bunch of ideas together and talk about them in familiar locative language.

2.4 G-vortices

The perceptual frames discussed by Varela and others have some kind of experimental validity. Although GO will use that validity as evidence, I wish to project into the frame notion a number of hypothesized properties. Hopefully other evidence will emerge to support GO's frame elaboration, but to avoid ambiguity it seems best to rechristen the frame concept with a special name. In GO it will be called a g-vortex. One historical meaning of "vortex" seems especially apt:

"Vortex, in old theories, as in Cartesian philosophy, [is] a rapid rotatory movement of cosmic matter about a centre, regarded as accounting for the origin or phenomena of bodies or systems of bodies in space." [Macquarie dictionary].

Thus the g-vortex is a linguistic mechanism analogous to visual perceptual frames of about 100 milliseconds duration. No specific claim is made for the duration of g-vortices, but I will propose that an important subtype contains clause-like constituents which are later projected into intonation units.

2.5 Ard-ents and mord-ents

As the constituents of an incipient g-vortex meet certain threshold conditions they are zapped for parallel processing by the ard-ent (autonomic rule drive). The ard-ent is a collective reference to the preset machinery of phrase structure and other clause-level constraints. The ard-ent generates something like a well-formed clause which is then manipulated as a unit by the mord-ent (monitor rule drive) for textual purposes.

"Mord-ent" is a cover-term for the very eclectic variety of rules and decisions which may impact on language that is to be projected into surface strings. Mord-ent operations would normally correlate with the second, sequential type of processing discussed above, and the typical output would be a stream of intonation units. These processes will be explained in more detail below.

The concept of an ard-ent is not entirely fanciful. Again, Varela's discussion of perceptual discrimination is relevant.

[Varela 1991:77-78] "From an information processing point of view in contemporary cognitive psychology, form and discernment would appear to specify each other. Form can be seen as the arising of something distinct from a background .... On the other hand, neurophysiological observations indicate that the initial stages of perceptual organization ... precede the more cognitively related electrical correlates by some 100-200 milliseconds. This time difference might be just too fast for detailed attention except when training in attention has stabilized sufficiently to notice the difference." [My italics]

Thus, within a perceptual frame, discriminations can only be made where they are familiar to the point of being autonomic. This phenomenon is common enough in linguistics with problems of phoneme recognition. The fact that discriminations are made means that some kind of autonomic filtering is taking place. The ard-ent describes this process, and extends it to filtering not only perceptions from the external environment, but also "perceptions" from the inner environment of cognition.

My iconic discription of what must be an immensely complex cognitive process here should be regarded, at best, as a kind of shorthand. It is nevertheless a useful shorthand whose slightly unfamiliar quality can remind us that the accumulated terminology of linguistics really performs a similar role. In Chapter 5 ard-ents are linked to the more traditional concept of linguistic meaning, as carriers for what I call encapsulated meaning, meaning which is stably coded and requires little or no further processing.

2.6 Ard-ents and the genesis of intelligence

At this point I believe the model has crystalized something of real explanatory power, and not just about language. All animals have autonomic behaviours. Most animals have contingent behaviours, and the more complex the animal the more complex its adaptive responses to an immediate situation. What non-human animals are unable to do, except in very restricted ways, is to learn adaptive behaviour to the point where it becomes autonomic. The striking thing about in humans is that with practice they can program extremely complex sets of stimuli to a point where they do become autonomic. The phrase structure rules of a language are perhaps the canonical example. Driving a car is another instance; and of course, the adherence to drilled routines by soldiers under fire is every general's dream. In other words humans evolved beyond some critical threshold where it became possible to translate mord-ent procedures into ard-ent filters.

Now here is the crunch. A herbivore spends all its time grazing to obtain enough nutrition. A carnivore has time to play at other things. Similarly, a complex animal with only a limited set of inherited ard-ent filters has to devote immense nervous resources to solving contingent problems on a one-off basis. My guess is that an animal which can translate complex but repetitive problems into ard-ent filter sets frees up those immense nervous resources for analytic activity above and beyond mere survival or propagation. Once the process begins it will become an accelerating spiral of cognitive resource economies. Herein, I believe, lies the key to the extraordinarily rapid emergence of human intelligence. Homo sapiens had built up huge central nervous systems to cope with increasingly complex survival needs. Suddenly these resources were left with idle time to make mischief. To borrow a computer engineering analogy, the moment biological ROMs <8> became biological EPROMs the species was unstoppable.

The hypothesis of intelligence from cognitive economy is attractive but extremely difficult to establish. Any proof would be indirect. If it could be shown, for example, that entirely base-generated languge exceeds human processing capacity and violates know heuristic cognitive procedures, then the case for something like an ard-ent would be very well motivated. Once the ard-ent was accepted as a necessary processing instrument, then the terms of its own evolution would tend to support the intelligence proposal.

If the ard-ent code (not merely a linguistic device here) is indeed a key to cognitive economy, the next question is how may we wield it at will to automate even more, and free up yet more of our intellectual resources. There is a further intriguing possibility here. Is there some way, some form of training, that can enable humans to discriminate the constituents of ard-ents which have become embedded as instinctual behaviour? Can we unpack the cognitive instruction sets that run our very life systems, and having learned their secrets, reset them for optimal performance? Perhaps the Buddhist's know something with their mindfulness/ awareness disciplines of meditation <9>. And perhaps for a linguist, formulaic language, dwelling on the margin of ard-ent types, almost but not quite at the level of autonomic performance, can tell us something about the boundary rules applying to this seminal phenomenon.

2.7 Generalization, symbolization, categorization

Another relationship to be explored is that between the process of generalization (as well as its derivatives, categorization and symbolization) and ard-ent types. Both generalizations and ardents have the common property of packing constituents into a single address for storage, recall and manipulation. The difference seems to be that ard-ent filter sets are not instantiated easily (at least, not after infancy). They are either innate, require an intensive learning process, or are a synthesis of the innate and learned. Generalization seems to be a far more facile process, easily packaged, modified or abandoned. The question is, which was the chicken and which the egg, generalization or ard-ents? My guess is that ard-ents were the egg: they are found in life forms at all levels, whereas generalization may be a particular gift of humans. If this is so, then the process that saw the biological ROM become a biological EPROM, may also have culminated in the much looser packaging of the generalization process.

If encapsulated meaning in ard-ents represents the special case of autonomic processing and recognition, meaning of any kind expresses a loss of innocence. This could be expressed in strong form by a "natural law of uniqueness": The Law of Uniqueness states that that which is unique in all its parts is invisible to nature. The real claim here is that meaning is fundamentally linked to the design of the organism, and the mechanism by which it is encoded is categorization:

[Varela et al. 1991:176] "One of the most fundamental cognitive activities that all organisms perform is categorization. By this means the uniqueness of each experience is transformed into the more limited set of learned, meaningful categories to which humans and other organisms respond. ...

[177] "In the enactive view, although mind and world arise together in enaction, their manner of arising in any particular situation is not arbitrary. Consider the object on which you are sitting, and ask yourself what it is. What is its name? If you are sitting on a chair, the chances are that your will have thought chair rather than furniture or armchair. Why? Rosch [1976] proposed that there was a basic level of categorization in taxonomies of concrete objects at which biology, culture, and cognitive needs for informativeness and economy all met. In a series of experiments, Rosch et al. found the basic level of categorization to be the most inclusive level at which category members (1) are used, or interacted with, by similar motor actions, 2) have similar perceived shapes and can be imaged, (3) have identifiable humanly meaningful attributes, (4) are categorized by young children, and (5) have linguistic primacy (in several senses). .. The basic level of categorization thus, appears to be the point at which cognition and environment become simultaneously enacted."

Some other linguists have also begun to see the categorization process which is so obvious in language as a marker for more gneral cognitive processes, and as with this study, have been forced to invent iconic machinery to discuss it:

[Varela et al 1991:177] "Mark Johnson proposed another very intriguing basic categorization process. Humans, he argues, have very general cognitive structures called kinesthetic image schemas: for example, the container schema, the part-whole schema, and the source-path schema. These schemas originate in bodily experience, can be defined in terms of certain structural elements, have a basic logic, and can be metaphorically projected to give structure to a wide variety of cognitve domains."

178. "Sweetzer provides specific case studies of [Johnson's] process in linguistics.She argues that historical changes of meanings of words in languages can be explained as metaphorical extensions from the concrete and bodily relevant senses of basic-level categories and image schemas to more abstract meanings - for example, "to see" comes to mean "to understand"."

2.8 G-ripples

The operation of the GO model is pervasively influenced by g-ripples (repeating entities). That which is to be given significance in language is always framed by known, hence repeated, elements. The organization of language is largely a matter of what is repeated, when, where, why, by whom, how and how often. The GO model takes a much broader view of repetition than is normally found in linguistics, considering a cline from local (often idiosyncratic) g-ripples to the global g-ripples such as lexical items which have become formal, generalized tokens in the language. Repetition so important to this thesis that it will be treated in a separate chapter.

2.9 Topics

The assembly plane is segmented by quasi-linguistic topics. These more or less correspond to certain organizing ideas we have as we are speaking or listening. In other words, topics act as mnemonic attractors (see attractors below) and provide foci for the (imperfect) reflection of focused ideas by linguistic form.

Note that "reflection" here is intended to avoid the patina of volition suggested by "representation". In the GO model we talk about forms co-arising in interactive environments, rather than subjective phenomena being modelled, by definition, on objective phenomena (which begs questions of ontology). This is not to deny that a kind of autonomic matching or adjustment between systems may occur, or when appropriate, that the adjustment may be volitional (in the sense of will-to-action discussed in chapter 1).

Topics also offer a stable framework for matching coherent patterns of idea-against-language while linguistic constituents are shaped into linear phonetic output, or as phonetic input is decoded.

Topics thus engage but are not fully defined by linguistic expression. There is always something more, such as background knowledge, to give them substance. They are never quite coextensive for speaker and listener, just as (in the GO model) conditions for cohesion are never quite coextensive; (cohesion and coherence are dealt with in Chapter 3).

Note that because we are not dealing with parallel but unlike universes of body and mind in the GO model, but rather with an inclusive ecology which embraces the outer and inner human properties, it is no special problem for an entity like topic to participate in both language and "ideas".

2.10 Attractors (dynamic systems attractors)

Attractors (dynamic systems attractors) in the GO model are a concept borrowed directly from the "attractors" of mathematical chaos theory. An attractor is essentially any factor which will influence the the development of a dynamic system. A post stuck in a river will shape the current flow downstream. More intriguingly, a "post", that is, a constant, stuck into any randomly developing (chaotic) environment will result in the emergence of ordered patterns, but not always the same patterns. A simple computer emulation of cellular behaviour described by Varela provides a good illustration:

[Varela 1991:88] "one of the most useful ways of capturing the emergent properties that various [connectionist] systems have in common is through the notion of an "attractor" in dynamical systems theory. .. Consider cellular automata ... a ring of cellular automata [which can have only two states, 0 & 1] acquires a dynamics by starting at some random state and letting each cell reach an updated state at each (discrete) moment of time in a synchornous fashion (ie. all the cells reach their respective states together). ...even this simple, almost minimal network has rich self-organizing capacities.... dynamically these rings fall into four major classes of attractors.. A first class exhibits a simple attractor, which leads all cells to become homogenously active or inactive. For a second .. class of rings the rules give rise to spatial periodicities, that is, some cells remain active while others do not. For a third class the rules give rise to spatiotemporal cycles of length two or longer. These last two classes correspond to cyclic attractors. Finally, for a few rules the dynamics seem to give rise to chaotic attractors, where one does not detect any regularities in space or time. ... in fact, it seems difficult for any densely connected aggregate to escape emergent properties; thus theories of such properties are a natural link for different levels of descriptions in natural and cognitive phenomena."

It is common enough in linguistics to talk of rules, constraints or filters. Connectionist theories would hold that in a massive connectionist system like the human brain, filters etc. are indeed attractors in the strict sense. In principle what a linguist will refer to abstractly as a "constraint" could have innumerable forms, from protein keys to synapse triggers to glandular secretions to harmonic electrical waveforms. In the complex processes of cognition all of these and more are likely to be involved. It is wildly beyond our technical ability to mathematically predict the interactive consequences of a few attractors in even simple lifeforms.

In the dynamic systems of natural language with their innumerable attractors of ever-varying significance, we can do no more than identify the most intrusive influences and the most general outcomes. Indeed, a certain level of indeterminacy is probably characteristic of the macro-system itself. Therefore discussions which follow in this thesis, and in any linguistic analysis, can never lay a truthful claim to precision.

2.11 Harmonic resonance attractors

"Attractor" is such a generally defined concept that more specific subsets of the class are needed for useful analysis. This is a good place to introduce a particular type which I will call "harmonic resonance attractors". In the GO model of language the notion of "harmonic resonance" often crops up as a way to describe certain effects of repetition, recall and synthesis; (other scholars have also found this term useful. See, for example, Hasan 1981, or Smolensky 1988).

The need for a model entity like harmonic resonance attractors arises from the observation that ideas, and language, often appear to be triggered by some minor event or recollection in ways that are quite disproportionate. The classic analogue for this in chaos literature is the butterfly in South America which causes a storm in the North Atlantic. Closer to home,it is in the same vein as a huge domestic fight being touched off by an innocent remark. In discourse we have the sudden branching into whole new topic areas as some apparently minor element of the conversation strikes a chord. Related processes might be at work sub-clausally in the quasi-random selection of particular phrase structures or expressions. This kind of discontinuity is not seriously broached in traditional linguistics, yet seems to play a major part in the generation of texts. At one level it might be predicted from the tension, the "generative oscillation", between representational language (that is, volitionally controlled language) and the ungoverned weed-like effloresence of linguistic strings within the organism, as discussed in Chap