Mediating Order in Chaos and Language
copyright Thor May 1994
Just as language change may be unpredictable up to a point, so the linguistic performance of any individual in a particular circumstance will be partly, but not entirely unpredictable. This study explores a few of the factors which can bear upon such constrained indeterminacy in synchronic language production, then proposes a model to account for the mixture of order and ad hoc adaptation found in typical examples of discourse.
Studies of linguistic variation have tracked performance ranges in minute elements of phonology and syntax (e.g. Labov 1975:65; Horvath 1985:67, 96); there have been discussions of pragmatic communicative limits in the speech group (e.g. Milroy 1985:339). Generative grammars have hypothesized inherent constraints on the form (and hence production) of possible grammars (Chomsky 1965, and hundreds of subsequent papers). Without discounting any of the work just mentioned, I want to take a slightly different tack.
First, I think it will be useful to firmly locate the complex dynamic
systems of language within the class of other such systems found in nature,
and to ask what properties all these systems share in common.
Second, I am going to ask some questions about the relationship
that the systems of language have to other cognitive systems, and to the
wider environment. The implication is that this relationship will have
important consequences for what we can say about the limits of indeterminacy
in linguistic performance.
Third, I will be searching for some analytic tools which may help in extracting evidence about the linguistic/wider-world relationship. One tool will be presupposition; another will be the use of formulaic utterance.
l. COMPLEX DYNAMIC SYSTEMS
Complex dynamic systems(l) often exist in a state of semi-equilibrium between a static state and chaos. It has been called "the edge of chaos", and is so finely balanced that the influence of a minute variable can have huge effects within the defining range of the system (Gleik 1987:16); (A geopolitical illustration would be the assassination in Sarajevo in 1914 triggering World War I). Since a large, complex system is influenced by an almost infinite number of minute interacting variables, it is impossible to predict outcomes precisely in any given instance. That is, there is an inherent indeterminacy about the system. This seems a prescription for chaos, but in practice there emerges a wonderfully complex iteration of patterns in which each apparent repetition clearly belongs to the whole, yet is never quite the same as any which came before. Apparently the myriad minor variations are rarely sufficient to alter the essential defining pattern.
1.1. Degrees of freedom
If the account of near-repetition just given sounds familiar to students of language, then that is no accident. What is particularly interesting about the behaviour of these systems is that gross deviations are not usually repeated: the indeterminacy operates within definite degrees of freedom. The mathematical equations of complexity theory generate some striking examples, but one of the most elegant is the figure-of-eight track which first caught the attention of Edward Lorenz ( 1963) as he studied weather cycles.
This indeterminacy of outcomes within limited degrees of freedom seems to operate at many levels within the systems of language. Patterns of diachronic language change and the synchronic language behaviour of individuals have already been mentioned. A less obvious example might be variations in the range of information encapsulated at a single lexical address by word-labels (i.e. the kind of variation addressed by so-called prototype theory: see Rosch 1978,1981, Lakoff 1987). Another important class of instances are near-formulaic expressions. We may notice repetition, clichés, parallel constructions and so on, yet a close examination of the data almost always shows variation from example to example within close limits. Constrained indeterminacy again.
1.2 The limited autonomy of systems
Another universal property of complex dynamic systems stems directly from the constrained indeterminacy of their behaviour. This property is a limited autonomy from all phenomena external to the system, an autonomy defined and guaranteed by independent variation. That is, although any number of external influences may impinge upon the system, the effect of any stimulation (or lack thereof) will be: ( 1) partly unpredictable; (2) often unrelated to the magnitude of the impinging effect; but (3) almost always within an expected range of variation for that phenomena (Gleik 1987:292-93; Lewin 1993:61 ).
This is Lorenz's butterfly effect (by now a cliché itself) where a beating butterfly's wings in one part of the world can whimsically result in a storm somewhere else, but not a storm which is significantly different from other storms. For the linguist who takes linguistic systems to be autonomous in some sense from other systems of cognition, this has to mean that stimulation by an idea, sensation etc. can lead to linguistic effects which are not always predictable, even to the speaker.
Equally, the language system itself might spontaneously emit structures through an ongoing performance momentum without intermediate stimulation at all. This notion is further explained and elaborated later. It extends to casual performance something that is already recognized in studies of language change. By way of analogy, some kinds of diachronic language change may be a product of the cognitive system's internal dynamic. An example is what Aitchison ( 1981:147) has called "pattern neating". Thus, English voiced and unvoiced fricatives have been tending towards a fully matched set since the eighteenth century : a process that has no communicative value at all. Pattern neating could be seen as a formal property of language. A claim in this paper is that not all utterance arises from a communicative need either. A natural language, in other words, is unlikely to be a purely representational vehicle for other realities, but is always likely to operate within language-like bounds.
1.3. Robustness of dynamic systems at the margin
The extraordinary thing is that while most complex dynamic systems appear to exercise their hair-trigger sensitivity to variation within limits, at the margin they become remarkably robust. Animal populations swing wildly, but rarely extinguish themselves; cyclones gather tremendous force but never beyond a certain point; ideologies dominate whole periods of human histories, but appear to be self-limiting. This behaviour seems to trace from what fractal mathematicians call "attractors", central defining influences which constrain the effects of linked variables (Gleik 1987:138). I wish to suggest that the production of language also has a certain self generating, unpredictable internal dynamic, within limits, but that those limits are secure enough to sustain a robust communication system.
1.4. A role for linguists
i) The limits of variability in language
If natural language is a complex dynamic system with some kind of autonomy and with the kinds of properties universal to such systems, as described in the preceding paragraphs, then the linguist's role is also proscribed in certain ways.
One task is to establish the limits of variability in all classes of linguistic phenomena. In many ways traditional grammars have surveyed this ground, although often without recognizing that statistical ranges rather than absolute boundaries were required. A related matter is to establish the dominant "attractors" in language generation, the defining influences which work to contain other variations within the acceptable bounds of a communicative system. The final equation for attractors in language is certain to be infinitely more subtle and complex than existing illustrations of iteration from fractal mathematics. In our current state of primitive understanding, we can take no more than a notional stab at what is going on.
ü) The limits of language as a system
Another task for the linguist is to show where limits are set by other cognitive systems, as opposed to the systems of language themselves. This is always a difficult matter, but there is a certain scope for hypothesis. For example, where the core rules of the grammar appear to permit certain constructions, but memory, social convention or other extrinsic factors preclude the usage, then we are arguably dealing with the limits of a system external to language. Of course, social (and especially) cognitive limitations have also shaped grammatical rules themselves over the whole evolutionary period. In other words, language systems are a product of co-evolutionary influences, although a dominant effect may be from environment to language (or vice versa) at different times. Because language and a multifarious collection of environmental factors are all in a never-ending oscillation of change and mutual adjustment, the historical process of grammaticalization for a total language system is never completed (even if specific grammaticalizations proceed to zero: Hopper & Traugott 1993).
Not all scholars accept an argument which assumes the relative autonomy of language from other cognitive processes. A more holistic view has been especially popular with psychologists (e.g. Piagetian followers such as Sinclair-de-Zwart 1969, and many others since). The arguments are often confused with debates about mind vs. body, physiology vs. "software", and so on. In practice the disagreements are often over research focus rather than substantive limits of cognitive operation. I do find it useful to distinguish some systematic autonomy. At the margin, particular examples of core grammar violation may be more ambiguous than others. Also note that a degree of autonomy in a system does not preclude that system from employing mechanisms which are common to other systems. For example, the Prototype Theory of Eleanor Rosch ( 1978), George Lakoff ( 1987:39) and others seems applicable to many areas of human cognition.
iii) The dynamics of relationships between systems
A third broad task for the linguist is to examine the dynamics of relationship between the macro system of language and external systems, and also between the sub-systems which exist within the macro system of language. Whereas general internal constraints on grammar, phonology etc. have been described persuasively in many models, these inter-systemic relationships have often been speculative. Physiological imperatives can have clear enough consequences for phonology (Ladefoged 1971:5), but the interdependence of phonology and morphology has always been more troublesome (e.g. Martinet 1965:91). The relationship between grammar and semantics or pragmatics is notoriously slippery. Nevertheless, the issues are significant, since relevant evidence will have consequences for understanding the manner and extent to which language is socially functional. Let us take the particular case of the language-to general-cognition interface:
(a) If language could be shown somehow solely to express instigating ideas or other impinging sensations, then its traditionally assumed role as a "mirror for nature" and symbolic vehicle for conveying ideas would be confirmed.
(b) On the other hand, if at least part of the time language emerged unmotivated by other cognitive stimulation, unrepresentative of ideas or topics, and purely as a function of its own internal dynamic, then the conditions applying to its management and prediction would be very different from (a).
(c) If language generation were initiated by external prompting, but then proceeded in ways and on a scale ungoverned by the original prompt, then to describe it as a "representational system" would be a severe distortion. I have already noted that complex dynamic systems are prone to such disproportionate "butterfly effects".
The kinds of relationship in b) and c) pose a major research problem. Acceptable scientific evidence in the conventional sense assumes a reasonably linear path between cause and effect, connections which can be made clear and predictable by constraining variables. However, where there is inherent indeterminacy within degrees of freedom, no single example can provide sure evidence. The researcher must finally work with aggregate statistical effects.
2. READING BETWEEN THE LINES
2.1. Presupposition as a translation point between language & cognitive systems. In discourse, knowledge which is assumed to be uncontroversial and known to both parties is said to be presupposed (Kempson 1977:70-72; Stalnaker 1978:321).
There are many conventions and distinctions attaching to the notion of presupposition, but we can note here that a
presupposition is bound to a speaker's present state of knowledge. Consider
(1) I realized that we didn't have a hope
(2) I realize that we don't have a hope
(3) # I will realize that we won't have a hope.
[Let # = infelicitous, as opposed to ungrammatical)
(4) I will acceptl? realize that we won't have a hope
if you can prove that their MIG 29s are really operational.
Sentence ( 1 ) is only acceptable if the past realization is concordant with the speaker's present belief; sentence (3) contains an implicit admission that the speaker is deluded, which the folk psychology accompanying all natural language use seems to exclude. Thus, although grammatical, it is utterly infelicitous.
The preceding example illustrates the fact that a language like English, following the logic of its own structures, has a the potential in certain configurations to misrepresent the boundaries of our experiential world. Furthermore, the failure is systematic, not incidental. There are points where, as it were, the language machine is bound to fail as a representational mechanism. Normally we would expect this to be no problem. If language is entirely the servant of other cognitive processes, sentences like (3) will simply not be uttered since they cannot be derived from any non-linguistic reality or any sane idea. Here we have an instance then of non-linguistic systems imposing a principled limit on the production of language.
Of course, if sentence (3), or something like it, were to occur then several hypotheses might arise about its source: (a) the speaker had suffered some mental aberration; (b) there had been a
"slip of the tongue"-a one-off error; (c) the systems of language were generating strings independently of and unmotivated by any external (non-linguistic) influence. In other words, the emergent language, although grammatical, was not representational.
In order to discount hypotheses (a) and (b), it would have to be shown that sentences like (3) were uttered by a sane person with reasonable frequency. In fact, I do not yet have evidence of a speaker behaving consistently in this way with that particular violation. However the kind of argument proposed, for identifying possible non-representational language generation might be applied to other contexts besides presupposition (an exploration I will forego in this paper). Presupposition itself, as an inter-systemic phenomena, also offers further scope for exploring the autonomy of language.
2.2. Presupposition versus counterfeit supposition?
The term "presupposition" claims to identify a link between social reality and coded linguistic forms. Linguistic studies have long distinguished between formal presupposition and discourse presupposition (Brown & Yule 1983:29). Formal presupposition is often claimed to be textually dependent and wholly defined by truth conditions (Keenan 1971:45). This interpretive certainty should actually make it a very reliable marker of social reality. Some verbs, for example, presuppose the truth of their sentential complement.
(4) I didn't realize (that the show was over)S
(4) presupposes that the show was over, whether or not I realized it. Other constructions are more ambiguous about the truth conditions holding between matrix phrases and their sentential complements. They require the consideration of non-textual factors, as Laurie Kartunnen once clearly showed in a discussion of "plugs, holes and filters". Consider his sentence (Kartunnen 1973:174):
(5) Sheila accuses Harry of beating his wife
The subordinate clause seems to presuppose that Harry has a wife, but strictly interpreted, the matrix clause does not. If Sheila were a tabloid journalist, it could well be the case that the inference from the subordinate clause was no more than a smear. In other words, the intent of the matrix agent can influence the truth of the complement. Such contexts often generate a postsuppositional effect which is more significant than the formal presuppositional claim. That is, the tabloid readers of Shirley's story "create" Harry's complicity: life imitates artifice.
The process of drawing inferences is an active one of deriving ideas by matching a form of words with known facts about the world. In the classic syllogism
(6) All men are mortal; Socrates is a man...
there is one clear inference to draw ("Socrates is mortal") and it may therefore remain unspoken. Such logically necessary relationships are at one end of a range which is marked at its opposite extreme by the linguistic assertion of relationship between a subject and a predicate:
(7) I am poor. [FOR ALL x, poor(x)], where x = speaker of utterance
A proposition like (7), standing alone, carries no further logically necessary inference. In between (6) and (7) is a vast population of possible suppositions that the decoder may extract from the form of words. Some of these suppositions may be almost coerced from the decoder, as in sentence (5); others may be based on no more than a personal prejudice about the world, such as
(8) I am poor [therefore] I am honest.
Although the pool of suppositions is indeed large however, it is not unbounded. Some linguistic suppositions are clearly more prototypic than others, to borrow Eleanor Rosch's concept (Rosch 1981 ), and it is questions about the range prototypicality to be analysed which underlies much disagreement over competing grammars.
Where does presupposition fit into the range of suppositions just described? It seems to me that a true presupposition is based on an unconsidered assumption by the encoder. That assumption is that the decoder will draw the same suppositions from the non-asserted elements of a message as the encoder holds. Hence the notion of a presupposition being uncontroversial.
What then do we call the coercive structure employed by the dishonest journalist in sentence (5)? This is a wolverine in sheep's clothing. Of course the intent of an encoder, like her private assumptions, is privileged information which the world cannot access directly. However our general experience of life should allow us to distinguish in principle between what is happening m sentence (5) and what is happening in sentence (4).
I propose to use the term counterfeit supposition to describe the kind of supposition that is forced onto a decoder by virtue of a form of words, but which is not shared innocently by the encoder. It is worth studying because where it dominates (for example, in propaganda) the communication cycle becomes corroded by cynicism. Note that the phenomenon differs from a regular proposition in which unshared information is openly asserted.
9. The Cycle of Communication
assertion & description of proposition
innocent supposition (presupposition)
acceptance / denial
How is it that what I have called counterfeit supposition is able to work as an element of communication? It may forego many of the shared conventions of language in a speech community, and covertly manipulate the more universal processes of strict logical inference. That is, the shonky journalist in sentence (5) could disclaim legal responsibility for any suppositions that readers draw about Harry's marital condition.
The prosecution might argue that this journalist had an
obligation to follow Grice's maxim , "Be Relevant" (Grice 1981). What both the maxim and the journalist trade on in different ways is that:
(a) deviant phenomena like counterfeit supposition can probably be slipped in while they remain only an intermittent element m the communication cycle.
(b) an interpretation will be imposed on any utterance (note the retrospective nature of this process), and
(c) the range of interpretation will be constrained by the expectations of decoders, even where normal linguistic conventions are overstepped. That is, decoders fit all messages, however deviant, to prototypic frames.
We might think back to Edward Lorenz's figure-of-eight equation and engage it as a metaphor. Take the first loop cycle as encoding and a subsequent loop cycle as decoding: the nature of the complex system itself will guarantee that the range of indeterminacy in either loop is robustly constrained within perceptual and experiential norms that are not particularly linguistic. This leads to a startling conclusion:
(d) The interpretation of language is able to exploit universal properties of complex systems without explicit provision in the grammar.
2.3. Supposition is superordinate to language
The process of supposition (including its specialized subset, formal inference) is one which typically employs the symbols, operators and patterns of language but which is itself superordinate to language. We can make suppositions from other kinds of information, such as visual cues. As the discussion about sentence (5) demonstrated, a defective linguistic frame may provoke supplementary interpretation from non-linguistic sources. However, the fact that there can be encoding and decoding to reliably extract suppositions from patterns of words is a kind of evidence for some general autonomy of language systems. It is conceivable that this autonomous system might sometimes take charge of the communication cycle itself
2.4. Emergent language versus representational language
I want to turn now to the interesting possibility of language which is not encoded, innocently or otherwise, but which nevertheless becomes available to be uttered. We are not dealing here with a relationship between interlocutors, but the relationship between a host cognitive system and its in-house linguistic systems.
For the sake of the discussion I define emergent language as linguistic strings which arise within a language system, but with no obvious motivation in the form of ideas, arguments, or other non-linguistic stimulation. Emergent language may or may not be actually spoken; (surely most of us also have "conversations in our head").
How credible is the idea of emergent language? What if it were the case sometimes that the language machine pre-empted any functional demand from other cognitive processes? What if phrases were projected into utterance merely because they had some salience in recent memory, rather than contributing to any social topic of conversation or train of argument? Or what if language came forth simply because the language machine was unable to stop, rather than because the host had anything to say? I do not find this proposal improbable. In fact, I will develop the speculation that such a process permeates language generation, and that the host gives retrospective meaning to utterance as often as considered meaning is given representation in language.
As an introductory experiment, readers are invited to apply a short test to themselves. Try not to think any language for five minutes. Trained meditators will know the problem. I can't do it unless absorbed in some motor activity like drawing; (for a general discussion of this kind of problem see Varela, Thompson & Rosch (1991:24). What is the significance of this exercise? It is not trivial. At first blush we might hypothesize that the cacophony in our head shows that short of becoming Zen Buddhist adepts, we cannot stop thought, thought which is then represented by language. But what do we mean by thought in this context? Is it necessarily the case that such language is representing anything?
We have just noted retrospective processes operating in all decoding which occurs between interlocutors. There seems no principled reason that an analogous relationship should not hold between a language system and its host cognitive system. Arguably however, the mind which decodes unmotivated language cannot be said to be dealing with assertive propositions (since no volitional intent is present), even if that is the grammatical shape of analogous conscious language. Nor does it seem sensible to talk of meaning "shared" between the macro-cognitive system and its autonomic language-system partner, in the manner of presupposition. Thus:
The process of decoding meaning from emergent language by a host cognitive system is perhaps best described as postsupposition by that cognitive system. Postsupposition is distinguished from propositions, presuppositions and counterfeit suppositions, which are sparked by Input from another speaker.
Postsupposition as a mechanical procedure is no different in principle from what we do in assigning meaning to the language of other speakers, as it impinges upon our senses through sound or writing. The mechanism is therefore perfectly credible as a behavioral category.
It is important to be aware of the differing ways in which the descriptive labels I have noted are identified in the grammar:
(a) Terms like "proposition" and "question" are inferred from certain lexico-grarnmatical configurations in English.
(b) The term "presupposition" is often derived from a lexico-grammatical configuration also, but so-called "discourse presuppositions" additionally rely on contingent knowledge of the world, shared by interlocutors.
(c) "Counterfeit supposition" is a type-category which cannot be identified with certainty in any particular instance, but can be recognized as a class from our general knowledge of the world. Formally, it is a special case of discourse presupposition in which the convention of shared assumption is violated and manipulated by the speaker.
(d) "Postsupposition" is a type-category which cannot be identified with certainty in any particular instance, but which may be recognized as a class from our general experience of mental behaviour. Postsuppositions can come in any lexico-grammatical configuration whatsoever. Formally, postsupposition is the private, post hoc process of assigning meaning to emergent (unmotivated) linguistic strings. They may or may not be generated as surface strings. The phenomenon can only be studied indirectly from analysing irregularities in the aggregate organization and meaning of texts, and the general psychology of language behaviour.
3. THE AUTONOMOUS LANGUAGE MACHINE
3.1. The problem of representation: historical views
An enterprise which sets out to give substance and credibility to the concepts of counterfeit supposition, emergent language and postsupposition cannot evade the existing assumptions which linguists bring to their study. In particular, notions of what language represents seem to be crucial.
One extreme view of the cognitive language machine is that it is a sort of tape recorder. A mass of data comes in through people's ears, is stored away, then selectively played back at an appropriate time. There are some obvious problems with this idea, especially when we think of creativity. Nevertheless, something like such a notion has been put by Michael Hoey (1991:154): "...each lexical item is stored more or less as received-in the context of the sentence in which it was used, rather in the way that it may be held in a computer store." There is no doubt that what is stored on an actual computer disk or tape is a "representation" of somebody's language: no speech, no recording. A tape recorder model of language in the mind would therefore be implicitly representational.
Another view of language has been that a finite set of rules is applied to a finite lexicon to generate a potentially infinite set of sentences. This mechanical generation of sentences could in principle be independent of any contingent reality, and at least early models of Chomskyan generative grammars placed emphasis upon the autonomy of syntax. Nevertheless, the systems of generative grammar (and the Artificial Intelligence models which were developed in close parallel), invariably claimed to be symbolic systems (Chomsky 1965:3, 222; Minsky 1986). That is, the generative engine was motivated by an extrinsic role, which was to represent a non-linguistic reality in coded form. The strong form of such models took thought itself to be a coded representation of external (public) reality, a notion sometimes called "the cognitivist hypothesis" (Varela, Thompson & Rosch 1991:8).
A third approach to the problem of representation comes from functionalists. "Function" is probably the most abused word in the social sciences. However, the general thrust of functional linguistics in all its permutations has been an underlying conviction that the patterns of language are pragmatic adaptations to the requirements of representing an external reality. Sometimes these functional adaptations are seen in the sociolinguistic minutiae of Labov-style variation, sometimes in sweeping categories of linguistic purpose such as Halliday's ideational, interpersonal and textual semantic functions (Labov 1975:65; Halliday 1985:53). In diachronic linguistics the functional approaches sometimes endorse neo-Darwinian evolutionary adaptation as a prime source of language change. What functional models all have in common is a belief that significant motivating sources of language structure and linguistic behaviour are external to the systems of language themselves.
The models of language just mentioned, the naïve recorder model, the generative and the functional, have assumed language to be, more or less, a passive instrument, reflecting the outer and mental worlds like a mirror reflects light. In most of these models, language itself (sometimes as opposed to imagination, memory etc.) can no more initiate reality than a non magic mirror can show other faces. At best it can constrain the representation of reality by its design limitations.
I find these analyses inadequate. Although language certainly is a superb tool for encoding and schematizing other experience, it also generates new experience in ways that are often unpredictable.
3.2. The likelihood and consequences of emergent language
Most linguists would probably accept that the internal dynamics of a language's structure give rise to some diachronic language change. These variations are expressed at every historical stage through performance which in single utterances might seem to be entirely driven by the representational purpose of the moment, but are not. If the language machine makes such constant small adjustments, what is to prevent it in discourse from inserting a constant embroidery of grammatical but pragmatically unmotivated phrases (which might be rationalized post hoc, like much other behaviour)?
Utterance is the natural systemic outcome of what I have called the "language machine". A total global constraint on the linguistic system itself initiating utterances seems intrinsically improbable, and out of kilter with the general behaviour of dynamic natural systems. inguistic research is dominated by teleological concepts: the final cause of our speaking is assumed to be that "we" "want" some communicative "effect". A meteorologist has less trouble accepting that the general cause of a storm derives from the locally unpredictable internal dynamic of weather systems themselves (though the religious insist on final cause by a deity). The point here is that in important respects, natural language systems and weather systems belong to a similar class of phenomena. Although language is affected by constant teleological interference, it also has the internal dynamic common to weather systems, demographic systems, economic systems and so on. This paper has already alluded to such a general theory of complex dynamic behaviour, and a recent branch of applied mathematics has been built upon it.
Consonant with the preceding paragraph, and what is now known about chaos and complexity theories, it seems doubtful that alone amongst natural systems the responses of the linguistic system to external stimuli should always be a precise, proportional and reliable reflection of those stimuli, which is what an extreme representational claim amounts to. Language often subverts our best intentions, either by omission or by uncontrolled acts of commission.
Although the notion of complex dynamic system itself has been partly defined in this paper by a principle of constrained indeterminacy, the robust limits of those constraints mean in practice that interacting natural systems are mutually correcting. In the cognitive sphere, this suggests that if utterances were to arise without any motivating ideas, the language generated would quickly be sustained for meaning by a process of interaction with memory, perception and so on. In other words, the speaker would keep his sanity by assigning a meaning to the utterance and following up by controlled discussion to give it a context. Such a process would be instantaneous and automatic. Here I am doing no more than extending to organism-internal procedures the behavior which automatically occurs when a person receives a message from someone else.
3.3. Is "emergent language" a credible description of certain mental experience?
I have seen a droll computer program called Babble (Korenthal Associates Inc. 1991 ) which generates something that looks like language. Since it is the product of an algorithm rather than an intelligent mind, I have some trouble attributing either presupposition or intentional assertion to its output. Here is a brief sample:
(9) Good news that comes your way might touch that comes your way might be but, speak again, soft! Arise, who is my love?
Example (9) is literally word-babble, yet perhaps not beyond an imposed meaning with an invented context. It is mentioned because something akin to the Babble program may well be at work in the brains of sentient human beings.
Suppose we take an infant, nurture it with love, give it every attention, talk to it constantly, watch it acquire our language to the point where by five or six years of age it has mastered a communicative instrument with skill and grace. Does this child now possess some independent self which entirely commands the language instrument? That is, can the language instrument be laid aside when there is nothing specific the child feels that it must say? What if there is no social imperative to speak? Will the language instrument be quiet when there is no problem to solve or no memory to dwell on? If my childhood was typical of most others, the answer to all of these questions is no.
Who hasn't spoken when every bone in their body screamed that it was smart to stay silent? Most of us have been caught on the crest of countless unexpected phrases and been forced to surf them to unhappy outcomes. We have listened miserably to our bumbling voice turning away potential friends, and wondered at the inexorable logic of our utterance as it painted our saner being into indefensible corners. Each of us feels somehow that "it wasn't ME creating this havoc". We could claim that it was the wretched language machine which, like Fantasia's Sorcerer's Apprentice(2) (Walt Disney 1951), having learned to carry buckets, wasn't going to stop for anyone.
If we accept, for the sake of argument that language is not an entirely dependent mechanism, that it is a dynamic instrument which has co-evolved with other dynamic systems of the organism, and has the capacity to initiate cognitive activity, then some quite difficult questions arise about the representational status of silent soliloquy, not to mention public utterance.
3.4. The rationality of emergent language
Why should emergent language amount to any more than the proverbial ninety-nine blind monkeys trying to write Hamlet on a typewriter? Why should it be cogent at all? There are several complementary answers to this.
· Firstly, most automatic output from the systems of language will conform to the general rules of those systems. In a sense, it should be more difficult for such a system to generate an ungrammatical string than a grammatical string, just as it is more difficult for us to utter a non-English phoneme than an English one.
· Secondly, emergent language should be more likely to accept a conventional meaning than a meaning of the "colourless ideas sleep furiously" variety. That is, the collocations which self select are almost bound to be conventional ones.
· Thirdly, for reasons yet to be discussed, emergent language should predominantly consist of only short stretches of language (even down to single words) and these will usually be seamlessly integrated post hoc into the normal representational discourse. We can predict, however, that when the post hoc integration breaks down (from a failure of concentration, or whatever), then the fragments of emergent language will have local grammaticality (the first point, preceding), but may exhibit a global incoherence.
Finally, all of the preceding arguments are local instances of the global principle which has given rise to this paper: the principle of indeterminacy within limits in complex dynamic systems. Emergent language will have a form which deviates from controlled representational language unpredictably, but only within certain degrees of freedom. Think again of Lorenz's figure-of-eight track.
If the language machine does behave in this way some of the time, then some of the constraints which define it will have evolved precisely to handle the self initiating, emergent aspect. This is a prospect which, in the form I have discussed it here, has escaped the attention of linguistics. Its closest cousin in existing research would be studies concerning the effect of structure on diachronic language change.
3.5. Pastiche talk
In practice we stumble in and out of emergent language, sometimes in control of a genuinely premeditated message, sometimes hostage to the unplanned form and structure of what has fallen from our lips. It is this scrambled rush of the calculated and the accidental that I call pastiche talk. A rough diagram of the process might be thus:
.......+.............==-- COGNITION: ACT TO SYMBOLIZE =-- TOPIC CONTROL ==--
==-- SYNTACTIC CONTROL ==-- CONTROLLED DISCOURSE
==-- LANGUAGE SYSTEMS
........................+....................==-- SELF- GENERATED LANG. ==-- PASTICHE TALK
(where ==-- is an arrow. It is hard to draw in ASCII!)
Does it really matter, from the point of view of linguistic modeling, whether our language is a purely representational system (as has been assumed up to now), or whether it is normally a hybrid system along the lines of pastiche talk? In fact, there are likely to be consequences for both synchronic and diachronic models.
In synchronic analysis a pastiche talk model will predict a principled discourse divergence between what the generative grammarians used to call competence and performance. Generative models were based on the supposed behaviour of an "ideal speaker-hearer". It was acknowledged that actual language behavior differed from this idealized source, but the differences were held to be accidental, unsystematic and of no serious consequence to the knowledge of the language that real interlocutors had. The principles of symbolic representation would not be affected by temporary glitches in output. This model retains some credibility at clausal level if my hypothesis about the conventional nature of emergent language is true. In an extended pastiche talk context however, the existing competence model becomes a nonsense.
If utterance, far from being a unidirectional representation of some cognitive reality, is in fact the outcome of constant negotiation for primacy between a language machine on the one hand, and impinging cognitive pressures of memory, topic, idea synthesis etc. on the other, then at any given moment the outcome will be a compromise of almost quantum unpredictability. The tidy categorial restrictions of generative models become no more than intermittent states. The wider pattern will inevitably translate as statistical tendencies in a pastiche model. In this sense Labov-type patterns of linguistic variation acquire an extra credibility. The twist is that such variation is no longer held to be almost entirely sociological, with a small quota of slips. Rather, a pastiche model will predict that even in a wholly homogeneous social environment (if such a thing were conceivable), significant linguistic variation is inevitable.
3.6. Some diachronic consequences of pastiche talk
If some synchronic variation in language generation is inevitable, regardless of social conditions, then we have one more principled reason for diachronic language change. The push-pull alternation of primacy between language and cognition would mean that the whole communication system had, as a design feature, a limited openness which could always admit some change. This would differentiate natural languages fundamentally from the linear representational behaviour of computing languages.
Historical linguists have had some success in correlating changes in social conditions with language change. Is it possible that changes also occur in populations, in the balance of emergent language to representational language? To speculate for a moment, the move to literacy might effect such a balance. So might the visual dependencies of television junkies. If so, this would predicate another kind of language change.
3.7. A sample of pastiche talk (?)
The following is an extract from the spontaneous utterance of a 55 year old Australian woman. Her conversation has drifted onto the topic of immigration, but I invite the reader to consider whether this speaker is thinking before she speaks, or speaking within the general penumbra of that topic before she thinks.
Extract 1 [Connie Hume Corpus]
Key:  signals an intonation unit boundary. Each boundary is numbered. ### Underline signals a probable formulaic structure, or highly preferred collocation. C: = Connie Hume.
Some related lexemes have been bolded.
(1) C: Yeah
(2) No well this is it
(3) the cou
(4) I know myself
(5) as I said to Alma when I first come over here vou
(6) to mv way of thinking then too
(7) as I say I've been in many a couple of them by the dagoes
(8) you know
(9) and thev've thrown out illegal
(11) and all this sort of thing
(12) and they're come in by the drav load
(13) we can't feed and house our own
(14) and jobs
(15) and this sort of thing
(16) and with inflation
(17) I thought God
(18) there're all going to get the dole
(19) or end it up
(20) you know it's a1l going to cost them in hospitals and things like this
(21) adding to it
(23) this was mv way of thinking then too
(24) and then vou see big televisions
(25) yöu know
(26) coming ofj`'the planes with fur coats and God knows what
(27) I'm thinking
(28) migrants! (29) refugees!
(30) have a look at it
(31) you know
(32) and this is iust how I think
(34) T: Actually I think
(35) C: so
(36) I can't blame anyone else for thinking
(37) and it's not
(39) you sort of stop to think
(40) and vou get into a thing like this that
(41) you know
(42) you start to realize
(43) vou know_
(44) things then
(45) that the difference
(46) with them.
If pastiche talk is a viable concept at all, this Connie Hume extract must surely illustrate one of its less controlled extremes. This language is the polar opposite of the premeditation found in counterfeit supposition. The listener must decode language which the speaker has generated without any apparent global plan. The listener can scarcely deconstruct what has not been constructed. Instead, he is forced to deconstruct the known cultural parameters of the speaker, and then suppose an interpretation within those bounds.
3.8. Global incoherence; local meaning
There are a number of features about the Connie Hume extract which are quite striking. The first is its overall incoherence. This is exceptional for the speaker, Connie Hume, whose pastiche talk is normally leavened by a more obvious control of topic. Even a listener who is tuned into the nominal topic in this extract has to work quite hard to guess what Connie is getting at. One's best hope seems to be to interpret the general atmospherics of local phrases and hypothesize a larger message. However this doesn't work well since the tone of earlier phrases we can't feed and house our own") is apparently contradicted by the more conciliatory ending ("you start to realize..."). Most of the ties, logical operators, sequencing markers and so on which might give the whole a thematic unity are missing.
The incoherence of Connie Hume's speech is utterly different from the incoherence encountered in Second-Language learners of English. For them the problem is inaccurate lexical selection, inappropriate collocation, and an insecure command of syntactic assignment. Connie Hume, on the other hand, has a highly fluent local command of the idiom. The main problem occurs above phrase level, where the ramshackle assembly has all the appearance of distracted workmanship. My guess is that the phrase-level utterance is indeed emergent language, and that Connie has been unable to take proper command of the potential postsuppositions with which she has been confronted. She is being driven by a language machine that is for the moment out of effective control.
3.9. Emergent preferred-collocations and formulas
A second property of the extract is that almost all the intonation units taken in isolation convey a sense of deja vu, as if we have heard them before somewhere. This kind of highly preferred collocation, shading into formulaism, is actually very difficult to establish statistically since the collocations are typically subject to minor shifts m form when repeated; (you could compare it geometrically to the affine transformation which occurs when you view a television screen from a slightly different angle, distorting it slightly). Nevertheless, these preferred collocations which are often opaque to searches on linear computer programs, are immediately evident to our own cognitive mechanism. We recognize not only the form of words, but the cultural patina and social interpretation which typically goes with each collocation.
It could be argued that preferred collocations and formulaic phrases of the kind found in the extract are "semi-encapsulated" in linguistic memory. The idea of encapsulation is that the internal constituents of such a unit are more or less pre-processed cognitively and don't have to be regarded analytically in normal usage. Thus the morphemes in a word are not normally subject to analysis (and hence a processing load) before we use the word. Formulaic phrases obviously vary across a continuum m the extent to which they are encapsulated Nor will all encapsulated phrases in an idiolect will be general to the speech community. There is no doubt however that all such constructions entail tightly bound collocations which are easily evoked as a unit.
Encapsulation has found some support amongst other researchers. Bybee ( 1985:7) suggests that "it is simply not necessary for human language users to segment every sequence into its minimal parts, because it is possible to acquire, store and access complex chunks of material without segmentation". She goes on to explore studies of errors in child language acquisition in various languages which support the storage and recall of phrasal chunks. (Bybee 1985:114).
Because phrasal formulas are not internally constructed according to the contingent needs of representing some other reality, their constituent lexemes may have little or no obvious referents in the context of situation. Extreme examples are frozen forms like "kick the bucket" where even the original metaphor is no longer transparent to lexical meaning.
The pre-formed nature of preferred collocations and phrasal formulas should make them prime fodder for a language machine running on automatic pilot. They carry a small packet of conventional meaning which is more substantial than individual words, and thus conveys some impression of local coherence. A heap of such formulas shoveled into the general context of a topic like immigration, as in the extract, can give a superficial impression of utterance driven by rather woolly ideas. It seems here in fact that there are some woolly ideas here driven by talk. The pressure to talk is primary.
3.10. Presuppositions, counterfeit suppositions or vagueness?
If language like the kind just discussed is indeed mostly emergent rather than representational, then it doesn't make much more sense to talk about it embedding presuppositions than it does to attribute presuppositions to the Babble computer program output referred to earlier. At least, the non-presuppositional content of emergent language holds when presuppositions are treated as qualities of thought rather than artifacts of linguistic form (such as where the sentential complement of realize is said, ipso facto, to embed a presupposition). To the extent that a listener can make sense of Connie's monologue above, it is by a process of hypothesizing a best fit for its fragments of meaning against existing knowledge of the world. The listener, in other words, can have no confidence in speaker-intent, as one does in interpreting true presupposition, and counterfeit supposition. The resulting vagueness requires action that is closer to an external version of the internal postsupposition practiced by a user of pastiche talk.
3.11. Some cognitive correlates of emergent language
Evidence for the frequency and nature of emergent language may have to be sought indirectly. This is a matter for empirical analysis, probably with a statistical emphasis. In more general terms however, we can identify the cognitive and environmental conditions which are likely to give rise to a significant amount of emergent language. I have already mentioned the mnemonic dependencies of pre-literate societies as one possible source.
a) Weak monitoring; the aphasic extreme
Amongst untrained users of emergent language (most of us) the surface text itself may be marked by erratic topic changes, poor semantic integration or obsessive repetition. Such signs of weak supporting ideas are sometimes accompanied by linguistic fluency, a rapid output of great quantity but low quality. Individuals vary greatly in the discipline of their social speech habits, and there is every reason to believe that this variation extends to mental life as well.
The excessive intrusion of emergent language would seem, intuitively, to be more likely in an undisciplined environment, either by reason of lifestyle or from the relaxation of the moment. An extreme de coupling of language and thought occurs in certain kinds of aphasia, where an individual talks with speed and fluency, but in an entirely vacuous manner; (we will all recall acquaintances who seem to dwell towards the vacuous end of this aphasic spectrum!).
The conventional view about both verbally rambling individuals and aphasics is that vacuous talk reflects vacuous or disassociated thought. No doubt there is some of that. It is not intended to claim that even a babbling charismatic in a trance is not drawing upon some train of memory and association. The suggestion is merely that the oscillation of pastiche talk may be less constrained under these kinds of conditions. Note that dissociative breaks could equally occur between perceptual reality and cognition, or between cognition and the language machine. Where the structures are disciplined but the content is vacuous then emergent language seems a fair hypothesis.
b) The dynamic momentum of pastiche oscillation
On the other hand, it is at least worth considering as a hypothesis that the generation of language which is exclusively dependent on prior thought, utterly representational, may be unsustainable in normal discourse. Of course, scripted lines can be learned or read for particular occasions but this is no model for ongoing natural language generation. If it were machines could replace humans in many tasks immediately.
It may be that a certain amount of looseness, an oscillation in the prioritizing of thought over language and its opposite is necessary for the system to maintain its dynamic and to evolve. Engineers should recognize this paradigm of controlled degrees of freedom: most complex moving machines depend upon it; (it is the buffer clearance between a piston and its cylinder that permits it to reciprocate).
Another kind of engineering analogy may also be relevant to the action and reaction between the language engine and the larger cognitive engine. It is the momentum of a flywheel which keeps an internal combustion engine rotating smoothly between cylinder explosions.
It could be the momentum of the language engine, freely generating structure, which takes up the slack in the flow of ideas, permitting a fluency that strict thought-to-symbolic-representation alone could not sustain. Yet the system is governed, prevented from spinning out of control by an oscillation between boundaries in cognition and language which is absolutely typical of the "edge of chaos" dynamic in complex natural systems.
Rapid reversals of symbolization, where thought rationalizes (i.e. represents ) language, as it were, may be a major and necessary contribution to the development of extended argumentation. In fact, this is more than hypothesis, for any researcher knows that the same process, externalized and slowed down, is at the heart of wanting. This paper could not be written without the constant stimulation of nascent ideas by existing language structures on the computer screen-and I say "language structures" on the computer screen rather than ideas" on the screen advisedly, for it is I who must construct ideas from the visible symbols.
The genesis of ideas expressed here is found in an explanatory and analytic problem I faced. That problem was a mismatch between theories of linguistic competence and the kinds of divergence from competence illustrated by Extract 1 in the text above. Clause-level units conform overwhelmingly with the structures we should expect in well formed language. However two currents of evidence seemed to call into question classical generative models. Firstly, a high percentage of the clausal strings were recognizable as frequent collocations in the language community, or in the speaker's idiolect. This suggested that some form of heuristic encapsulation played an important part in the process of language generation. Secondly, speakers showed an intermittent tendency to lose control of text generation above the clausal level. Various explanations could be proposed for such failures of coherence and cohesion. Could it be that the encapsulation actually subverted the generation of communicatively appropriate structures?
I chose to explore the possibility that semi-encapsulated structures in language could emerge spontaneously and irrespective of the non-linguistic context of situation. Language thus became a co-creative system which did not always merge with the communicative intent of a speaker. This proposal raised very substantive philosophical questions. It seemed necessary to contextualize the issue within general patterns of natural phenomena.
The presentation of this paper thus began at a point to which the linguistic problem had driven me: a search for general phenomena m nature resembling the patterns of natural language generation. In fact, language turned out to behave in ways analogous to countless other complex dynamic systems. Everywhere, from the genetic code to weather cycles, there was a conformity of micro-generative rules marred by a macro unpredictability within strict limits. Was such indeterminacy a problem, or was it the property which gave such systems their very dynamic? The latter seemed more likely.
If language were a quasi-independent entity, then the very margins of its interpretation could be the best place to look for evidence of breakaway from the strict role of representation. Studies of presupposition have always been controversial in this regard. It was shown that well-formed grammatical structures could systematically violate folk conceptions of reality. The notion of counterfeit supposition was then introduced to identify the deliberate perversion of presuppositional conventions. Finally dialogue negotiation between interlocutors was shifted analogously to a negotiation between a sentient mind and its available "language machine". Who was representing what to whom in this inner arena? Where the language machine generated spontaneous, unmotivated strings, it was inappropriate to talk either of propositions or presuppositions. The term "postsupposition" was proposed to explain post hoc rationalization by the sentient mind of such emergent language.
Pastiche talk was the outcome proposed for a dynamic oscillation between the linguistic representation of ideas and a postsuppositional logic imposed by the mind on emergent language. Pastiche talk was a perpetual compromise that sustained a fluent though imperfect flow of discourse which was never quite predictable or ordered, yet kept just within bounds on the stable side of chaos.
Mathematical complexity theory has demonstrated that large, dynamic systems have self organizing properties (Lewin l993:27). In what is called a random Boolean network of interacting items, "attractors" always emerge (patterned nodes which resist re absorption or mutation), and the number of attractors is always roughly the square root of the number of items in the system. For example, the 100,000 genes in the human genome give rise to 254 cell types, credibly close to the square root value of 370.
Those who grew up on Walt Disney will know his epic cartoon film Fantasia, where animated creatures, working to an inexorable progression in the music, show the dreadful logic of unthinking efficiency run amok.
Aitchison, J.1981. Language Change: Progress or Decay? Bungay, Suffolk: Fontana.
Brown, G. & Yule, G.1983. Discourse Analysis. Cambridge: Cambridge University Press.
Bybee, J. 1985. Morphology: a Study in the Relationship between Meaning and Form. Amsterdam: John Benjamins Publishing Company.
Chomsky, N. 1965. Aspects of the Theory of Syntax. Boston, Mass.: M.I.T. Press. Cole, P. (ed.) 1978. Syntax & Semantics 9: Pragmatics. New York: Academic Press.
Cole, P. (ed.) 1981. Radical Pragmatics. New York: Academic Press.
Gleik, J.1987. Chaos. London: Cardinal.
Grice, H.P.1981. Presupposition and Conversational Implicature. In Cole (ed.)
Halliday, M.A.K.1985. An Introduction to Functional Grammar. London: Edward Arnold.
Hoey, M.1991. Patterns of Lexis in Text. Oxford: Oxford University Press.
Hopper, P. & Traugott, E.1993. Grammaticalization. Cambridge: Cambridge University Press.
Horvath, B.1985. Variation in Australian English. Cambridge: Cambridge University Press.
Kartunnen, L.1973. Presuppositions of Compound Sentences. Linguistic Enquiry 4,2:169-193.
Kauffman, S. 1992. The Origins of Order: Self organization and Selection in Evolution. New York: Oxford University Press.
Keenan, E. (ed.) 1975. Formal Semantics in Natural Language. Cambridge: Cambridge University Press.
Kempson, R.1977. Semantic Theory. Cambridge: Cambridge University Press.
Korenthal Associates Inc.1991. Babble. Version 2.0 (computer program). New York.
Labov, W.1975. Language in the Inner City. Philadelphia: University of Pennsylvania Press.
Ladefoged, P.1971. Preliminaries to Linguistic Phonetics. Chicago: University of Chicago Press.
Lakoff, G. 1987. Women, Fire and Dangerous Things: What Categories Reveal About the Mind. Chicago: University of Chicago Press.
Lewin, R.1993. Complexity: Life on the edge of chaos. London: MacMillan, Phoenix.
Lorenz, E.1963. Deterministic Nonperiodic Flow. Journal of Atmospheric Sciences 20:130-141. Lorenz, E.1963. The Mechanics of Vacillation. Journal of Atmospheric Sciences 20:44864.
Lorenz, E.1964. The Problem of Deducing the Climate from the Governing Equations. Tellus 16:1-11.
Martinet, A.1973 . Morphophonemics. In Fudge, E (ed) Phonology. London: Penguin.
May T. 1993. Aspects of Repetition. In University of Melbourne Working Papers in Linguistics 13:79-90.
Milroy, J. & Milroy, L.1985. Linguistic change. Social network and speaker innovation. Journal of Linguistics 21:339-384.
Minsky, M.1986. The Society of Mind. New York: Simon & Schuster.
Pawley A.1985. Lexicalization. In Tannen D & J Alatis (eds.) Languages & linguistics : The interdependence of theory; data and application. Georgetown University Round Table on Languages & Linguistics.
Rosch, E. 1981. Prototype Classification and Logical Classification: The Two Systems, in E. Scholnick (ed.) New Trends in Cognitive Representation: Challenges to Piaget's Theory. Hillsdale, N.J.: Lawrence Erlbaum Assoc.
Schourup, L.1985. Common Discourse Particles in English Conversation. New York: Garland.
Sinclair-de-Zwart, H. 1969. Developmental Psycholinguistics. In Elkind, D. & Flavell, J. (eds.) Studies in Cognitive Development. New York: Oxford University Press.
Stalnaker, R.1978. Assertion. In P. Cole (ed.) Syntax & Semantics 9: Pragmatics. New York: Academic Press.
Varela, F. J, Thompson, E. & Rosch, E.1991. The Embodied Mind. Boston, Mass.: M.I.T. Press.