Chapter III, part 2

III.D. Evolution in language and discourse

104. Since ancient times, wide-ranging accounts have been offered for the origin of language. Perhaps language was a gift of the gods; or a sudden invention of an inspired caveman; or an extension of human grunts and cries; or an imitation of sounds in the environment, such as animal noises. But, divine intervention can explain everything in general and hence nothing in particular; inspired inventions cannot be made until there are concepts to connect; the noises made by humans on their own or by imitating animals don’t produce a language but at most a primitive ‘proto-language’, such as human infants develop (III.111, 130-33). Such accounts dismally fail to explain the evolution up to the rich amplified design that all known languages manifest (cf. III.111).

105. The account developed in sections III.A-C might offer a more auspicious basis. As before, we can °start from a highly sparse model and enrich it with constraints° (cf. III.54, 69, 103). The emergence of order out of either stasis or chaos was made possible, indeed probable, through the evolution of °adaptive self-organizing complex systems° (cf. III.53-59). Human cognition could have emerged from a similar though far more amplified evolution wherein minimal entities evolved into objects, these into phenomena and all three categories into systems, schemas, and world-models, which evolved in their turn (III.70, 85). Each °critical mass in this evolution brought its own emergent properties and amplified the dialectical interaction between the substrates of material and data by enriching the constraints and softening the coupling°.

106. The same account might be applied to language °emerging from the data codes already inherent in all life-systems° (III.79ff) but highly amplified only in humans and in some species of animals (cf. III.107). Language would have been elaborately prepared for emergence through the evolving design of certain life-systems. Far from being somebody’s individual invention, language was a product of an already established system of social interaction. This system co-evolved with human processing capacities until the stage when the code could pass from operations within a system to operations between systems and could vastly enrich the social interactions by enabling humans to coordinate plans or goals and to share and tune knowledge and experience far better than before.

107. The °critical mass for this ‘externalization’ would give language the emergent properties° needed for one life-system (the text producer) to °condense a rich data field into the sparse data field of sequential expressions coupled to a material substrate of uttered sounds°, and for another life-system (the text receiver) to °re-amplify this sparse data into a rich data field whose design is tuned to fit the original field° (Fig. III.28) (I.34; III.48, 54, 92).  The  potential  of  the  language would

spontaneously amplify with the design of the life-system supporting it. ‘Lower-order’ animals like ants or bees are harder-coupled and communicate by investing high energy to transmit sparse data about scripted situations (e.g., location of food or enemies), while the communal text converges from group behavior (e.g. lines of movement). ‘Higher-order’ animals like humans are softer-coupled and invest low energy to transmit rich data, while the text appears more individual than communal because the individual is more explicit and the communal  is  more  implicit  (cf. I.39). The gradation between ‘higher’ versus ‘lower’ animals has often been misconstrued as a total dichotomy, whence the old anthropocentric notion that only humans have language.

108. In this account, the design of the ‘external’ code should resemble the design of the ‘internal’ one. So communication would not fit the familiar model of ‘encoding’, ‘sending’, ‘receiving’,  and  ‘decoding’  a  ‘message’  from some non-code (I.53; V.96; VII.209) (Fig. III.29a), but would more

properly be a transaction of recoding. The speaker as ‘recoder’ starts from a representation in  internal  coding and moves to external coding, while the hearer as  ‘recoder’  moves  back to internal coding (Fig. III.29b). The internal coding is the operational implementation for cognitive entities such as ‘meaning’, ‘thought’, ‘ideas’, and ‘concepts’; these are all rich data fields whose shared design principles offset potential differences among the modalities of coding (e.g. visual versus acoustic). In principle, they are all recodable into language, whether or not they actually get recoded. The greatest advantage of language lies in being the most comprehensive and adaptable modality of meaning and the most conducive to explicit, intentional transactions of control (cf. III.216; VII.97, 291; VIII.131). Talking about things and what they ‘mean’ helps define, monitor, and manage them far more effectiv ely than just thinking about them or imagining their visual appearance (cf. III.216). Or, you can easily decide not to talk about them, even when you should be talking about them or when you can’t help thinking about them (cf. III.88). Evidently, the social is better controlled than the cognitive, so that discourse is made the chief modality for sharing knowledge, even though just how it might best do so is still poorly understood (cf. VII.93).

109. What modes of supportive evolution could produce, sustain, and change a language? We can identify at least four modes, each with its own participants and temporality. First, language evolves in the species and supports its genetic and adaptive capacities. Second, language evolves in the community of speakers and supports their culture. Third, language evolves in the individual during ontogenesis and maturation and supports socialization and education. And fourth, language evolves in the participants during discourse interaction and supports their orientation. Inquiring how these four modes of evolution could affect the organization of language might help to clarify the tough questions of who ‘knows’ a language, what it is they ‘know’, how they come to know it — and how more people can enhance their knowledge of language (cf. V.15; VI.30).

110. For evolution in the species, we can inquire what internal design would be required to support the emergence and survival of a language. Either the design would have to be genetically transmittable; or the emergence would have to reoccur for each generation; or the language would have to rely solely on being taught by adults to offspring. A language which kept re-emerging or had to be expressly taught to offspring who had no genetic support for it, would be insecure and would have to remain unproductively simple. Surely genetic transmission would be more reliable and productive, and developing it would favor the survival of the species. But how language-specific might the genetic material be in view of the stunning discovery that populations with different language families also differ in their genetics (II.19)? The physiological capacities would be specified for the construction of the vocal tract, which apparently stabilized for consonants before vowels; and dental evidence indicates specifications of teeth structure too. The cognitive capacities are much harder  to 

document  except  by drawing inferences from tangential evidence about brain size and shape, tool artifacts, cave-wall drawings, and so on, up until the emergence of the forerunners of writing systems, which were probably stones of various sizes and shapes used in bartering.

111. How much data about the organization of language could the genetic material transmit? Modest data  would  suffice  for  a  proto-language (Fig. III.30a) with  a sparse repertory of sounds plus gestures, each of them being hard-coupled to one specific concrete meaning. A full-fledged natural language  has  an  intermediary °lexicogrammar that softens the coupling and enables multiple coding° for expressing meanings that range between specific and general and between concrete and abstract (Fig. III.30b). Here, two closely related questions arise. First, when and how did the lexicogrammar evolve: at what rate and in what order of intermediary stages? Second, why do we find no proto-language being used among any community in the world: why do all known languages have a fairly elaborate lexicogrammar, irrespective of how ‘advanced’ the culture might appear by other criteria? We might hypothesize that (a) language is unlikely to emerge prior to a certain ‘complication threshold’; or (b) human genetics have evolved to the point of supporting only full-fledged natural languages; or (c) language undergoes a steep amplification relatively soon after emergence, a bit like the ‘inflation’ of the universe soon after the big bang.

112. For evolution in the community, we can inquire how an established natural language evolves and under what adaptive pressures from the evolution of shared models of world and society. The discovery of new modes of producing foods or tools would obviously affect the lexicon, and we have plenty of evidence. Less obvious but still reconstructable would be some social changes affecting the grammar in specific domains such as forms of address or commands and requests (cf. IV.38, 60). But we are hard-pressed to account for the sweeping evolution of the phonemic and morphemic systems, which can profoundly transform the total appearance of a language, e.g., from Latin to French or from Anglo-Saxon to Modern English. Wouldn’t there be substantial adaptive value in keeping these systems stable so as to prevent the loss of information or the breakup of one language into several? Or would people just not perceive the evolution and imagine that these systems are stable?

113. The problem of stability has long puzzled linguists. In the early days, Saussure swiftly moved to consolidate the °formalist mainstream in a ‘static synchronic’ linguistics°, whose object of study was ‘the true and only reality to the community of speakers’, and to consign language change to ‘evolutionary diachronic linguistics’ as the ‘study of relations that bind together successive terms not perceived by the collective mind’. He irritably associated changes with ‘deteriorations’, ‘vicissitudes’, ‘damage’, ‘disturbance’, ‘breaking’, and ‘effacement’; and conjured a dark vision of ‘diachronic facts’ ‘setting a blind force against the organization of a system of signs’. These moves rendered °language by itself (‘langue’) as a uniform, stable, and abstract system° particularly austere, remote, and cut off from time and change, and obscured many rich constraints that apply to a language precisely when it is undergoing selective evolution, e.g., when the English Verb system is moving from °transitive toward ergative° (cf. IV.36, 82, 203). Here as elsewhere, formalism erases the memory of culture in order to describe forms without inquiring how and why they developed.

114. In a °functionalist view of language as a dynamic system°, in contrast, evolution is inevitable. Language is a °non-deterministic system with fuzzy contours and boundaries°, so that concurrent alternatives can persist and move in or out of the ‘standard’; the same dynamics hold for social groups of speakers with alternative language varieties. Also, °language co-evolves with shared models of world and society°, which are adapted as experiences are proliferated and people are regrouped into new constellations of roles. We could view this parallel ‘co-evolution’ as three versions of the °dialectic whereby the virtual system; controls by specifying constraints, while the actual system controls by manifesting constraints° (cf. Fig. III.6 in III.20), such as: in language, the unmarked word order of English Clauses versus the specific phrasing of a real utterance; in a world-model, the normal causality of events versus the connection between two changes in the current visual display; in society, the standard roles of teacher and pupil versus the ongoing interaction in an ecologically oriented classroom. Within each dialectic, the actual side °interfaces the standing constraints of the virtual side with the emergent constraints of its own context°. And constraints that frequently emerge can become standing ones when people assign them some °adaptive value°, e.g., when new patterns are adopted to indicate solidarity (cf. III.136).

115. The formality of language foregrounded by mainstream linguistics is the most °arbitrary aspect of language in being the least motivated by models of world or society° — and not, as Saussure suggested by making ‘grammar’ the opposite pole to the ‘lexicon’, the aspect that does the most to ‘limit arbitrariness’ (after he had naively remarked that meanings do not compel specific sound patterns) (II.48). Moreover, the formality of a language evolves through a long series of °frozen accidents° whereby the sounds and shapes of words undergo fortuitous changes that help set the conditions for the subsequent history of the language (cf. II.45ff, 83; III.53, 61, 87). Only when we take seriously the question of how the forms relate to their functions, and examine the organization of the unified lexicogrammar can we recognize the rich limits on arbitrariness and the rich capacity of language to adaptively exploit its frozen accidents. The functions (the ends) ‘control’ the language more richly than do the forms (the means) and can thus adapt the forms to suit their evolving needs and the needs of speakers in the culture. Speakers will in turn consider their own ends and select the adaptive means, even if these are not exactly the same means selected in the past or are not formally consistent with other means next to which they might be grouped in a linguist’s description. It is not that ‘language continues to function in spite of evolution’, as Saussure surmised, but that language evolves in order that it can continue to function.

116. The greatest formal stability is concentrated in the frozen islands, i.e., in convergent patterns of stabilized standing constraints, which can emerge in complex systems of all kinds. Evolution can ‘freeze’ the islands and can also ‘relax’ them again, e.g., by making them into merely °unmarked choices°. Some added stability accumulates when the language has a widespread °written standard° maintained by institutions such as schools and publishing houses. But the apparent stability of the written standard is misleading about the state of the whole language and mystifies the competition or conflict among alternative varieties and among the cultures that use them. The °adaptive value° of the standard for power through social, educational, or professional ‘mobility’ competes with the adaptive value of non-standard varieties for solidarity within their respective cultures (III.121, 171; VII.128; VIII.96, 150). The formalist campaign of °language guardians° for stability and ‘purity’ (e.g. for ‘correct grammar’) is a gesture for cultural or institutional power, and the changes or fluctuations it seeks to hold back and proscribe are just the ones occurring in the varieties spoken by °disempowered cultures or subcultures° (cf. VII.61, 169). The campaign is therefore a °mystified denial of linguistic human rights° (cf. VI.61; VII.344ff).

117. From another perspective, we could inquire under what conditions a distribution of functions might or might not evolve in ways that favor a stable and consistent distribution of forms. Instead of taking the formality of language as a certified and free-standing ‘synchronic fact’, we can ask where it came from, how it survives during evolution, and what adaptive values it might have and for whom. We could then probe the °emergence of the lexicogrammar in between the sounds and the meanings of a proto-language° (III.111). It is utterly improbable that this emergence was controlled, let alone planned; how could speakers set about developing a lexicogrammar if they didn’t yet know what it can do? So it must have °self-organized° to evolve the form-function correlations it eventually attained in its full-fledged ‘natural’ state. It emerged in the spaces, so favorable for self-organization, in between stasis, i.e., frozen one-to-one bonds of form to function, and chaos, i.e., unpredictable bonds of form to function; °increasing returns amplified minor and embryonic formal patterns into major and elaborated ones°. We cannot predict which sorts of forms will emerge, but we can predict that a pattern which begins to amplify will continue and will act as an ‘attractor’ for others. For example, it is probable that some word-class for objects would evolve the forms to distinguish what is done to things, where they are, who has them, and so on; and that another word-class for actions would evolve the forms to distinguish who performs them and when. But we cannot predict that the language will get specific ‘declensions’ of Nouns and ‘conjugations’ of Verbs looking like the much-studied ‘inflective’ models of Latin or Greek, with particular sets of ‘cases’ and ‘tenses’. Nor can we predict how long such systems will remain stable and when they will begin to flatten their formal differentiations, though we can see that some languages like English or French have done so more readily than others like Lithuanian or Finnish (cf. IV.214). The fact that the forms got elaborated shows that functionality can make good use of them, whereas the fact that the forms got flattened shows that functionality does not urgently require them. Viewed as form alone, a language in its evolution seems indifferent to creation and destruction, as accident wills (Saussure’s dark notion, III.113); viewed as function controlling form, language in its evolution is seen to keep pursuing human ends while varying its means.

118. What predictions might we make about the evolution in the design of a language? We might predict that the rates and types of change would vary among the subsystems of a language to reflect how richly a given subsystem °adapts to the constraints of world and society°. In the lexicogrammar, we can distinguish between the lexical side having a more rapid evolution, a higher ratio of deliberate innovation, and a less systematic emergence of word-classes than does the grammatical side (cf. IV.13). The evolutionary change in the lexicon (or ‘vocabulary’ in the usual sense) is propelled and constrained by such factors as the rise of technology and the resulting proliferation of new commodities and professions (V.74). Less leeway for change would be predicted in the grammar, which is more prone to engender frozen islands and is more indirectly constrained, e.g., when evolving modes of social interaction lead to new or more specialized functions of the Modal Verbs (cf. IV.38, 60). The least change would be predicted in the sound system of phonemes, which is scarcely expected to reflect world and society at all.

119. Yet sound systems do change, sometimes more radically than grammar; cognitive and social forces may not propel sound change but neither would they have compelling grounds or means to repress it. And change is always feasible because every sound system is °non-deterministic° in several modes. In fine-grained detail, no two sound events are ever exactly the same, even for the same speaker uttering the same words (cf. III.234; V.16.5). In coarser-grained detail, sounds adapt to their environments, e.g., whether a vowel like /o:/ (long ‘o’) has a briefer duration before a stop like /k/ (e.g. ‘stoke’) than before a nasal like /n/ (e.g. ‘stone’). Moreover, articulatory criteria do not function the same as auditory criteria, since sounds pronounced rather differently may sound rather similar (e.g. /l/ and /r/) or even get interchanged (e.g. when native speakers of German replace the /v/ in their English with a /w/). And finally, wide variations in pitch, loudness, and quality of voice are tolerated without disruption, as are many sound differences among concurrent dialects.

120. These various modes of non-determinism suggest that each sound-unit with its own place in the system — each phoneme in phonology — subsists as a physiological target around which the actual realizations in speech are fuzzily clustered (III.92). If the phoneme is analogous to a genotype (a set of instructions), the realization is like a phenotype (the result of carrying out the instructions) (III.60, 66); this analogy suggests that some random mutations can appear and get ‘selected’ to survive in the sound system. As long as the uttered sounds can be fitted to their targets with the aid of rich constraints from words and co-texts, minor variations and changes will not be disruptive. Yet within the same leeway, minor changes can persist and gradually amplify through increasing returns (III.117). Whereas minor transitions may be barely noticeable at one point in time, the total change over time can be dramatic, e.g., when the old Indo-European unvoiced stops written as ‘p’, ‘t’, and ‘k’ got shifted to the corresponding fricatives written as ‘f’, ‘s’, and ‘ch’, as sketched in II.17. Such a change could hardly have been predicted and, as we see from unaffected languages like Sanskrit and Latin, was plainly not enforced by world and society. But once the change was well under way, we can predict that it would spread regularly through the whole language because world and society have no deliberate mandate to stop it.

121. For grammar, changes ought to be more noticeable and hence easier to repress, but also more strongly propelled to adapt to changes in world and society than are sound systems. The campaign of generations of °language guardians° — academics, grammarians, schoolteachers, or parents — has failed in combatting the changes, ushered in by widening literacy and general education, that flattened the finicky distinctions in English cherished by the aristocracy and clergy, e.g. in forms of address, Pro-Nouns, and Modal Verbs; still, this very failure has its own adaptive value for the elites in making the more stabilized standard variety into the prime medium for °discourses of power° (cf. III.116; VII.61). More recent resistance has also proven futile against the attraction among English word-classes, notably the technology-driven shifting of Nouns into Verbs (e.g. ‘input’, ‘network’, ‘conference’); the primary users have more real power than do the language guardians.

122. The widest leeway within grammar comes from the fluctuating ratio between the °standing constraints, the stablest of which are the ‘rules’ applying to frozen islands, versus the emergent constraints generated on-line in appropriate contexts°. In grammatical patterns where speakers may have no firm sense of what ‘rules’ to apply, changes would meet little resistance. And even distinct ‘rules’ do get changed, e.g., the gradual elimination, in spoken English, of the contrasting forms ‘who’ and ‘whom’. Or, a change in the ‘core’ grammar may be prepared along the ‘margins’. For instance, a change may be introduced from another variety or dialect, as when the Third Person ‘‑s’ ending was introduced from Northumbrian and spread south to displace the sound written ‘‑th’ in the southern dialect of Middle English that became the basis for Modern English.

123. Or, we might face about and predict that language change, once it is under way, would be most thorough when constraints from world and society are the least direct, and vice-versa. So we should and do find sound changes sweeping all across the language rather than affecting just some words here and there. In grammar, in contrast, a change is more prone to leave an unchanged residue, e.g., when the innovative ‘‑s’ plurals in English failed to squeeze out all the old vowel-changing plurals like ‘goose/geese’ or the invariant plurals like ‘swine/swine’ although anyone could understand ‘gooses’ or ‘swines’. Some constraints may come from the affected domains of vocabulary, e.g., the old plurals surviving in animal husbandry, a profession unconcerned with grammatical niceties. But we still find isolated hold-outs (like ‘foot/feet’ among body parts), probably protected by parents and teachers and accepted by children as a cheap token to show respect for ‘grammar’ and ‘manners’. Also protected are some plurals of borrowed foreign words, e.g., ‘corpus’ keeping the Latin plural ‘corpora’ alongside ‘corpuses’; these are typically technical terms whose usage is stabilized by specialists, such as the schema theorists who keep the Greek plural ‘schemata’ as a signal of status, though they don’t know Greek and wouldn’t use analogous plurals like ‘commata’ for ‘comma’.

124. The least thorough and consistent changes should and do appear in the lexicon. Even moderately systematic innovations have been uncommon, such as introducing alongside some common English animal names some French-based names to designate the edible meat, e.g., for ‘cow/beef’, ‘calf/veal’, ‘pig/pork’, or ‘sheep/mutton’ but not for ‘lamb’ (and ‘chicken/poultry’ doesn’t match up) (cf. VI.8; VII.152). Much commoner are gaps and skewings of vocabulary, e.g., that we can run ‘slowly’ but not ‘fastly’, or disguise things ‘thinly’ but not ‘fatly’, though again anybody could understand what is meant.

125. Furthermore, lexical changes and innovations are the hardest to oppose or repress. Even the language guardians who strive to preserve a battery of finicky grammatical ‘rules’ must realize the hopelessness of regulating the lexicon — a task whose scope is overwhelming and whose principles are far from clear anyway. The appearance, say, of ‘maternity suit’ as an item of clothing and ‘paternity suit’ as a legal proceeding was nowhere denounced as ‘illogical’ or ‘incorrect’; and doing so would be pointless because the innovations correspond to two different constraints, one biological and one socio-economic. Ominously, lexical innovations may gain currency even when they deserve to be soundly repressed, e.g., the gruesome term ‘ethnic cleansing’ for mass killing and rape of non-Serbs by the Serbian army in the early 1990s — filthy deeds made to sound clean.

126. Still, if we view the lexicogrammar as an integrated whole subjected to the rich sets of constraints that are discoverable in a large corpus of actual texts, such as those sketched for ‘warrant’ in II.68-76, we may recognize some sources of stability that resist certain types of changes or adaptations. For example, the semantics and pragmatics of ‘warrant’ make it hard to adapt for trivial occasions, e.g., ‘my eggs warrant salt and pepper’. In contrast, we find motivated innovations like ‘warrant inclusion in a wheelchair’ in [26] in II.74, adapting to social constraints against harsher terms like ‘confinement’.

127. Large-corpus research is likely to affect models of language evolution simply by offering such detailed views of it. A ‘monitor corpus’ that keeps adding to its coverage naturally turns up new expressions whose motivation is not hard to see. Apart from the inevitable neologisms for new technological inventions or discoveries, e.g., ‘glassphalt’ in building materials and ‘riflips’ in genetics, we see reflections of social pressures, e.g., ‘glass ceiling’ for a company with invisible barriers to promotion, especially for women; or ‘right to life’ for the right-wing genderist denial of the woman’s right to terminate a pregnancy; or ‘pigmentocracy’ for South Africa’s obsessively nuanced racism. But extensive linguistic, cognitive, and social research will be needed on the deeper issue of how far the general direction of innovations or changes, whether lexical or grammatical, are actually steered by the constraints which a corpus indicates but which are not readily open to the intuition or introspection of speakers (cf. II.65f, 76).

128. We can now turn to the evolution of language during the life of the individual. Notions of how children learn a language have long been shaped by notions of how they learn in general, as described in III.90. For °realism and empiricism°, infants begin with no language and acquire it by imitating what they hear and matching this with their own experiences. For °idealism and rationalism°, infants begin with innate categories of language for organizing the language examples they hear. In linguistics, this split has dominated the controversy between the °behaviorists° who claimed that language is ‘learned’ as a part of learning how to behave, versus the °mentalists° who claimed that language is ‘acquired’ by specifying the constraints of an ‘innate universal grammar’. Insofar as both sides focused heavily on language as an already established system — the behaviorists on phonology and the mentalists on syntax — their accounts were sketchy about the explicit operations of learning and building up such a system. The behaviorists had to argue that children just imitate the utterances they hear; but mere imitation will not enable the system to emerge. The mentalists had to argue that language is acquired by itself, independently of situational contexts, thanks to an innate system like a theory prior to practice (VII.132ff). But instead of mapping out how such a system can get from theory to practice, the mentalists filled in the gap by proceeding as if the whole system were acquired at once — the ‘idealized “instantaneous” model’ where ‘successful language acquisition’ happens in one ‘moment’ (VII.134, 327).

129. A °systemic functional account of language learning° would highlight the °interaction of language with world and society° and would acknowledge evolution as a primary organizing force. Like all life-systems, human children have a built-in organization riding on chemical, biological, or neurological codes that can be compared to languages (III.58, 81, 106). But unlike other life-systems, they have a transmitted genetic capacity to learn °natural languages°. Beyond the physiology for articulation, e.g. vocal tract and teeth (III.110), we can’t yet tell exactly what the genetic material specifies and what has to be done with it or added to it. It can evidently operate with an enormous range of sounds, forms, and patterns, such as are found in the world’s languages; even if we could invent some that occur in no known language, we couldn’t prove they are ruled out by genetics alone. Still, genetics offers a solid basis to reassess the notion of °language universals°, which may have originally been put forward to paper over the question of how the complex and abstract system envisioned by formalism could be ‘acquired’ (cf. VII.133).

130. Recent empirical studies of the emergence of language in infants are relevant here (cf. VII.H). Intriguingly, the evidence indicates that infants do start off from a °proto-language°, whose design was shown in Fig. III.30a in III.111. Each one of a small set of spontaneously invented sounds (plus gestures) not derived from actual words of the ‘mother language’ is hard-coupled to a single specific concrete meaning. The act of uttering only serves for crudely monitoring and managing typical situations to signal distress, get fed, etc. This provisional system gradually evolves an intermediary ‘grammar’, again spontaneously designed rather than copied from adults, for elaborating richer functions and for transferring them from single sound-units to sound-patterns. But once the infant sets about shifting from spontaneously designed sounds over to real words, the grammar begins assimilating to the mother language fairly soon — another indication that the lexicogrammar has a functional unity (II.32; III.237; IV.13; VII.145). The grammar increasingly softens the coupling between utterances and meanings and enables the child to cover a wider range of meanings from specific to general and from concrete to abstract (cf. Fig. III.30b), thereby combining more means and pursuing more ends. As Halliday remarks, the grammar supplies ‘the potential for combining different kinds of meaning in one utterance — using language to think with and to act with at the same time’ (cf. III.203).

131. During the child’s evolution, adaptive pressure from world and society actively impels first the emergence of a spontaneous proto-language and then the assimilation to the mother language. The early sound system is arbitrary in respect to the mother language but is motivated by exploiting the spontaneous sounds infants produce when they are excited, happy, hungry, and so on. The early ‘vocabulary’ is directly motivated and constrained by the infant’s needs to understand the world and to interact with other people. The eventual ‘grammar’ adapts to the pressure for diversifying and specifying the child’s resources while making them more uniform with the language community, as if ‘growing up’ for the individual were like ‘modernizing’ for a society (III.3; VII.22).

132. The assimilation to the mother-language occurs both under genetic pressure and under cognitive and social pressure to enrich control over interactions with the family and with outsiders who wouldn’t be familiar with the proto-language. These joint pressures massively support the ‘cheap amplification’ of communicative capacities, whence the famous talent of children for learning non-native languages (VII.324). For adults, non-native language learning is much harder once the pressures have subsided and their language capacities now interact with a far more elaborated model of world and society, which is costly to modify (VII.87). Still, some degree of native language learning could continue throughout life as people enhance their sensitivity for such collocations and co-texts as we can recognize in large corpus data (II.67ff; IV.27). Or, specialized training may be administered, e.g., to make Margaret Thatcher sound more dignified and intelligent (a tall order); but aside from pronunciation and a few issues of usage, specialists still lack detailed models of how to enhance language competence.

133. We can now ask again why we don’t find proto-languages being used by a human community anywhere in the world (III.111). Using one would freeze the community into a permanently ‘infantile’ mentality; and its chances of survival would be low, because speakers would lack crucial adaptive resources for organizing world and society. As the genetic coding upon which language is based gets changed through mutation, genetic programs that restricted humans to proto-languages would get selected out. Further enhancement of the design of language through the evolution of the genetic base might occur, but only quite slowly.

134. Finally, we can turn to the evolution of language during discourse. As we saw in Ch. I, the project of °mainstream linguistics to describe language as a uniform, stable, and abstract system implied a frozen design°, which might look like a complicated version of the  network  in  Fig.  III.31a.  The view of language as a continually evolving system, in contrast, would look like a a complicated version of Fig. III.31b. an evolving network whose connections are temporarily activated and

,regulated among items within the °currently active version of the system°. So far, the mainstream project has failed to achieve coverage, convergence, and consensus. However, the prospect of describing language as a system continually evolving from one currently active version to another so as to support the ongoing discourse is still unfamiliar and may be rather disturbing to many linguists. The language might keep on changing while we are trying to describe it; and our descriptions might be unduly specific to just one version.

135. Undoubtedly, we will need to modify our long-standing conceptions of theory and method. If we cannot freeze language, we need to build models of how it evolves. If we cannot describe all the speakers’ versions of the language, we can describe the principles whereby such versions self-organize and evolve. If we cannot state a full set of formal rules for all sentences of a language, we can state a representative set of constraints that apply to the collocations and co-texts in a large corpus of real discourse (cf. II.64-79).

 136. If, like other complex systems, language runs on a °dialectic between the virtual control that specifies constraints and the actual control that manifests constraints° (cf. III.20, 66, 71, 114), the general model shown as Fig. III.6 back in III.20 would be specified as in Fig. III.32: ‘theory-driven’ is now ‘language-driven’, and ‘data-driven’  is  ‘discourse-driven’. But we may need a more complicated model, as suggested in Fig. III.33. There, the discourse is controlled by the °standing

 language constraints and by the emergent language constraints specified in the context°, as well as by the relevant °cognitive and social constraints from models of world and society°. Following the flow lines in Fig. III.33, the discourse exerts control back on the language by adding more co-texts and adjusting the sensitivities of speakers. A direct alteration of the language by a single  discourse  would be rare; even when a new lexical item is coined by one speaker, it can enter the language only if it has a motivation in the world (e.g. a new invention) or in society (e.g. fashionable prestige). This more complex dialectic rides on a °subcritical baseline of processing° (in the sense of III.80) continually ready to evolve a °current partial version of the language° (of the virtual system  ) to support the text (the actual system  ). This baseline would supply a rich and adaptive ‘pre-organization’ to assist the ‘self-organization’ of the current version of the language, while knowledge of world and society would be enlisted to maintain the °text-world model°, the total array of knowledge activated while processing a text (III.229). This model is not just the sum of the word-meanings as they might be listed in a dictionary; it would be a °richly interconnected data field° shading off into the person’s store of knowledge and experience. Instead of saying that words ‘refer to’ or ‘point to’ things in the world, we could say that the text-world model corresponds to some relevant activated domains in the model of world and society (cf. II.91).

137. For linguistics, the top questions might now be: how much coverage of the language does a speaker need, and how similar or consistent do the various versions of the language need to be to support a convergence among the data during discourse processing and a consensus among speakers (III.256)? Also, how could the ongoing evolution of language versions during discourse change the language system in significant and long-range ways? Such changes can obviously happen to the individual speaker’s knowledge of the language through ‘knowledge-driven’ modifications of the lexicon, particularly through °specialization°, e.g., entering a technical profession and joining in its discourse (V.76-79; VII.125). Less often, engaging in discourse with speakers of another variety or dialect can bring about lasting changes in a speaker’s grammar or pronunciation. But how could on-line discourse practices change the language of a whole community? And if discourse practices are not the agent of the huge changes we do see, e.g., from Anglo-Saxon down to Modern English, what is the agent? How could a language change when people aren’t using it? We might recall philologists like the Neo-Grammarians attributing change to ‘natural laws’ quite apart from discourse (II.17).

138. We are back again to the question of historical evolution in the community (cf. III.112ff). The shared knowledge of the language would also be the main source of the constraints shared from version to version, unless deliberate changes are performed, e.g., by a modernist poet (cf. III.140; V.16.5). Presumably, a large historical change can hardly be achieved deliberately; it would come from a series of minor changes, each of them going unnoticed or being judged insignificant. Successive generations of children might adopt and extend the changes made by their parents and grandparents and never become aware of the unchanged stage.

139. A community can also share constraints applying not to the whole language but to such °intermediary control systems as style (e.g. colloquial), text type (e.g. interview), and discourse domain (e.g. public politics)° (Ch. V). These constraints would be more likely to differ from version to version, and at least some constraints would be shared only by specific sub-communities, who would be the dominant contributors of the corresponding corpus data, e.g., the specific uses of ‘warrant’ in legal discourse as a term for official documents. And speakers would be more conscious of the constraints, especially during transitions, e.g., when learning to talk like a lawyer and use terms like ‘contemporaneous oral agreement’ in [607] in V.89.

140. An ‘individual style’ is often attributed to just one person, usually to a literary author. But defining just what constitutes a style remains a stubborn problem, as we shall see in section V.A. Some individuals like Gerald Manley Hopkins work hard to change from the language they use in ordinary discourse, whereas others like Ernest Hemingway work hard not to. But how different is the current version of their language when they are producing a text in their style, and how do we tune our own version when we are receiving it? The widest variation is not so much in the resources of language (e.g., in the whole lexicon) as in the patterns of collocation. As we saw for ‘warrant’ in II.68-76, the constraints on collocation can be quite delicate, and minor innovations might be detected, although perhaps not consciously. Whether these would be accepted as signals of an individual style would depend on whether they act as part of a personalized pattern and on whether the participants judge them to be signals for innovative meanings rather than mere solecisms due to carelessness or to inadequate control of the language. A successful style attracts imitators, but they are hampered by the widespread uncertainty about what a style consists of and how it should be implemented.

141. Some eminent authors such as John Milton have been subjected to a statistical analysis of the corpus of their works. But the results don’t tell us which proportions reflect the general usage of Milton’s time versus his own evolving self-conscious usage. We would also need a very large corpus of the general usage, such as the ‘Bank of English’ mentioned in II.64f for current English. The requirement would be feasible at all only if we relied on written records and tried to reconstruct the spoken usage of past eras, a method which entails substantial risk. 

142. As we can see, the various modes of language evolution and change raise numerous difficult problems that still need to be accounted for by a theory or model. The same adaptability and range that make language so efficient and effective also make it hard to sort out what aspects might be more or less stable or evolvable, and why. Against the formalists’ ‘synchronic’ vision of a static language frozen outside of time, there arises the functionalists’ ‘socio-semiotic’ vision of a dynamic language in the thick of social and historical practices. In the second vision, language and meaning are eminently °non-classical phenomena° whose connectedness, temporality, locality, observability, and so on, are exquisitely sensitive to contexts. Discourse constrains and is constrained by both the °interobjectivity of shared experience and the intersubjectivity of shared meaning°, without being anchored at any one time or place (cf. III.84, 94). A language is not tangibly secured in its identity, connectedness, temporality, locality, substantiality, observability, measurability, or predictability; yet it can effortlessly acquire any or all these factors when discourse takes place. Also, language exploits concurrent alternatives and local interactions to bring about the transitory but potent critical mass of context (III.83, 231, 242; IV.8). And it does not support total certainty or determinacy, but rather a convergence of data making some meanings or understandings much more probable than others — rather like the situation in the quantum world as described in III.12f. 

143. But if language and meaning are eminently non-classical, the keen skills of human data processing make them appear fairly classical to the speaker and hearer — mainly simple, determinate, familiar, and stable (cf. III.30, 151). Discourse can act as a ‘great classicalizer’ whereby you can apparently enhance the ‘realness’ of objects or events merely by talking about them, and sort out ‘truth’ or ‘fact’ with no involvement in constructing them, by just ‘telling it like it is’ or ‘saying what you see’ (cf. III.7). The ‘vocabulary’ of language might appear to project a °reified reality of tangible, free-standing entities°, each secured under its own name.

144. However, discourse can also act as a ‘great bracketer’ whereby you can disengage from reality without becoming disoriented. Discourse can support long-range planning of events that may or may not happen, or project countless alternatives to the current state, or run through imaginary contingencies and extravagant fantasies. The lexicogrammar of discourse offers form-function repertories (e.g. Modals, Conditionals) to step back from ‘truth’ or ‘fact’ and to signal nuanced beliefs and attitudes (Ch. IV). Hence, discourse can both °construct and deconstruct reality°, but this dual potential has been abridged by °common-sense realism and by mainstream linguistic and semantic research°.

145. In this section, a post-classical evolutionary perspective has been applied to the organization of language and discourse in its interactions with knowledge of world and society. Discarding the old view of language as a static system, we can probe the conditions for its evolution and evolvability in life-systems, both internal and external, both individual and social. This unconventional perspective may help us reassess some long-standing problems and questions, such as how language could originate and change, and how it participates and adapts in human genetics, culture, history, and maturation. We may better understand how language and meanings are sharable; why the use of language in discourse makes such modest demands on time and effort; why language learning is easy for children but hard for adults; and why proto-languages are used by infants but not by whole adult communities. We may gain an operational model for conventional but unspecified notions like ‘meaning’, ‘thoughts’, ‘ideas’, ‘awareness’, or ‘consciousness’. And finally, we may formulate a °transdisciplinary ecologist program° that leads to designing more equitable methods for using language and discourse in socialization and education.

 III.E. Communication among the sciences

146. Describing science as a communicative enterprise of °discoursal, cognitive, and social transactions° (III.198) might not be welcomed by °classical scientists°. As resolute °realists, they prefer to see themselves discovering objective facts rather than constructing subjective knowledge and producing discourse about knowledge° (V.76). Their notions of discourse are often naive, witness the °positivist or physicalist notion in unified science° assigning ‘the greatest importance’ to ‘speaking about physical things’ (Carnap) (III.10). They are not in the business of doing discourse analysis, cognitive science, or sociology of knowledge; and even had they decided to start, they would have gotten little help until recently from such sciences as a °behaviorist psychology centered on observable actions° or a °formalist linguistics centered on language by itself°. But the prospects are brighter for a °post-classical science of text and discourse centered on access to knowledge°, expressly including scientific knowledge. We can °bracket the content° of science and explore the communicative status of theories and models as °accredited schemas emerging from modeling styles and paradigms°, and can consider the opportunities for °dialectical interaction between such familiar dichotomies as theory versus practice, realism versus idealism, objectivity versus subjectivity, physicalism versus mentalism°, and so on (cf. III.191).

147. Could the philosophy of science be useful here? In the conventional exposition (outlined in II.2), a normal science shares a paradigm of assumptions about theories, methods, and terms, thus offering a stable scenario for gathering and handling accredited types of data. Eventually, each theory reaches stagnation by depleting its potential for getting fresh data and for solving interesting problems, whereas alternative data and problems look messy or chaotic. The final stage enters a crisis, which is definitively resolved only when a scientific revolution displaces the current paradigm and supplies a more productive scenario for the next stage of normal science.

148. This philosophical exposition has unintentionally led to some noisy and unproductive rounds of ‘science-bashing’ that simply feed the °anti-intellectualism oddly coexisting in modern society with devout scientism° (I.3). We need to highlight the inadequacies of the exposition itself in failing to balance the threefold status of science in terms of its cognitive, discoursal, and social moves. The decisive role of discourse among scientists or between scientists and the society at large in constructing and negotiating both reality and theories about reality is vastly underrated (V.76-79). The chief orientation has remained classical in assuming that our own modern knowledge of reality is so secure that we can use our hindsight to dismiss past paradigms, often carefully chosen ‘straw man’ cases from the natural sciences, e.g., the ‘phlogiston’ theory of combustion (cf. III.171f). Essentially, a theory is assumed to be driven by reality as scientists then saw it, i.e., by one united and shared domain of phenomena. From a post-classical standpoint, reality is never given for everybody, but is continually passing through a series of current versions that support ongoing activities such as discourse moves. These activities dialectically tune and are tuned by reality to suit their own goals. Classical science has tuned reality its own way by disconnecting its theories from the practices of ordinary experience in search of underlying constraints, such as the ‘laws of nature’ (III.11). These can be discovered and tested only through specialized practices with training and technology, which in turn forcefully shape the version of reality they only claim to reveal; indeed, they assert a reality you can access only by ignoring much of what you can actually observe, e.g., the roundness of the earth when it looks flat, or the indestructibility of matter when household objects like key rings or notebooks have obviously dematerialized. It remained for the new physics to make scientists genuinely conscious of our participation in shaping reality. We can now °bracket the content of theories and focus on how their design evolves during the ongoing dialectics between alternative theories, including both ones put forth by the sciences and ones arising from ordinary experience and commonsense knowledge of world and society°. Here, the °scientific revolution° we require is not one claiming to get us ‘back to reality’ following some now painfully obvious error, but one that °reconnects theory with practice in a dynamic dialectic° and strips away the premature and now counter-productive ‘realness’ of theories and terms that purported to state the definitive essence of reality.

149. Viewed from a post-classical standpoint, °normal science would be a communal modeling style °. Its major moves are °controlling and classicalizing by means of a paradigm of theories and practices°. Its °control centers rest upon the basic postulates of identity, connectedness, locality, observability and so on, and upon the corresponding moves of identifying, connecting, measuring dimensions in time, space, and substance, and observing and predicting° (cf. III.8). These postulates and moves are necessary to construct and test any scientific theory but they cannot themselves be tested by a normal science whose methods of °verifying and falsifying° always entail them. You cannot make observations and predictions that refute the legitimacy of observing and predicting in principle, because you would only refute yourself. At most, you could deploy such moves to show the limits of such postulates, e.g., for the ‘conjugate variables’ whose precision must be traded off in the ‘uncertainty principle’ (III.14), or for the °thresholds of critical mass and critical dispersion where identity, connectedness, and so on become unpredictable because major emergent properties are gained or lost°. Our next step would be to design explicit models of these moves to explore human moves like observing or predicting favor certain types of results over others.

150. Normal science makes routine progress when research completes and specifies the current store of knowledge by adding on facts like so many building-blocks and by making them more exact. Each fact is to be demonstrated by observing objects and events and by assigning temporal and local connections, chiefly cause and effect. Data that can readily be gathered and handled this way are accredited for investigation and for inclusion within the established categories. Other data are neglected or manipulated into insignificance by declaring them to be either (a) ‘anomalies ’ occurring ‘rarely’ or ‘randomly’; (b) ‘errors ’ caused by ‘noisy’ or ‘contaminated’ conditions; or (c) ‘distortions ’ by scientists who hold ‘unscientific biases’ or ‘wrong theories’ (cf. III.23f). These moves legitimize normal science to sustain its ‘objectivity’ even while refusing to accredit a margin of the data it has actually observed (III.148).

151. The total project of °classical science is thus to construct and reinforce its own version of classical reality by discovering new facts° and explaining away anomalies, errors, and distortions. Ultimately, science should prove the universe to be stable and determinate, and provide the theories to render its events familiar and simple enough to be fully understood and accurately predicted (cf. III.30, 143). The °goodness of fit° between the universe ruled by pervasive order versus the model science constructs of it should steadily improve as progress continues; and °coverage, convergence, and consensus° should steadily rise (cf. II.28; III.6). Only the rate of progress would fluctuate, with phases of rapid discovery punctuated by phases of the stagnation that sets in when the neat preserve of accredited data has been worked over. In terms of the design parameters proposed back in III.42ff, °stagnation is resistive stability°, where no new and important facts are discovered and °diminishing returns settle the slope of novel events down into equilibrium°. No provisions are made for the converse phase of the °chaos in resistive fluctuation°, where predictability breaks down, anomalies are proliferated, and °increasing returns amplify small events into large drifts going far from equilibrium°. Here, the goodness of fit between the universe and science’s model of it starts to disintegrate, with some domains of the model making it higher and others making it lower. The classical project for attaining a complete goodness of fit then seems profoundly problematic.

152. In a post-classical account, science does not just reveal the order of the universe, but makes its swiftest progress by navigating at the °boundary between order and chaos and by discovering order in chaos° — patterns in chance, regularity in fortuity, richness and complexity in accident. In this way, post-classical science reflects the universe more comprehensively than classical science because we can now engage with regularities in phenomena that do not become predictable or settle down into equilibrium, e.g., weather, dripping faucets, coastlines, and sand dunes in ordinary experience (III.15) as well as some special reactions in chemistry, particle interactions in physics, or equations in mathematics. By acknowledging and exploiting the boundary between order and chaos, science can gain some cheap self-organization by discovering the principles for the emergence and evolution of highly complex adaptive systems such as human language, whose design resolutely resists being frozen for purposes of description.

153. A post-classical approach might also model the evolution of scientific theories using the design parameters proposed in III.42ff. Each prevailing paradigm projects a °trade-off between supportive versus resistive°  (Fig. III.34). An °accredited theory  in a normal science° is on

the °supportive side and appears orderly°: its °stability makes it reliable, and its determinacy  makes it perspicuous°. A non-accredited theory, in contrast, is on the °resistive side and appears chaotic°: now its °novelty makes   it   seem volatile,   and   its indeterminacy makes it vague°. When °anomalous data reach the proportions of a crisis and the stage is set for a scientific revolution°, these values are subtly inverted: resistive becomes supportive, and vice-versa. The new theory is now creative, and its indeterminacy is open°. The °stability of the old theory is now called stagnant, and its determinacy called biased°. So the new theory is judged progressive and is accredited, while the old theory is judged regressive and is discredited. Here also, order has emerged at the boundary of chaos: a new theory once deemed chaotic is now taken to be a new perspective on order. In this account, a °scientific crisis would be a state where the accredited theory has propelled the science far from equilibrium°; such a state is °subcritical, and a new theory that can restore equilibrium has a good chance of reaching the critical mass for a scientific revolution°. Thereafter, the °renormalized science° restabilizes in a new ‘order’ and shies away from data that look ‘chaotic’ — maybe just the data that will later lend impetus to a still newer theory.

154. Similar modeling might be applied to the career of the °normal scientist°. Classical training by teachers and textbooks foregrounds order and makes the current paradigm seem timelessly true and unshakeable. But in the later competition to enter and advance in the science, aspirants may be preferred with a new, seemingly ‘chaotic’ theory that does not come from their training. In the event, most revolutions are carried out by younger scientists with scant commitments to the accredited theory, who can notice a chance for ‘progress’ toward a new ‘order’ where others only see total ‘chaos’. So the ‘revolutionary’ theorizing will for a time be confined to a ‘radical’ minority whose efforts are widely recognized to be ‘progress’ only once their theory is shown to exert more °supportive control for order° than the accredited theory does. Then the new theory is revalued and brought into the mainstream of teaching and textbooks while the more cautious scientists accept it (or at least stop opposing it in public). Only rarely do distinguished ‘permanent revolutionaries’ stay at the forefront of progress, where they both exploit and produce new technologies such as supercomputers, linear accelerators, satellite radar telescopes, and nanotechnology workstations, plus the mountains of new data these generate. In our own times, we might point to Richard Feynman, Murray Gell-Mann, John Wheeler, David Bohm, and Stephen Hawking in physics; Ilya Prigogine in chemistry; Francis Crick, James Watson, and Stuart Kauffmann in biology; Allan Sandage, Margaret Geller, and Boris Zeldovich in astronomy; John von Neuman, John McCarthy, John Holland, Charles Bennett, and Douglas Lenat in computer science; Allen Newell, Herbert Simon, John R. Anderson, David Rumelhart, and Walter Kintsch in cognitive science; Kenneth Arrow and Brian Arthur in economics; Robert Longacre, Peter Hartmann, Harald Weinrich, Michael Halliday, John Sinclair, and Teun van Dijk in linguistics; or Michel Foucault, Stephen Toulmin, Jacques Derrida, Siegfried J. Schmidt, and Luce Irigaray in philosophy. Their family resemblances include an uncanny knack for moving beyond the borders of single issues or disciplines and for creating dynamic transdisciplinary spaces wherein, as for Alice and the Red Queen, it takes all the running you can do just to stay in one place.

155. No doubt the classical ambience of science strongly encourages successful ‘revolutionaries’ to subside into a fresh equilibrium and become ‘mainstreamers’ defending their theory against newer alternatives. Paradoxically, the cognitive ambience of classical science is progressive whereas the social ambience of ‘academic politics’ is conservative (cf. VIII.18). Classical scientists are content to live in the submerged contradiction between their historical hindsight that every theory so far has eventually been revised or discarded, versus their timeless ambition to prove that their latest theory is, finally, the ‘correct’ one. Their ideal is a story of progress going up to their own theory but not beyond it. And, once established, their own theory faces about and defends its privileges by highlighting constancy and stability over change and variety, e.g., when the revolutionary rhetoric of ‘generative’ linguistics subsided into a monumental orthodoxy of privileged technical constructions like ‘competence’ and ’language acquisition device’.

156. Most classical scientists prefer a °data-driven approach wherein theories and models are built on accredited ‘facts’ from the bottom up°, because this approach seems the safest from being overturned. A °theory-driven approach building from the top down° is viewed more cautiously, except in such abstract disciplines as mathematics and philosophy, whose capacity for being driven by reality is limited in principle (III.158). Even so, science is obliged to merge the two approaches within a °dialectical relation, which is the essential precondition for all cognition and communication° (III.20). Another paradox: theory by itself is ‘undercontrolled’ by data because several theories are possible for the same data and the data sample can never be complete; yet theory is also ‘overcontrolled’ because the occurrence of authentic data always entails some circumstances which are entirely accidental or which fall in the domain of another theory. So even the most resolutely classical science has to share control between theory and data, and to build theories that draw upon other theories, however implicitly. And these two recourses should be explicitly addressed in our own models of how science works.

157. The actual practices of science have long been seeking °parsimony in theory-building by exploiting theoretical constructs that have proven their merits elsewhere°. A standard ‘classical’ move has been a °global movement among the sciences between sparser versus richer °. Fig. III.35 displays the array of the °natural sciences°, which, in English usage, may get equated with ‘science’ itself; viewed horizontally, the display resembles the vertical layering shown in Fig. III.19 back in III.58 for the enrichment that led to the emergence of life and eventually of cognition and communication, except that we can now safely include mathematics, whose role in the emergence of life has not been settled. The left-hand sciences with sparser constraints  supply control  for  the right-hand  sciences with richer constraints. Here, accreditation is sought by showing that the construction of a richer theory or a model is strictly  derivable from some sparser one by incorporating precisely stated constraints.

158. Mathematics has been the sparsest source domain. It exploits the postulates of identity and dimensionality or measurability so assiduously that it depends only modestly on connectedness, temporality, locality, observability, and predictability, and not at all on substantiality. In terms of the dialectic described back in III.20 and Fig. III.6, mathematics specifies constraints but does not manifest them in the world. Its entities are objects only in an extremely sparse sense, e.g., the lines, planes, and solids of Euclidian geometry as contrasted with real objects like wires, table tops, and ice cubes; nor are its entities phenomena insofar as their properties can be defined quite independently of sensory manifestations (cf. III.70). So mathematics is best able to reconcile abstraction with precision and to state general or even universal constraints that are detached from experience yet fully accurate (cf. II.12), e.g.: ‘if a = b and b = c, then a = c’ (the ‘law of transitivity’). °Realism and mechanism° are not strictly entailed, but they can gain accreditation by using mathematics to quantify the dimensionality of objects, e.g. size or shape, and the temporality of cause-effect connections among events, e.g. frequency or velocity. This accreditation has encouraged classical science to seek control by ‘reducing’ qualitative to quantitative, and observation to measurement (III.179, 191).

159. Physics is the next source domain in line. It emphatically exploits the postulates of connectedness, temporality, locality, substantiality, observability, and predictability. So physics both specifies constraints and manifests them in the world, but still quite sparse ones, witness the fine-grained electrons and photons of particle physics as contrasted with coarse-grained ordinary objects like apples and oranges. Whereas mathematics is the closest approximation of a pure data substrate, physics incorporates a material substrate of matter and energy whose extremely hard coupling had, prior to the new physics, led scientists to consider it fully independent of the data substrate (III.17ff). So physics can state general or even universal constraints on objects and events and on causes and effects, e.g.: ‘energy cannot be created or destroyed’, or ‘the distribution of energy becomes steadily more uniform’ (the first and second laws of thermodynamics). Classical physics had forcefully enhanced °realism and mechanism° by extending them to aspects of reality that are quite inaccessible to ordinary experience (III.11, 148; VIII.46), such as the long-range motions of stars and planets (e.g., the orbit of Uranus) or the short-range motions of tiny particles (e.g., the Brownian movement caused by collisions of molecules in a gas or liquid). Physics deploys mathematics to impose precision upon abstraction — not just through ‘pure’ numbers and formulas but through inexorably ‘real’ though non-experiential’ processes. This tactic supports accreditation for theories that are expressible in mathematical formalisms, such as superstring theory (cf. III.46, 94).

160. Chemistry would be the third source domain, its sparser inorganic side engaging more with physics, and its richer organic side more with biology. It supportively enriches the central postulates in physics, especially substantiality, witness the polymers and proteins of organic chemistry being far closer to life-systems than are the electrons and photons of particle physics. So chemistry can state the properties of matter for distinctive layers of richness, e.g.: ‘all organic compounds contain carbon’, or ‘all polymers have repeating structures’. As we saw in III.59f, molecules that can catalyze others are capable of self-organization, possibly including the critical mass whereby life-systems emerged. An amplified instance would be the enzymes which provoke or accelerate reactions without being used up themselves, and, by controlling reactions, also control ‘signal response systems’ among the proteins specified by genes (cf. III.85, 176).

161. Biology would be the fourth source domain, its postulates once again vastly enriched, witness the genome in each living cell as contrasted with chemical proteins and enzymes. Regarding our basic postulates, a life-system is richly constrained: (a) in its temporality by its histories of evolution as a species and as a society, and of birth, maturation, reproduction, and death as an individual; (b) in its locality by sharply distinguishing inside from outside and by reconfiguring itself in parallel with its environment; (c) in its connectedness by interaction, co-evolution, and competition with others, e.g., in a food chain of predators and prey; (d) in its substantiality in having cell types with specialized functions, e.g., white corpuscles to attack intruders; (e) in its observability in being obtrusive and yet able to conceal itself, e.g., by taking on mimicry or protective coloring; and (f) in its predictability in forming habits, e.g., mating only at certain times of year.

162. So biology exploits the same postulates as the other natural sciences but in far richer ways; it both specifies constraints and manifests them in the world, including many °frozen accidents that created the basis for regularities in further evolution, some of which produced more adaptive life-systems° (cf. III.61). It can make general statements that are enrichments of physics and chemistry, e.g.: ‘all life-systems have a metabolism that dissipates energy’, but also more specific statements on its own terms, e.g.: ‘on the average, the larger the size of a plant or animal is, the more likely it is to have a long life span, a late onset of reproduction, and a modest number of offspring that are protected and cared for during their early stages’. The latter statement again shows how the material aspect of dimensionality is also a data aspect: the size of the organism is soft-coupled to a set of biological instructions about how it will probably live and behave. Also, the size encodes the organism’s evolving relation to its environment, e.g., when an increase in size enhances its ability to compete with other species and to survive in the face of adverse conditions such as extended dry seasons. And the genetic changes in those individuals (appearing as phenotype) that take the lead in this evolution of size will favor their survival and increase their frequency vis-a-vis other members of the whole species (constituted as genotype) (III.60, 63, 66). Eventually, though, the same instructions that helped to prolong the species can make it vulnerable to extinction, e.g., when it has reached a ceiling where its size or behavior is maladaptive and ecologically unsustainable for its environment, as may eventually happen to humans.

163. This overall movement among the sciences between sparser and richer is °classical insofar as one science exploits another by anchoring its own theories and models on their shared control postulates with the express intent to impose determinacy and stability° (cf. III.11). Notable instances include the linear differential equations in mathematics, whose rigorous ratios anchored the classical laws of motion and gravitation by Isaac Newton; the closed thermodynamic systems in physics, whose balanced regulation and equilibrium inspired the cybernetics of Norbert Wiener; and the animal conditioning in biology, whose mechanistic stimulus-response chaining grounded the behaviorist psychology of J.B. Watson and B.F. Skinner. Conversely, less classical issues were largely neglected until their revolutionary potential for deeper insights was eventually recognized through technological advances, e.g., the mathematics of °non-linear equations°, the physics of °chaotic systems° not seeking equilibrium, and the biology of °self-organization° in the evolution of species. These were all widely acknowledged only when high-powered computers made it feasible to do the required calculations and to test models by simulating them and watching the outcome.

164. Insofar as moving between sparser and richer sciences will affect design, we can expect some important °trade-offs among modes of control°. Since physics states the universal and inexorable constraints that apply to physical systems, it supports models that control many cases but in sparse detail. Since biology states richer and more variable constraints on life-systems, it supports models that control fewer cases but in richer detail. Mathematics states constraints so general and abstract that they must be enriched before they can control any real objects and events, while these must be made very sparse in order to be fitted to mathematics, e.g., when statistics constructs one idealized ‘average citizen’ for all the real people in a society.

165. A major challenge for science is to steadily enhance control despite all these fluctuations between richer and sparser constraints (cf. Fig. III.35 in III.157). If we push too radically toward sparseness, we may cross °critical dispersion and lose just those essential emergent properties° of the phenomenon we want to explain, e.g., if we try to account for the human being as a mix of chemicals (cf. III.51). If we push too radically toward richness, we may cross °critical mass and gain emergent properties° that transform the phenomenon, e.g., if we try to account for the evolution of an animal species as the result of its deliberate plan to take on adaptive features rather than as a gradual selection favoring those species or individuals that happen to acquire such features through unplanned mutation (cf. III.62, 77). To retain control across the sciences, we need explicit cognitive, discoursal, and social models of scientific communication, exploiting the °dialectics between data-driven versus theory-driven, realistic versus idealistic, classical versus non-classical, ordinary versus scientific reasoning°, to indicate how making the constraints richer or sparser affects control — understanding, insight, progress, accreditation, validity, and so on. Here, we might pursue analogies between a rich natural language like English versus the sparser biological languages in life-systems, such as genetic coding (III.66, 81); or versus the sparser chemical languages of proteins, enzymes, and so on (III.160); or versus the still sparser physical languages of messenger particles whereby the four basic forces of nature perform interactions (III.46, 95); or even versus the sparsest mathematical languages of symbols, formulas, and equations (III.54).

166. °Mainstream linguistics° has already envisioned some enrichments of the natural sciences, but, due the restrictive quest for °language by itself°, mainly in sketchy or programmatic analogies. For Hjelmslev, the ‘main task’ of ‘linguistics’ is to propound ‘an immanent algebra of language’ drawing on ‘meta-mathematics’ and a ‘logical theory of signs’. For Bloomfield, ‘human actions’, ‘including speech’, ‘are part of cause-and-effect sequences exactly like those we may observe’ ‘in the study of physics’. For Chomsky, a ‘chemical theory’ ‘serving as a theoretical basis for techniques of qualitative analysis and synthesis of specific compounds’ is analogous to a ‘grammar’ for ‘the investigation’ of the ‘analysis and synthesis of particular utterances’. For Firth, ‘linguistic behavior’ as a way of ‘maintaining appropriate patterns of life’ is a topic for ‘the study of the whole man by biologists’, ‘anatomists, physiologists’, and ‘neurologists’. But none of these analogies was worked out either in a detailed theory or in the practices for treating language data, such as the data-handling moves listed back in II.25. Instead, words and phrases were often treated like simple physical objects we can cut up or move about, and not like complex symbols that are richly interconnected to contexts.

167. Over the last 30 years, linguistics has programmatically exploited mathematics and appropriated the venerable tradition of formal logic , defined most broadly as any system whereby the value of a term (e.g. true or false) can be strictly determined through formal steps from the known value of other terms (cf. II.10). No doubt this trend was encouraged by the ‘language-like’ appearance of mathematical notations, which seem to offer more rigorous modes for representing language data; and by the long-standing aspiration to consider logic the prime model for all human thought and reasoning (II.13). Hence, numerous linguists have considered °modeling natural language (e.g. English) upon formal language° (e.g., first-order predicate calculus). This °classicalizing move salutes the formalist project of capturing meaning as form° (II.57ff; III.25); as a °modeling style, it seeks to impose determinacy upon complexity° by rendering the data formalized enough for precise computations. Once rich language data (even isolated sentences) are made into sparse symbol chains or formulas, we can compute which symbols occur together or in stable environments or form a ‘set’; which symbols represent ‘constants’ or ‘variables’; how one chain can be recoded as the ‘transformation’ of another; and so on. Eventually, we might state a set of formal ‘rules’ for the selection and combination of symbols into the ‘well-formed chains’.

168. The price for this rigorous design is the indeterminacy and fluctuation entailed in going from words and sentences, whose meanings are naturally constrained by the interaction of language with knowledge of world and society, over to the symbols and formulas, whose meanings need to be artificially constrained by the formalizer. Despite the investment of impressive talent and effort, the hypothesis that this daunting task can be done for any major corpus of language, let alone for an entire language, is so far wholly unsupported; and, if °coverage, convergence, and consensus° are the tests, it stands refuted along with all radical versions of linguistic formalism (cf. II.28, 41, 54; III.184).

169. It is vital to appreciate the reasons for this impasse. Analogizing with mathematics by treating natural language as if it were a formal language is a category mistake that short-circuits the notation directly onto the domain, much as when science takes its own procedures for designing stringent theories and formal representations to be the privileged model of the human processes and practices of thought and language (cf. II.13, 96; III.180; V.124; VII.63). The short-circuit skips over the evolution of life and language through the principles described by physics, chemistry, and biology, as if language could have somehow evolved straight out of mathematical notation instead of exactly the reverse. And just as the notation evolved out of language through being steeply rarefied to remove richness, we must steeply rarefy language data whenever we rewrite them into formal notation. Aside from a few fairly mathematical issues like ‘quantification’ by enumerators (e.g., ‘each’, ‘every’, ‘all’, cf. II.50), °critical dispersion° impends: the data lose many of the emergent properties whereby language enables discourse to continually activate or generate emergent constraints upon what is ‘meant’ (cf. IV.19). The only remedy would be to enrich the notation to the same degree as a natural language by assigning to each symbol the same meaning(s) and function(s) as a real word. The notation would then be just a natural language in disguise, much like English rewritten by replacing each letter in the alphabet with a number, like a row of 1s and Øs in computer languages. Obviously, the ‘number-words’ would not interact in a sentence the way the numbers would interact in a mathematical calculation such as adding or averaging. And the relation between sparse numbers and rich meanings would be just as °arbitrary° as the relation between the series of sounds or letters and the meaning of the word (cf. II.48). So instead of explaining the meaning in language, we would only have duplicated the form of language in a far less readable modality — a scant return for all the effort.

170. We should therefore expect intractable problems when formal systems or ‘grammars’ treat natural language by ‘rewriting’ sentences into symbol ‘chains’ or ‘strings’ and looking for ‘underlying regularities’ in the patterns of ‘noun phrases’ (‘NPs’) and ‘verb phrases’ (‘VPs’) rather than in the actual words of a sentence. Formality becomes a goal in its own right, instating the formalist credo that a theory can be replaced only by another that is at least as formal and preferably more so (II.57, 115; VII.188). The description remains uneven, concentrated on °frozen islands° in English grammar (cf. II.45ff, 83, 86), e.g., tidy patterns like ‘adjective + noun’ or ‘auxiliary + verb’, and shying away from °functional constraints° on the mutual positions of words or word-parts in authentic co-texts, e.g., the ordering of several ‘adjectives’ before the ‘noun’ (‘the old grey mare’ rather than ‘the grey old mare’) (cf. IV.142). Many functional constraints are primarily concerned with factors like meaning, belief, attitude, and focus, and we cannot hope to deduce them solely from their effects on formal position, which are secondary — °epiphenomena° in the sense of III.45.

171. Perhaps the notions of °grammaticality and well-formedness° have a status in linguistics rather like that of ‘phlogiston’, which chemistry postulated, up into the 18th century, to be the correct explanation for combustion. ‘Phlogiston’ was the hypothetical substance contained by all burnable materials and released when they actually burnt. The theory fit the observed fact that the product of burning usually weighs less than the original material. Chemists saw scant motives to inquire whether they had merely pasted a handy label on the observed phenomena via circular reasoning: certain things burn and lose weight because they contain phlogiston, and phlogiston must exist because certain things burn and lose weight. It remained for advances in technology to break out of the circle, namely the rise of pneumatic chemistry that could retain the gaseous products of reactions.

172. Formal linguistics seems to have followed a similarly circular argument. The technical construction of ‘grammar’ is legitimized by the evidence of ‘grammatical sentences’, while the ‘grammaticality of sentences’ is decided by the ‘grammar’. But since much of the evidence has to be carefully normalized (II.41), grammaticality is not discovered but imposed (II.41; III.195); so the ‘grammar’ is legitimized by evidence expressly fabricated for the purpose. From a °systemic functional standpoint°, in contrast, human utterances assume the shapes they do not because people are trying, in a barely ‘competent’ way, to uphold ‘grammaticality’ (IV.2), but because they are trying, in a highly competent way, to °interface the constraints of language, world, and society°. Converting functional constraints into formal ones and rewriting sentences into formulas heavily °overcomplicates the role of linear positions° (cf. II.84, 106). Devising and enriching the constraints upon the formal notation comes to compete with rather than support the project of describing the organization of natural language. Because the formulas look unfamiliar and arbitrary, they paradoxically seem more complex than the sentences but are actually much simpler, whereas because the sentences look familiar and ordinary they seem simpler but are actually much more complex. Due to this paradox between apparent versus real complexity, the search for richer constraints gets deflected by the demands of the notation for sparse constraints. We see here the °trade-off whereby the complexity and indeterminacy ‘squeezed out’ of the model (and the notation) just flow over into the connection between the model and its domain° — and into the procedures for using the notation (cf. III.40 and Fig. III.12 there).

173. To work with enrichments of mathematics, linguistics or a science of text and discourse should not short-circuit language onto formal notation and skip over the physical, chemical, and biological principles of language evolution within and among life-systems. We might seek to design computable models that can evolve toward the richness of natural languages through the °self-organization of interactions among local constraints°, such as the cellular automaton sketched in III.54ff. Or, we might deploy statistics across large language corpuses, in order to assess the °probability° that a given set or type of constraints will be more strictly or loosely applied; we might then determine which degrees of formality are appropriate to a description of a given language while keeping the supportive functionality in full view (II.78). This project would be much richer than the sparse formal use of statistics in ‘information theory’ (II.98), which computed only the relative probability of any given transition between one state in a sequence and another (e.g., two words in a sentence) but did not state the constraints that decide how predictable or surprising some words or longer stretches of text might be in a discourse; here too, the role of formal positions got overcomplicated.

174. Our enrichment of physics could address the general properties of non-linear systems that can condense or amplify, rarefy or enrich, stabilize or fluctuate, simplify or complicate, and so on. Linear systems, such as gentle streams, slowly reacting chemicals, or engines operating at low speeds, exhibit predictable behavior, can be broken down into simpler parts, and evolve in smooth and gradual ways. Non-linear systems, such as rushing rivers, explosive chemical reactions, and engines running at top speed, exhibit novel behavior and generate new structures, act as a whole, and evolve in abrupt and discontinuous ways that have been designated ‘chaotic’. Although nature is replete with non-linear systems, they were long neglected in physics because they required new approaches to mathematics like the interrelated theories of °complexity, chaos, and dynamical systems°, which demand extensive computer support (cf. III.54ff, 154, 163). Unexpectedly, a simple or deterministic system was found to manifest complex or non-deterministic behavior, and self-organization to emerge near the boundary between order and chaos (III.28, 44, 152f). Some chaotic systems attain astonishing precision from the order of self-similarity: they contain precise ‘models’ of themselves extending recursively downward to indefinite degrees of local detail, a bit like the ‘hologram’ wherein each part contains a copy of the whole (cf. III.68, 182). The celebrated ‘Mandelbrot set’ generates wondrous color graphics wherein swirls and islands move through endlessly finer and finer scales. The set corresponds to a ‘non-classical’ geometry in which a point is plotted not when it satisfies an equation but when it shows a certain evolutionary behavior during ‘iteration’ (repeating an operation). The arithmetic can be quite simple, e.g., if you take a number, square it and add in the result, then square the total and add that in, and so forth; yet the behavior can be breathtakingly complex. Intriguingly, chaos turns out to be not the total absence of order, but an extremely fine-tuned order of a previously unimagined kind, possibly connected to the fundamental indeterminacy of °quantum mechanics in the quasi-classical universe° (cf. III.15).

175. New analogies to language are not hard to envision. The formalists’ concept of ‘semantic features’ like ‘±Human’ could be replaced with the concept of °subsymbols that can converge to produce the meaning of a symbol without having their own meaning° (III.95). These are non-classical units that do not merely act as parts of a whole, but constitute a °subcritical baseline° for the rapid emergence of a meaning, much as order suddenly emerges along the boundary of chaos. The content of the meaning can thus adapt to suit numerous contexts, e.g., when we speak of ‘human language’ as distinct from ‘animal language’ (cf. III.106f). In other contexts, the meaning could be ‘possessing a highly evolved brain’, e.g., when we speak of ‘human intelligence’ as a biological condition for amplifying one’s genetic potential. In still other contexts, meaning could be ‘subject to limitations on accuracy or endurance’, as when we excuse our mistakes because we are ‘only human’. Perhaps post-classical science will supply some novel meanings for ‘human’, e.g., ‘having unlimited capacity for adaptive self-organization’.

176. Our enrichment of chemistry could address the evolution from inorganic to organic as a leading instance of enrichment in physical systems. The °catalyzing  potential of certain compounds for self-organization and critical mass° (III.59ff, 160) suggests how life-systems emerged and how the chemical base led evolution forward, as when amino acids evolved into living protein, or the metabolism spontaneously emerged. Here, an impressive analogy to language is found in the °genome, whose sparse material ‘alphabet’ of four chemical bases° supplies enormous data to send rich ‘messages’ all over the body (III.63ff). Indeed, nature is riddled with processes of control and communication upon a chemical basis, witness the cell types with specific ‘receptor molecules’ made of proteins and designed to interact in a ‘signal response system’ (cf. III.85, 160).

177. Our enrichment of biology could explore how the evolving organization of life-systems supports cognitive and communicative processes, which are prefigured in the genome and in the extraordinarily refined human nervous system and its in operations of °neurotransmission° (cf. III.58, 85; VIII.100). At some point, the processes reached the critical mass needed for language to be ‘externalized’ among whole communities of organisms (III.106f). Significant new properties emerged here, notably for condensing rich data onto a sparse material base of articulatory events and sound waves, but the design principles should be comparable to those of ‘biological languages’, e.g., when evolution amplifies the material base of a life-system to support richer data (III.49).

178. These enrichments of the several sciences might help reintegrate linguistics into the scientific community from which it has drifted somewhat in its quest for °language by itself°. The principle that all life-systems and their operations, including human beings and language, are subject to the constraints of physics, chemistry, and biology is hardly disputable, though it seems far more significant today than in more classical times. The real challenge is to show how those basic constraints apply and how they could be further enriched to support human cognition and communication in society. The enrichments can be sought in the constraints of culture, society, ideology, gender, emotion, and so on, and in the semiotic systems related to each of these, e.g., in the strategic interactions between talk and emotional displays (cf. IV.103-10; VIII.110ff).

179. We must also turn to the °human sciences°, such as anthropology, sociology, and psychology, along with their offshoots and intermediaries like sociolinguistics, social psychology psycholinguistics, cognitive psychology, artificial intelligence, and artificial life — all of which are converging under such headings as °semiotics, cognitive science, and discourse processing° (cf. section II.E). This turn is not actually an enrichment on our part, since these domains are at least as rich today as linguistics and discourse analysis. In the past, they were at times kept sparse by their own attempts at stringently classical enrichments of the natural sciences, notably in the °realistic and mechanistic models inspired by physicalism, positivism, and behaviorism in unified science°. There, scientists strove to construct sparsely objective models, methods, and explanations for richly subjective human processes and activities. Qualitative was rarefied into quantitative, and interpretation into observation and measurement (III.99, 158). Theories were argued from well-behaved special cases, e.g., a theory of human behavior argued from sparse experiments in animal conditioning (III.163). Such trends produced the usual °trade-off whereby the complexity squeezed out of the model reappeared in the connection between the model and its domain° (III.40, 172). The analogy between, say, stimulus-response chains and ordinary human activities like carrying on a telephone conversation to ask for help from a friend (V.57ff) can only be frankly metaphoric, e.g., in B.F. Skinner’s arid suggestion that people who are thinking about what to say or devising rational reasons are ‘automatically reinforcing themselves’ when his whole claim to scientific accreditation rested on the direct observability of action plus reinforcement. Such trade-offs are more damaging in the human sciences than in the natural sciences because the richer the phenomena, the sooner our simplification reaches critical dispersion and loses sight of essential properties, e.g., of the human capacity to generate or adapt constraints in context instead of just re-enacting ‘conditioned behavior’..

180. More recently, the human sciences have realized the adverse implications of reducing the complex to the simple and proposed instead to °convert complexity from resistive and disintegrative over to supportive and integrative°, viz.:

So far as the study of man [sic] is concerned, one may […] argue that explanation consists of substituting complex pictures for simple ones while striving somehow to retain the persuasive clarity that went with the simple ones. […] the French anthropologist Lévi-Strauss remarks that scientific explanation does not consist, as we have been led to imagine, in the reduction of the complex to the simple [but] in the substitution of a complexity more intelligible for one that is less. […] intelligibility, and thus explanatory power, comes to rest on the possibility of substituting the involved but comprehensible for the involved but incomprehensible.

Here too, a productive agenda would be to enrich the constraints within our models of human action and interaction and explore how such complexity might emerge during either the long-range or the short-range evolution of a culture or society. We might then replace the °category mistake of taking scientific and logical procedures to be the privileged model of the human processes and practices of thought and language° (III.169) with a model of ordinary reasoning and everyday activity whose immense richness is masked by their familiarity and efficien cy (II.101; III.190). Even, or especially, the apparently chaotic aspects can produce impressive order through self-organization.

181. On a higher plane, the ‘chaos’ that classical human science might behold in a post-classical human science of society, culture, and discourse, may also prove richly productive for the emergence of deeper-lying order. Post-classical science might underwrite a °permanent revolution that does not settle down into a routine normal science° (cf. III.147-55). The recent critiques by philosophers are an occasion not just to sharpen our awareness of paradigms but to °integrate the diversified alternatives°. The impending °revolution° would not be another Darwinian struggle for academic ‘survival’, but a recognition that the competitive conception of paradigm is obsolete and can yield to an integrative conception (III.24). We can now join to develop a °transdisciplinary meta-paradigm° that would gain support not by refuting and discrediting the others, but by subsuming their results and findings into wider contexts (cf. II.2, 132). We may yet discover some ‘universal’ design principles far deeper and more significant than the ‘linguistic universals’ postulated by the mentalists (cf. III.128f; V.2; VIII.12).

182. Such a science would at last be °self-similar° enough to span within its scope the cognitive, discoursal, and social moves of constructing theories and models, covering the whole range from science itself over to ordinary interaction. We can deepen our understanding of the various vital °dialectics°: °virtual versus actual, operation versus storage, theory-driven versus data-driven, top-down versus bottom-up, specifying constraints versus manifesting constraints°, and so on. We have seen the perils of trying to °freeze these dialectics into dichotomies° and viewing just one side in isolation. Now, we can explore how their full dynamics support both the unity and the diversity of knowledge and language.

III.F. Criteria for designing models

183. Like the concept of ‘system’, the concept of model is highly general and diverse. In ordinary discourse, a ‘model’ intuitively corresponds to an implied standard, e.g., a ‘role model’ or a ‘model husband’. In the discourse of engineering, a ‘model’ is a miniature or blueprint for something to be built, e.g., an ‘architect’s model’ of a new shopping mall. In the discourse of mathematics, a ‘model’ is a system whose ‘objects’ obey all the ‘axioms’ of the domain, e.g., of Euclidian geometry. From a post-classical standpoint, all these ‘models’ would be schemas: patterns of knowledge, ranging from common sense over to science, to whose °goodness of fit° may be abstract or concrete, and fine-grained or coarse-grained, and supports cognitive, discoursal, and social moves.

184. Whereas the term ‘model’ includes informal and practical schemas, °classical science° has preferred the more prestigious term theory, suggesting a formalized and theory-driven schema (cf. III.98). A ‘scientific theory’ should explicitly state which domain it describes or explains and how it can be tested and verified or falsified; in °classical reality, every domain should have a single correct theory whose content and design are built into the domain°. These requirements have fueled the development of scientific scripts stipulating how to carry out tests and how to design testable theories (III.1): a problematic result has been the widespread tendency to construct isolated or fragmented theories and to accentuate confrontations between alternative theories on the grounds that all but one must be errors (cf. III.7, 23f). Even so, the terms ‘theory’ and ‘theoretical’ have been used in diffuse and inflated ways for many schemas that were not tested or even testable. Some ‘linguistic theories’ have been essentially just scripts for using formal notations to write ‘rules’ and ‘rewrite’ invented sentences into symbol strings (III.172). Stated as a testable theory, such a script hypothesizes that language is °uniform, stable, and abstract and subsists on ‘purely linguistic’ constraints°; and this hypothesis has been soundly refuted by the three test scales of °coverage, convergence, and consensus° (III.168).

185. Here, it seems more prudent to adopt the term ’model’ in a neutral broad sense: any schematic representation of one system by another system, a sparse instance being the ‘internal model’ a simple life-system has of itself and its environment, and a rich instance being a human’s ‘model of the world’. We can also adopt the post-classical tenet, promoted by the °cognitive revolution°, that models are implicated in all knowledge and discourse. They cannot be independently tested and verified for their correctness in any absolute sense. But their °goodness of fit can be assessed by their supportive capacity for orientation and interaction within their domain°; and they can be °progressively tuned to yield finer approximations°. It may often prove strategic to apply multiple models to the same domain or to integrate among modeling styles such as psychological and sociological (cf. III.24, 38, 181).

186. To design its models, a °science of text and discourse° could profitably set an explicit agenda. I shall propose ten points such an agenda might have. First, we can formulate a statement of our ecological goals, expounding the motivations for our work and its relevance to the human situation. If our goal is to enhance the °freedom of access to knowledge and society through discourse°, our models need to identify the major factors, such as the on-line resources and constraints, that decide how well or poorly people can develop and use their °communicative competence° (I.59f; III.97, 109; IV.163; V.15, 70, 89; VI.30). If our goal is to formulate an °ecologist program°, then the model also needs to emphasize the co-operative and constructive uses of this competence over the confrontational and destructive uses (VII.37).

187. Second, we can define key terms and concepts within a systematic terminology and then use them consistently (cf. III.103; V.10, 87ff, 91f, 96-99, 102f). Whereas classical science has viewed terminology mainly as a set of straightforward labels that ‘refer to’ individual well-defined objects in the world, post-classical method views terms as °control centers for activating a schema or network of cognitive, discoursal, and social knowledge that dialectically interacts as a whole with some domain of a world-model° (cf. III.91, 136). Because this control is crucial for constructing and accessing the model-domain, the formulation and development of terminology must not continue to be a mere intuitive sideline of ‘doing science’, but should become a conscientious procedure to regulate the design of terms and to promote transdisciplinary consensus about their meaning, e.g., between cognitive science and discourse processing (cf. V.C). Instead of just coining erudite terms from Latin and Greek stems, as is fashionable in ‘scientific English’ (although — or is it because? — most English speakers have poor knowledge of Latin and Greek), we can adapt user-friendly terms from ordinary usage by either generalizing them (e.g. ‘system’, ‘model’, ‘meaning’, ‘problem’) or specializing them (e.g., ‘complication’, ‘chaos’, ‘progress’, ‘enrich’). We must then take care not to slip back into offhand ordinary usage, e.g., ‘progress’ meaning only ‘advances in technology’ (III.43, 103).

188. Third, we can assess the activities entailed in model-building as °cognitive, discoursal, and social moves° in their own right (cf. III.30, 182). Again, we could start from our basic postulates of knowledge and their correlated moves of °identifying, connecting, experiencing time, space, and substance, observing, measuring, and predicting° (III.8). These moves lead not just to the °convergence that engenders the object we can observe as a phenomenon, but also to the dual consensus of interobjectivity among each subject’s multiple experiences with the same object, plus intersubjectivity among the subjects who perceive or know objects as phenomena° (cf. III.84, 142). In science, the phenomena are typically °enriched and amplified° through ‘scientific moves’ like these:

(a) Data-gathering assembles prospective instances of the phenomenon, e.g., spontaneously occurring samples of English discourse, hoping to reach the critical mass where the total corpus can be considered a representative approximation (cf. II.78).

(b) Description tries to name the essential qualities of a corpus of phenomena, e.g., the functional categories and patterns of English discourse (cf. IV.B).

(c) Analysis seeks insights by breaking down the phenomenon into components, stages, levels, and so on, seeking to capture the relevant divisions or transitions in the organization, e.g., examining a conversation to see how speaking turns are allotted (cf. V.54-59).

(d) Comparison situates the phenomenon alongside others and draws insights from the similarities and differences, e.g., between the authentic text and alternative versions of it (cf. I.19).

(e) Generalization seeks to make statements about broad classes of instances, e.g., which Clause patterns are typical in English discourse (cf. I.21; IV.214-18).

(f) Interpretation assigns expansive significances to a phenomenon, e.g., that the distribution of English Imperatives is a sensitive indicator of commonsense assumptions about which ‘Processes’ can be actively initiated or refrained from (cf. IV.38ff).

(g) Explanation expounds the reasons why phenomena might be the way they are, e.g., that the diversification of ‘commands’ through Modals, Interrogatives, etc. reflects the increasingly diverse and specialized divisions of labor in English-speaking societies (cf. IV.38, 60).

(h) Hypothesis-testing treats the assumptions of a model as predictions and confronts them with fresh evidence, e.g., that all meanings associated with a word are initially activated and their connection strengths are then adjusted to suit contextual relevance (cf. III.81, 249-52).

(i) Experimentation actively elicits phenomena by constructing a stringently controlled setting where ‘test subjects’ do specified tasks such as executing an action or making a decision, e.g., when a text is presented on a moving screen and periodically stopped, and the reader has to say if a probe item flashed on another display is a real word (cf. III.249).

(j) Simulation builds an artificial analogue of a phenomenon so we can set it running and see how well it works and evolves, e.g., a computer program that runs simple simulated animals through millions of generations (cf. III.56).

(k) Accreditation decides what data or facts are worthy of being investigated and accounted for by established theories and categories, e.g., when a linguistic theory postulates a domain of underlying structure not foreseen in previous theories (cf. II.42, 84).

(l) Application works back to control the organization or operation of the phenomenal domain, e.g., when a description of the discourse strategies for expository texts is used in coursework for training naive writers (cf. VII. 268-88).

When a model gets constructed, these scientific moves are typically done in combination, though some may get more emphasis when priorities compete, e.g., when linguists dispute the merits of ‘descriptive’ versus ‘explanatory’ models. So the moves would not run in a tidy temporal or procedural sequence: data-gathering usually entails some provisional analysis, comparison, interpretation, and so on, while application should be anticipated during the whole course of research. And the moves can be recursive, e.g., describing descriptions, analyzing analyses, explaining explanations, etc.; and they can frame each other, e.g., comparing descriptions, evaluating analyses, interpreting explanations, etc.

189. Each °normal science° specifies the moves to suit its own goals. The °data-handling moves specified by mainstream linguistics for language by itself° (cited in II.25) have primarily been aimed at description and have proceeded by data-gathering and analysis, with special emphasis on collating, rarefying, and decontextualizing. The accredited results have been subjected to comparison and generalization in order to state the highest regularities of a language. In return, the hesitation to venture beyond language by itself made linguists wary of embarking on interpretation and explanation, which would have to engage the rich interaction of language with society and culture. Hypothesis-testing was mostly limited to consulting informants and, later, to introspecting (cf. II.25ff). The remaining moves were delegated to adjacent disciplines: experimentation to phonetics, psycholinguistics, and cognitive psychology; simulation to artificial intelligence; and application to applied linguistics and sociolinguistics. The delegation remained problematic as long as ‘theoretical linguistics’ persisted in regarding language as a °technical construction of a purely linguistic nature° and discounting the needs of the other disciplines, especially for applications (VII.320f).

190. Fourth, we can build models of our own activities as investigators in terms of the scientific moves enumerated above. Such a self-similar model can also show how our activities are analogous to general human moves for building models of world and society (III.182). Ordinary reasoning and ordinary discourse are rather informal and spontaneous modes for doing so, yet are surely much more systematic and efficien t than is widely credited (I.39; II.101; III.180). What they chiefly lack are explicit procedures for seeking a convergence of data, for negotiating consensus among various persons’ models, and for resolving salient differences or contradictions. The lack could be significantly offset if education were expressly reoriented around using discourse for actively and jointly constructing and integrating knowledge rather than for passively and singly absorbing and reproducing facts and figures (cf. VII.H).

191. Fifth, we can formulate programmatic priorities expressed as parameters for models. A research project or method might be characterized as:

(a) more data-driven to exploit manifested constraints, e.g., observing planets with a telescope, versus more theory-driven to specify expected constraints, e.g., writing differential equations for celestial motion (cf. III.163);

(b) more quantitative to count and measure, e.g., naming winds by velocity on the Beaufort scale (‘strong gale’ at 47-54 mph., ‘whole gale’ at 55-63 mph., etc.), versus more qualitative to characterize and classify, e.g., naming winds by their effects (chimney pots blown off by a ‘strong gale’, trees uprooted by a ‘whole gale’, etc.);

(c) more fine-grained for elementary entities such as quarks, versus more coarse-grained for sophisticated ones like jaguars (cf. III.15);

(d) more analytic in dismantling the phenomenon, e.g., immediate constituent analysis, versus more synthetic in assembling it, e.g., a biological model of the emergence of an internal ‘language’ (cf. II.34; III.81);

(e) more generalized for wider coverage in sparser detail, e.g., the laws of thermodynamics in physics, versus more specialized for narrower coverage in richer detail e.g., the processes of the metabolism in biology (cf. III.163, 176);

(f) more formal for the shapes and arrangements of objects, e.g., in a structuralist syntax, versus more functional for their uses as means for ends, e.g., in a systemic lexicogrammar (cf. II.32);

(g) more modular for a single subdomain, e.g., linguistic levels studied independently, versus more integrative for the whole domain, e.g., discourse processing studied comprehensively (cf. II.59ff, 112);

(h) more deterministic for exact conditions and inexorable causalities, e.g., Newton’s laws of gravitation, versus more non-deterministic for context-sensitive conditions and probable causalities, e.g., Feynman’s sum-over-histories model (cf. III.36, 163);

(i) more static for a single state, e.g., a ‘synchronic’ description of a language outside of time, versus more dynamic for action and evolution, e.g., the meaning of a discourse defined by the ongoing interaction of constraints (cf. II.24; III.134ff);

(j) more closed-ended for fixed boundaries of objects or events, e.g., a logical system bounded by axioms, versus more open-ended for indefinite ranges of objects or events, e.g., a cognitive schema for understanding folktales (cf. V.118);

(k) more short-range for a modest span of time, e.g., how an infant acquires language between the ages of about 12 to 24 months, versus more long-range for an extended span, e.g., how language evolved in the human species (cf. III.114ff; VII.142ff);

(l) more etic for actions and events as they are performed, e.g., the seizure of wealth by a repressive ruling class, versus more emic for actions and events as they are understood within the cultural system, e.g., a caste system decreed by God (cf. II.92; VIII.24-27);

(m) more materialistic for attributing events to material constraints, e.g., food supplies, versus more idealistic for attributing events to mental constructs, e.g., binary oppositions (cf. VIII.16, 20);

(n) more motivated for inquiring how phenomena came to be organized as they are, e.g., language constrained by knowledge of world and society, versus more arbitrary for taking the organization as an abstract pattern in its own right, e.g., language by itself as a system of mutually different signs (cf. II.44, 48f, 89);

(o) more natural for seeking the motivations in the processes of nature, e.g., sound systems in human articulation, versus more transcendental for looking beyond nature, e.g., all languages based upon one grand ‘universal grammar’ (cf. II.29; III.128f).

Models in °classical science° have often been more data-driven, quantitative, fine-grained, analytic, generalized, formal, modular, deterministic, static, and closed-ended. These priorities tend to be mutually supportive, e.g., analytic and formal approaches favoring a fine-grained modular, deterministic, and static design. They simplify designing and testing also, and uphold the °classical reality of free-standing objects and objective facts°. °Post-classical science°, in contrast, can make progress by continually resetting priorities to promote a strategic fit with domain. For a °science of cognition and communication in society°, models should be more qualitative, synthetic, functional, integrative, non-deterministic, dynamic, and open-ended, as well as materialistic, motivated, and natural; and should work to balance data-driven with theory-driven, fine-grained with coarse-grained, generalized with specialized, short-range with long-range, and etic with emic (III.218) (III.220). Instead of taking classical reality as our obligatory model-domain, we °bracket it and investigate how it could emerge and evolve through a modeling style such as realism°. And we can cross-refer our priorities, e.g., by exploring the function of formalism (II.59), or the material motives for giving idealistic accounts (VIII.16).

192. Sixth, we can formulate questions to assess the merits of models that have already been constructed and presented, e.g.:

(a) What is the goodness of fit between the model and its domain, e.g., in enabling effective applications for human strategies of discourse?

(b) How parsimonious is the model in representing a wide range of knowledge with compact means, e.g., both cognition and communication?

(c) What is the ecological validity of the model in relating the results of science to the conditions of doing science, e.g., relating experimental findings on cognition to the experimental procedures for knowledge-gathering?

Classical science, which has concentrated on °testing specific hypotheses (‘theories’, ‘predictions’ etc.) to be verified (‘validated’, ‘proven’ etc.) or falsified (‘refuted’, ‘disproven’ etc.)°, has emphasized parsimony and goodness of fit, sometimes even when doing so meant studying °epiphenomena more than the main phenomena° (III.45, 170), e.g., abstract sentence structure rather than utterances in context (II.38). Today, ecological validity is prominent (II.133), especially when science undertakes to support an °ecologist program for sustainable life°.

193. Seventh, we can propose a scheme for coordinating the roles of the investigator, such as:

(a) The investigator makes observations without participating, e.g., taking notes or making acoustic and video tapes of a conversation (cf. V.57ff).

(b) The investigator makes observations while participating in an ordinary role, e.g., joining in the conversation (cf. VIII.149ff).

(c) The investigator makes observations while participating in a special role, e.g., as a fact-finding interviewer or as a therapist giving counsel to a neurotic client (VIII.142ff).

(d) The investigator participates as the constructor of a controlled situation for making observations, e.g., of a laboratory experiment for test subjects who may have been sorted by their age, social background, personality, and so on (III.197f).

(e) The investigator participates as the interpreter of data displays from a large corpus, e.g., by comparing typical collocations and formulating the constraints they follow (II.76).

In classical science, the investigator was to observe while participating only as the constructor of experiments. Post-classical science, above all the °new physics°, has shown that purely data-driven detached observation is not meaningful (III.12f, 17). In the human sciences, the role of the observer as participant has been more widely acknowledged, because some rich domains of data can be accessed no other way, e.g., a fieldworker who wants to learn and describe an unrecorded language will have to join in cultural activities (cf. II.27).

194. Eighth, we can we can seek convergence and consensus by diversifying and integrating our sources of data and our procedures for gathering data. Instead of assuming, as mainstream linguistics has, that all data are equally indicative of the state of a language, we can explore how different data sources reflect their particular conditions. Unannounced recordings of conversations might yield more spontaneous data, whereas announced recordings might make people feel self-conscious about displaying ‘good manners’ or ‘elegant style’, or feel hesitant to speak at all. Or, people might soon grow accustomed to being recorded, e.g., if it is done in an everyday setting like their homes; or they may enhance their self-assurance by imitating the interviews they often see in public media such as television talk shows or opinion polls. Or again, we can conduct interviews with varying stances, e.g., active versus passive, or sympathetic versus detached; and also collect interviews conducted by regular participants, e.g., therapist with client (cf. VIII.142ff).

195. Handling data from ordinary conversation is harder than you think until you’ve tried. The data already begin to get °normalized when we write them down in standard orthography°. Divisions into words and clauses are visually enhanced by spacing and punctuation, which thereby perform some unacknowledged ‘explanatory work’; in return, data are diluted about the prosody and intonation of speaking, e.g., pacing, pitch, and volume, which punctuation and typography can only approximate. A different and far more drastic normalization is imposed by °formal notations such as syntactic ‘trees’ or symbolic formulas that accept only ‘well-formed’ data° (II.41; III.172). We can resist such normalizations by scrupulously transcribing spontaneous speech with the hesitations, pauses, false starts, non-word sounds, and so on, found even in the discourse of academics, e.g. (/ = brief pause, // = longer pause):

[33] as far as I know / no one has done the / in a way obvious now and interesting problem of // doing a / in a sense frequency study of the alternative // syntactical / uh / in a given language / say like English / the alternative // uh / possible structures

A further way to resist, one popular among sociologists, is to transcribe into an orthography that imitates spoken sounds more than the standard does (V.56), as in Roy Watson’s study of the discourse of police investigators with suspects:

[34] ‘N ah (sed) wo::w yiknow (0.3) ‘n den she own the’ sahd a’ God yih know  [and I  said well, you know  and then she [is] on the side of God you know]

Another °paradoxical trade-off°: the more richly natural data are transcribed, the less ‘readable’ they become for non-specialists, including the recorded speakers themselves (cf. II.58). A transcription also marking pitch, volume, facial expression, gestures, emotional signals, and so on, would be still richer but less readable, so that our informants could hardly be expected to verify it.

196. As we saw in Ch. II, the perennial problems of how to handle language data are now being profoundly reshaped by the advent of large corpuses. We are now attaining intensely data-driven methods for identifying types of constraints, such as: (a) lexical constraints applying to collocations of specific words (e.g., ‘search warrant’ rather than, say, ‘hunt warrant’); (b) grammatical constraints applying to categories (e.g., the high occurrence of ‘warrant’ with Infinitives and with non-human Third Persons); (c) semantic constraints applying to the cognitive organization of processes in the world (e.g., Pejorative Events and Actions ‘warranting’ reactions); or (d) pragmatic constraints applying to the social organization of human interaction (e.g., who has the authority to say what is or is not ‘warranted’) (cf. III.209). °Mainstream models with separated levels defined by the size and constituency of their respective formal units° should be reframed by models which highlight the °functional interaction of multiple means and ends° (cf. II.59ff). And the time-worn dichotomy between °language by itself versus language use° can finally be reconciled.

197. Ninth we can investigate the respective conditions under which data are gathered. Think-aloud protocols of people planning and deciding what to say or write can supply a running commentary, but only on those aspects that can be made conscious (cf.II.239). Experiments can coordinate such tasks as recognition of form (e.g., did this exact sentence appear in the text?), recognition of content (e.g., did this event happen in the story?), cued recall (e.g., which actions did this character perform?), free recall (e.g., tell everything you can remember), and prediction (e.g., what will happen next in the story?). The language data for the experiments can be varied among word lists, sentences, paragraphs, and whole texts of specified types (e.g. stories, instructions), and can be presented as recorded speech or as writing on a page or on a timed computer display, or filtered against a background of distractors or noise. Simulations can test the computability of on-line models, e.g., ones that analyze language by formal rules versus ones that use connectionist networks. Experimental findings or computer simulations in the laboratory can provide hypotheses about spontaneous discourse, while findings on spontaneous discourse can suggest what experiments or simulations might be worth doing.

198. The conditions of the scientific experiment merit special attention as three modes of transaction. In the social transaction, one group (experimenters) is empowered to select and shape the task for the other group (test subjects), either discounting their social status or else fitting it to a few specific criteria (e.g. middle class versus working class). In the cognitive transaction, specialized methods are deployed to ‘read’ the interaction between material (e.g. a muscle response) and data (e.g. a state of anxiety). In the discoursal transaction, experimenters give instructions, collect verbal responses, interpret and publish findings, etc. The experiment allows us to ‘read’ data in physics from physical systems (e.g. photons in the ‘EPRB’ set-up, III.13), and in biology from non-human life-systems (e.g. frogs), even though neither type of system ‘communicates’ with us in the ordinary sense. In psychology, sociology, and anthropology, both humans and the data about them are so ‘communicative’ as to surpass our strength to ‘read’ it all.

199. Tenth and last, we should draw up a comprehensive transdisciplinary research plan for exploring the wide range of cognition and communication in society, not by prescribing certain procedures and proscribing others, but by °diversifying and integrating our procedures in dialectic ways° (III.194). In a °bottom-up or data-driven mode°, we assemble large corpuses of data for respective text types and try to distill out the relevant aspects of each. In a °top-down or theory-driven mode°, we construct process-models of the knowledge networks, schemas, constraints, and so on, being applied to texts. Our research plan succeeds to the degree that we can balance the two modes within an account matching the strategies human participants also use for balancing the two in everyday discourse (cf. III.156). And we can seek to reconcile the orientations of theoretical versus applied research by expressly designing theories of language and discourse that support applications to social or educational problems, e.g., the teaching of native or non-native languages, or the uses of language and discourse for special purposes in technical fields (Chs. V and VII).

200. Such a ten-point research plan might relieve us from relying on personal or institutional commitments, decisions, or coincidences, e.g., serendipitous gatherings of text linguists or discourse analysts at a few progressive universities. As these regional concentrations yield to global distributions, larger projects become feasible, provided that investigators can interact over longer distances and shorter time intervals, e.g., through computer networks like Internet.

201. Formulating and implementing such a transdisciplinary research plan may seem arduous if not utopian, but also pressing and productive. Perhaps all science has a utopian prospect in seeking a fuller understanding of the universe, but has often camouflaged it by working on small, tidy research issues. The large, messy issues we face today can only be managed by a concerted and regularly reassessed plan of research for coordinating special models within a general model and for integrating empirical and theoretical insights from relevant disciplines all across the academic and social spectrum (II.132f). In this context, seeking ‘new foundations’ might not seem disproportionate.

III.G. Modeling linearity

202. The most obvious ‘fact’ about language might appear to its linearity. Words and phrases must come either before or after each other in a sequence, whether in the more temporal acoustic dimension of speaking or in the more spatial visual dimension of writing. Many linguistic theories and methods have exploited the ‘fact’ by devising procedures for identifying units and segmenting sequences, usually after writing them down in a visual format that enhances clarity and does some ‘cheap explanatory work’ (II.24, 29, 34, 38; III.195; IV.1). Sometimes, the linearity of the notation got short-circuited onto the terminology, viz. ‘chain’, ‘string’, ‘left’, and ‘right’ in generative grammar. Still, the primary dimension of linearity is surely time, whereas space is a secondary visual projection of alphabetic orthography that seems natural only because our literacy has rendered it so familiar.

203. Linearity appears to obey the material constraint that people can utter or perceive only one word at a time. But language is actually our best medium for simultaneously doing multiple actions and integrating multiple meanings and functions, just as a text maintains multiple economies (I.17; III.130; VII.141). And the current word is just one ‘cue’ that also typically has multiple functions , like a set of elaborate instructions to perform several operations on meanings by grammaticalizing, lexicalizing, linearizing, and integrating them (cf. II.48, 75; III.62, 92, 232ff). The linear sequence emerges from a °steep condensation° by the speaker/writer and has no meaning or significance — indeed, is not even language but merely an artifact — except through an equally °steep amplification° by a hearer/reader (cf. I.41; III.107, 177). So the linearity of language is not an obvious ‘fact’ to be taken for granted but a complex dimension we cannot isolate without moving to °critical dispersion and losing essential properties°. We may ignore our own amplifying moves while we analyze or manipulate the written artifact, but we cannot refrain from amplifying it.

204. As °complex systems operating in real time°, language and discourse require linear principles whereby °constraints° can be accessed and shared among sets of operations or choices at various points in the sequence. These principles would coordinate the constraints in processing domains whose linearity is quite disparate, e.g., the articulation of sounds versus the flow of topics. Seven general-purpose linear principles appear plausible:

(1) The pacing principle enables the on-line sequence to be temporarily speeded up, slowed down, or suspended.

(2) The look-back principle enables the retrospective access to the prior text.

(3) The look-ahead principle enables the prospective access to the later text.

(4) The merging principle chooses one among each set of competing options or combines the set into one option.

(5) The listing principle handles the juxtaposition of comparable items in a sequence.

(6) The core-and-adjunct principle distinguishes between central versus peripheral entities.

(7) The loading principle regulates the degrees of focus, emphasis, or salience that draw varying amounts of processing resources.

These principles are visually portrayed in Fig. III.36. Pacing regulates the rate and timing of the surface sequencing by speeding up when the current data are fairly familiar, simple, and determinate, slowing down or pausing when they are not, and otherwise moving at a moderate rate. Look-back and look-ahead exchange constraints with non-current stages of the sequence. Merging reduces branching options to just one,  whereas listing organizes a series of recurrent or  parallel options. Core-and-adjunct makes some items into control centers for the rest, whereas loading decides how and where available resources should be distributed.

205. These principles can be supported with both anecdotal and experimental evidence. Pacing is the most accessible ‘front end’ of the overall scheduling of operations and can be carefully measured in psychological and psycholinguistic research. Typical measures include ‘reaction time’ or ‘latency’ (interval between perceiving the cue and starting to perform), ‘fluency’ (the speed and smoothness within the sequence), and ‘time on task’ (interval from start to finish). By themselves, time measures are more problematic than is widely imagined, because test persons might be using up time for multiple reasons. We see this problem clearly in the many factors uncovered by empirical research on speech pauses. Textual factors included the text type and the degree of integration, coherence, abstractness, or ambiguity. Physiological and psychic factors included breathing patterns, anxiety, and stress. Task-based factors included length, difficulty, and modality (speaking versus reading aloud). Speaker-based factors included age, social class, exhibitionism, deceitfulness, habitual rate of speaking or writing, and inclination to adjust to other speakers. Hearer-based factors included whether hearers were present (versus talking to oneself), and whether they were visible or invisible (e.g., in a face-to-face or a telephone conversation), whether they exhibited approval or disapproval, and whether they were disposed to interrupt. Also, the organization of conversation into turns among speakers can be influential, e.g., when you pause to encourage a response, or do not pause at the ends of clauses lest you lose your turn (cf. V.57ff).

206. Involuntary evidence can be found for look-back in ‘reactivation’ when an item reappears out of order [35-36], and for look-ahead in ‘pre-activation’ when an item appears too soon [37-38]. ‘Restarts’ are common too, where the speaker repeats parts of a sequence either with the same words [39] or with changes [40]. These ‘misplacements’ indicate that the operations for linearizing are somewhat non-deterministic though highly efficien t.

[35] John gave the goy — John gave the boy

[36] The field goal is done by kicking the ball, which is held up by a team mate, ball through the uprights of the goal posts.

[37] the sird and — the bird — the third and surviving brother

[38] The Federal Defense Budget is the only government that program [program that] will be increased.

[39] the people using the / using the election to make decisions

[40] the strange thing is that / the hypocritical thing is that these people are the ones who are for the death penalty

Look-ahead is also responsible for ‘miscues’ when people alter a text they are reading aloud, especially when, as in [41-42], the interpolated item (e.g. ‘dries’ and ‘gain’) both resembles the printed one (e.g. ‘dies’ and ‘gain’) and is °collocative° with text items (e.g. ‘dust’ and ‘corn’). Miscues offer clues about which words or concepts are °activated in memory°: for a misleading version of a text about ‘rocket flight’, whose ‘style’ variations are discussed in V.20-38, college-student readers said ‘rocket’ for [43] and ‘launching pad’ for [44], and later reported having read those items. Here we get at least some glimpses of the normally submerged operations of discourse processing (I.34).

[41] Through dust where the blacksnake dries [dies] (James Dickey)

[42] My crop of corn is but a field of tares,

 And all my good is but vain hope of grain [gain] (Chidiock Tichborne)

[43] With a roar and a burst of flares, the giant rocked on its pad.

[44] It plunged down to the earth’s surface 40 miles from the launching padded by landing gear.

Voluntary evidence for look-back and look-ahead is found in major resources of cohesion for maintaining and compacting the discourse material of texts, e.g., °recurrence, parallelism, paraphrase, pro-forms, and ellipsis°, which will be surveyed in section IV.C.

207. Involuntary evidence for merging can be found in ‘hybrids’ like [45] and in unintentional indeterminacies like [46] in English newspaper headlines with reduced phrasing. Or, merging may be suspended to create intentional puns or double-entendres like [47-48].

[45] they have to make a decide [‘make a decision’ vs. ‘decide’]

[46] Congress votes for running trains over union workers ([Lafayette IN] Courier and Journal 28/6/1992)

[47] One of the biggest things to do in New Orleans during Mardi Gras is sucking heads and pinching tails. I am referring to eating crawfish.

[48] Ben Battle was a soldier bold and used to war’s alarms;

       But a cannon-ball took off his legs and he laid down his arms! (Thomas Hood)

Puns are traditionally disdained, perhaps because they threaten to disrupt the relation of a word to its meaning with °arbitrary coincidences and frozen accidents°.

208. Straightforward evidence for listing can be seen in the sequence of listed items. Listing complements merging by allowing alternatives to be selected one by one. In English, the greater self-sufficiency of °Content Words° like Nouns, Verbs, or Modifiers [49], makes them more suitable for listing than °Function Words° like Articles or Prepositions [50] (cf. II.35, 69). By far the most frequent listing is two items, and the next is three. After that, frequency plunges sharply, and lists of four or more may seem verbose [51] or bureaucratic [52] unless marked effects are desired, e.g., the nine present participles in [53] invoking a foolish excess of ‘taboos’:

[49] the Beduin had assets of mobility, toughness, self-assurance, and courage 231

[50] government of the people, by the people, and for the people (Gettysburg Address)

[51] her lovely skin, like dear sweet white old silk

[52] any comment, request, suggestion, or proposal which is obscene, lewd, lascivious, filthy, or indecent (Florida Statute)

[53] the entire cupboard of Protestant taboos against drinking, lusting, gambling, staying out late, getting up late, loafing, idling, lollygagging around the streets, and wearing Capri pants (Pump House Gang)

Processing exploits familiarity to gain simplicity by applying comparable operations to each list member, thus interfacing with look-back and look-ahead as well.

209. The assignment of core or adjunct status in a sequence follows from the concept of °control centers° (III.204). Typical cores would be the Noun in Subject position plus the main Verb expressing the Process (e.g., ‘giant rocked’ in [43]), while rest would be adjuncts (e.g., ‘With a roar and a burst of flares’ plus ‘on its pad’ in [43]). Many visual graphics used in representing sentence structure reflect this design, e.g., the networks displayed in IV.18. Empirical evidence might be found in the tendency of text recipients to recall such cores better than adjuncts and to use them more often in making a paraphrase or summary of a text. But other factors have been observed too, such as adjuncts being recalled because they evoke salient mental images, e.g., Adjectives for colors. Such observations again raise the troublesome question of which types of constraints apply (cf. III.196): whether the effects are due to (a) the specific lexical item (e.g. ‘warrant’), (b) the grammatical category (e.g. Verb), (c) the semantic concept (e.g. world knowledge about how warrants are served), or (d) the pragmatic action (e.g., announcing a warrant for the arrest of an Italian Prime Minister on charges of corruption). And again, we recognize the need for °functional models that can integrate the diversity of means and ends in language°.

210. Loading reflects and regulates the degrees to which items are strategic, e.g. crucial for a goal [54]; salient, e.g. strongly emotional [55-56]; or surprising, e.g. a flat contradiction [57].

[54] ‘The trial cannot proceed’, said the King, in a very grave voice, ‘until all the jurymen are back in their proper places — all’, he repeated with great emphasis, looking very hard at Alice, [who] had put the Lizard in head downwards

[55] ‘It’s new I tell you — I bought it yesterday — my nice new rattle!’ and his voice rose to a perfect scream.

[56] ‘The horror of that moment’, the King went on, ‘I shall never forget!’

[57] ‘I could show you hills in comparison with which you’d call that a valley’. ‘No, I shouldn’t’, said Alice, surprised into contradicting her at last: ‘a hill can’t be a valley, you know. That would be nonsense —’

Loaded items can be signalled in speaking by intonation, e.g., increasing the stress, pitch, or volume (‘grave voice’, ‘great emphasis’, ‘perfect scream’), and in writing by typographics, shown in the discourse samples in this volume as small capitals but in other printed matter also shown by bold, italics, LARGE CAPITALS, or exclamation marks (cf. IV.209, 219; VIII.120). The loaded items may be either cores (e.g. ‘rattle’) or adjuncts (e.g. ‘all’, ‘can’t’) and may appear in various positions of an utterance or ‘tone group’, most often toward the end for °end weight°, e.g. [55], but also °fronted (e.g. Direct Object ahead of the Verb [56]) or recurring by themselves, e.g. [54].

211. Loading may also depend on what is ‘close to the ego’. Reflecting °patriarchal mainstream culture°, frozen two-part collocations typically start with the element that suggests ‘here, now, adult, male, positive, singular, living, friendly, solid, agentive, powerful, at home, and patriotic’. Hence, we find ‘this and that’, not ‘that and this’, or ‘come and go’, not ‘go and come’ (‘here’ feature); ‘sooner or later’, not ‘later or sooner’ (‘now’ feature); ‘husband and wife’, not ‘wife and husband’ (‘male’ feature); and so forth. Closeness to the ‘ego’ can also influence the arrangement of Clauses, as when newspaper reports about soccer matches consistently mentioned players of the home team in Subject position and players of the opposing team in the Predicate.

212. The seven principles of linearity can be correlated with our four °design parameters° in III.25ff. Look-back and look-ahead can control fluctuation and novelty. Merging and core-and-adjunct can control complexity and indeterminacy. And pacing, listing, and loading can control all four parameters. When control drops sharply, the linear principles would °degrade°. Operations wouldn’t look very far back or ahead, and would be beset by reactivations, pre-activations, restarts, or miscues; merging would leave seriously resistive indeterminacies; core-adjunct patterns might yield to mere lists, as in the ‘telegraphic style’ of some aphasics; and so on. These problems signal the °degradation that sets in during overload° — a more significant factor in human performance than has been widely appreciated (cf. III.214; VII.23, 66f, 83).

III.H. Modeling resource constraints

213. Issues of loading and overloading indicate how cognitive, discoursal, and social processes must respect the constraints on behavioral and cognitive resources (cf. V.30; VII.23, 84, 236; VIII.106). At least ten major factors have been found to regulate the supply and demand of resources, as shown in Fig. III.37; demand increases with an ‘upward’ move on each parameter and decreases with a ‘downward’ move.

213.1 Motor precision concerns how exactly the activation of neurons, tissues, muscles, and so on should fit detailed specifications. Greater precision is essential for difficult high-performance tasks such as throwing a javelin, but lesser precision suffices for routine tasks such as walking.

213.2  For attention, attentive tasks compete with concurrent tasks, whereas automatic tasks do not. Difficult tasks shift from largely attentional for novices toward largely automatic for experts. In a task as complex as discourse processing, numerous aspects need to be automatized, e.g., the articulation of speech sounds, so that attention can be devoted to less controllable aspects, such as how to persuade a reluctant audience to accept your beliefs or attitudes.

213.3 Direct perception concerns how directly operations are connected to what you can perceive. Tasks for obtrusively manipulating °coarse-grained hard-coupled objects° like stones (III.84) are more immediate and mentally less demanding (though physically more demanding on motor resources), while tasks for unobtrusively organizing °fine-grained soft-coupled objects° like philosophical essays are less immediate and mentally more demanding.

213.4 Feedback is data which the environment supplies about your current actions and which you use to control your further actions: the more you need to gather and process, the harder you have to work. In conversation, participants use all sorts of feedback from each other, including silences to infer how well the interaction is going (V.57f). In writing, some children or novice writers are so distracted by getting ‘surface’ feedback on their handwriting, spelling, and punctuation that inadequate resources are left to plan out the topic flow (VII.231). A skilled writer would need intense feedback only when writing down difficult or erudite words or when proofreading final galleys for publication.

213.5 Stress concerns the excitement or tension under which the performer is acting, higher stress consuming more resources than lower. Major contributing factors include °cognitive uncertainty and emotional anxiety°. Stress tends to go in cycles: a rise in anxiety interferes with performance, which raises the anxiety, which intensifies the interference, and so on (cf. VII.81, 84).

213.6 Noise designates irrelevant but competing events in the environment, higher noise levels demanding more resources to disregard distractions. The term ‘noise’ extends from the everyday sense of loud, confused sounds to the technical sense of unwanted signals in communication systems.

213.7 Motivation subsumes all the purposes for doing the task, e.g., to gain a reward, become more skilled, or help your friends or colleagues. Lower motivation makes you expend more resources on forcing yourself to do °alienating tasks that are forced upon you°; higher motivation assists you when doing °actualizing tasks that enhance your potential° (cf. VII.5).

213.8 Interest is the degree to which the task seems enjoyable, engaging, and challenging as opposed to tedious, boring, and facile, whether or not you stand to gain by doing it. As with motivation, more resources are needed for lower degrees than for higher degrees. Even a routine, well-paid task can be a strain once you’ve lost interest in it and your skilled performances seem empty.

213.9 A serious problem makes the attainment of your goals unlikely and drains more resources, whereas a mild problem is a minor hindrance. If you look ahead to failure, you make the problem more serious and are more prone to struggle and flounder than if you look ahead to success (VII. 83).

213.10 Timing is the rate, speed, or duration allotted to the performance of the task, tighter timing being more costly to maintain than looser timing. On °open-ended tasks°, such as writing a scientific journal article, the performer can set the thresholds for initiating and terminating the activity and can return to continue or change later on. On °closed-ended tasks°, the thresholds are either built into the sequence, such as jumping over a hurdle, or are arbitrarily fixed by some institution, such as having exactly sixty minutes to take a standardized test that was deliberately designed to be too long (VII.80). Though speed is popularly regarded as a signal of skill and intelligence, experiments show that °speed is often traded off against accuracy°.

214. When too many demands pile up, overload sets in and performance degrades below the performer’s potential (cf. II.99; III.21; VII.23, 66f, 83). Overload can be modeled as a threshold of °critical dispersion°, where operations disintegrate or break down altogether. Common causes include a stress-interference cycle, a burst of noise, a drastically serious problem, or a compulsion to work at top speed. A gradual build-up of disorientation may peak in a sudden flip-over into utter bewilderment or alienation.

215. Still, an increased demand on one factor need not worsen performance if another factor can compensate. A certain noise level, for example, has been found to improve performance when people allot more attention. Motor precision can be upheld by adjusting the timing. Rising motivation may offset the seriousness of a problem; and rising interest may help to resist stress. Moreover, some people can plainly work better within their resource limitations than others can, and seldom undergo degradation or overload. To say they have ‘high intelligence or aptitude’ provides no account of how they attained their advantage, which ought to be the central issue for strategically designing methods of socialization and education but, due to the °fundamental contradiction between inclusive theory versus exclusive practice° (I.6), has been persistently neglected (Ch. VII).

216. Such an account must explain how the design of a system can evolve to do the same tasks with less resources or to do new tasks that were too demanding before (cf. III.97; VI.30). In terms of our four design parameters in III.25, a °classical mechanistic account° assumes that resource demands always rise along with rises in fluctuation, novelty, complexity, and indeterminacy. But a °post-classical account° would stipulate that such rises would not draw heavier resources if they are °supportive rather than resistive° (III.42ff). Optimally, the design of processing should adapt to fit the design of the task or domain, e.g., by assuming the appropriate degrees of simplicity or complexity. Quite plausibly, discourse processing is a system with precisely this adaptive capacity, such that talking about tasks, e.g. describing and explaining them, can supply strong support for doing them without overload (cf. VII.114, 194, 211, 235, 263). Discourse can adjust not merely its own design parameters but also those of the activities it accompanies (cf. III.108) — again provided we understand how design can be encouraged to evolve (III.97).

217. Our understanding would be much enhanced if we could reliably estimate whether the respective resource demands should decrease or increase along with the parameters of operations. On the whole, fluctuation may go either way, decreasing demands if it raises degrees of freedom (e.g., when you pace your own work) or increasing them if it lowers predictability (e.g., when the task keeps changing); the same holds for complexity, decreasing if it is supportive and integrative (e.g., when you produce a genuine work of art) or increasing if it is resistive or disintegrative (e.g., when you are learning a difficult sport). Novelty, in contrast, should sharply increase demands except for motivation and interest, where it decreases them by spurring the performer on (e.g., when you are hiking in exotic places). And indeterminacy is quite demanding for handling hard-coupled data (e.g., when you operate a console with many switches), but may be supportive for handling soft-coupled data (e.g., when you read a poem). Noise would be most demanding when if it keeps fluctuating (e.g., when you are surrounded by noisy neighbors with small talkative children); it may not even get processed for novelty, complexity, and determinacy. But you may neutralize it all by masking it with a source of stable noise (e.g., by jamming a TV or radio between channels and turning up the white noise).

218. Though quite provisional, these estimates suggest that evolutions of design to significantly enhance general performance capacities are too elaborate to occur easily, let alone by fortuitous chance (cf. III.3, 97, 163; VII.95, 174; VIII.118). If people shun overload by seeking a stable, familiar, simple, and determinate environment, their capacities cannot evolve and their sagging motivation and interest can further depress their performance. To embark in search of novelty and complexity, as artists and scientists often do, people should be highly motivated (e.g., in actualizing their potential) and interested (e.g., in what will happen next) to help focus attention and gather feedback (e.g., reading back over your science text to see if you have been clear and consistent). Major evolution would require developing conscious strategies for operating well within your outer limits while managing fluctuation, novelty, complexity, and indeterminacy in supportive rather than resistive ways. Again, such evolutions can be ably assisted by the strategic potential of discourse processing (III.108, 216). Discourse can enhance motor precision, focus attention, monitor immediate perception, provide feedback, relieve stress, blend out noise, improve motivation, excite interest, solve problems, and budget timing. Showing how to achieve this will be another central question for our °science of text and discourse°.

III.I An interactive model for text and discourse processing

219. We can now return once more to the design of a model of discourse processing. In terms of the programmatic priorities listed in III.191, most models of language propounded in °mainstream linguistics° have been similar to those of °classical science° in being fine-grained, analytic, generalized, formal, modular, deterministic, and static, but different in being qualitative, idealistic, arbitrary, and sometimes transcendental. The task was to describe the language system and language data as they are already given, whether as a °corpus of authentic data° or as an °invented set of idealized grammatical sentences°, rather than to inquire how and why the data got produced or received. Methods highlighted the ‘shallow’ linearity of language, visually enhanced by ordinary orthography or special notations. The °formalist scheme of levels defined by their segmentable language units° would project discourse to be a °bottom-up process of fetching or recognizing smaller units and stringing them into bigger units like a row of building blocks° (II.59ff), without support from the °top-down semantic and pragmatic control of topics and goals applying the constraints of world and society°. If, as argued in II.D, this projection has blocked °coverage, convergence, and consensus°, we need to look elsewhere for a model of discourse processing.

220. We return to the proposal in III.191 for a °science of cognition and communication in society° to develop models which are more qualitative, synthetic, functional, integrative, non-deterministic, dynamic, and open-ended, as well as materialistic, motivated, and natural; and which seek to balance data-driven with theory-driven, fine-grained with coarse-grained, generalized with specialized, short-range with long-range, and etic with emic. Here, discourse could be modeled as a °dynamic multi-system comprising several processing domains° defined not by their formal linguistic units but by their contributions to the production and reception of text and discourse. These domains °interact, cooperate, and interchange constraints, coordinating bottom-up with top-down and local with global°. The discourse is supported not by the °total language system with a frozen design°, but by a °current partial version with an evolving design to merge standing constraints with emergent constraints, and language constraints with constraints of world and society°.

221. Such a model could plainly not have the format of a °serial relay of ‘black boxes’° we can’t see inside, where each box does just takes the ‘input’, does its own task, and sends its final ‘output’ to the next (Fig. III.38a) (cf. VII.77). Instead,  the  model  would  have  a  °parallel  interactive  format°  of  processing domains that run concurrently and share their ongoing provisional data (Fig. III.38b). 

At least two  main  types of executive  operations  would oversee the evolution of the model’s design. Scheduling would set up workable sequences of processes in real time and correlate them with the °pacing of the ‘surface sequence’°. Complementarily, packaging would arrange processes in interactive arrays or ‘programs’ to suit current conditions and resource demands, ‘larger packages’ being for a °lighter loading (e.g., complex sentences  for  a  simpletopic) and ‘smaller packages’ for a heavier loading (e.g., simple sentences for a complex topic)° (cf. I.21). These two operation types can  distribute processing among many local sites rather than working at just one site; and °parallel distributed processing° can produce some vital °emergent properties°, such as the capacity to simultaneously apply many partially fitting constraints.

222. Some schedules and packages might be °unmarked routines°, e.g.: try to process each word as far as you can for all domains while you are reading it on the page. Others could be specified on demand, e.g.: resolve the contradiction between this argument and one you stated before. The routines would tend to dominate ordinary communication, with a high proportion of unmarked options and commonplace collocations, e.g., when a teacher is giving a standard explanation to young children. For specialized tasks, in contrast, operations could be °rescheduled and unpacked° to be done more slowly and individually without incurring °overload°, e.g., when a science writer tries to state an original theory of communication while defining the technical terms clearly and using them consistently (it’s harder than you’d think!). You might just write down a few phrases, sentences, or paragraphs and then critique and rephrase them, trying to anticipate the problems for the readers, who won’t already have the ideas organized in their minds as well as you do. Or you might not even have them organized yourself and might be waiting to see where they lead — the ‘brainstorming’ that °classical scientific discourse° conceals from public view (cf. VII.63).

  223. Operationally, each processing domain would be charged with organizing, distributing, and integrating operations and storage and with allotting control and resources while the discourse proceeds. Each domain would be functionally rather than temporally or locally unified: its operations would not run all at once at a single site and then relay complete and final output to the next in line but would get °scheduled and packaged° whenever they get activated. Still, one domain would be dominant when it is allotted significantly more resources than the others, especially when a problem arises, e.g., when a speaker self-consciously seeks a ‘polished style’ to elicit solidarity from an elite audience or to distract attention from the banality of the content (cf. II.9

224. A process model could arrange its domains by their °processing depth°, a term matching the popular division between ‘deep’ versus ‘surface structure’. A plausible graphic might be Fig. 

III.39. The ‘shallower’ domains shown toward the top organize local ‘micro-units’ and have a °harder coupling to the linear sequence (or ‘surface text’)°, whereas the ‘deeper’ ones shown toward the bottom organize global ‘macro-units’ have a °softer coupling to the linear sequence (more toward the ‘deep structure’)°.

225. Goal-planning (GP) would be the ‘deepest’ and most °pragmatic° domain and subsumes the agenda of the discourse with the plans and goals of the participants, e.g., getting a job. The participants weigh the alternative histories they anticipate and set up plans to make their own goal more probable, consisting of discourse actions such as utterances, discourse moves such as managing the situation, and discourse strategies such as being cooperative. Since a probability of not attaining a goal creates a problem by definition (III.76, 213.9), this domain is most clearly analogous to °problem-solving° (cf. III.49), although the analogy may also be useful for all domains that seek to establish and maintain connections in an activated network (cf. III.234). Three planning strategies are commonly recognized in computational research on problem-solving (cf. VII.6). In breadth-first search, you evaluate all the alternative paths before choosing one (Fig. III.40a), e.g., considering the various jobs you’re qualified for. In depth-first search, you pursue just one in a rush toward the goal (Fig. III.40b), e.g., picking one job and putting all your efforts into getting it. In means-end analysis, you contrast the current state with the goal state and perform actions to reduce the differences (Fig. III.40c), e.g., improving your qualifications until they match the job you want. Breadth is safer but demands more time and data, whereas depth needs less time and data but is riskier. Means-end demands a well-fitting description of  both  current state  and  goal  state.

.  A bit like  the  °alternative histories in quantum reality° (III.12), the more probable paths ‘interact’ in the sense that what you do is meaningful in respect to what you don’t do but might do. Highly improbable paths are largely ignored (or °summed over°) and, if they unpredictably happen anyway, they catch the planner unprepared.

     226. Goal-planning is closely attuned to situationality by correlating actions with the scenario or situation model constructed by participants to represent the current situation and its expected or desired evolution. This scenario orients the participants in choosing and performing their discoursal moves. You can monitor the evolution of the situation (e.g., watching whether your listener seems disposed to help you) and manage it by steering it toward your plans and goals (e.g., dropping hints that you need help) (cf. V.57). Also, you can invoke content by presenting it as already known (e.g., that your listeners ‘love Harvard beets’), or you can inform by presenting content as not yet known (e.g., that the manufacturer ‘treats the beets ever so royally’) (cf. I.21, 27f). Or your can strategically merge these moves. You can seem to monitor when you are managing, e.g., seem to say what is going on when you are saying what you want to happen. Also, you can seem to invoke when you are informing, e.g., to suggest that what you’re saying is well-known or that your listeners are highly knowledgeable (even when they aren’t). The lexicogrammar of discourse offers subtle means for doing all this.

227. Goal-planning also takes account of the roles of the participants, i.e., their status as a type of agent with characteristic activities, e.g., whether they are colleagues in your business or members of your family (cf. VII.2). Roles which are clearly unequal in status support power, whereas roles which are fairly equal support solidarity (I.7ff; VII.1). These roles and relations strongly control whose plans and goals get priority, and who is entitled to perform discoursal moves in certain ways, e.g., to state what is ‘true’ or to legitimize what is ‘warranted’ (II.71, 94, 119f; III.24). Power encourages people to solve problems and seek goals through confrontation, whereas solidarity encourages them to do so through cooperation. Here, we might add a °social enrichment° to our concept of progress defined so far as an °evolution that softens the coupling, enriches constraints, and moves from resistive toward supportive design° (cf. III.43; VII.7); we can further stipulate that it also °evolves from confrontation among agents (resistive complexity in society) toward cooperation among agents (supportive complexity in society)° (VII.7). Regress does just the reverse, so that you solve your problems and seek your goals by blocking other people’s. Whether modernization will be dominated by progress or by regress is the burning question in most societies today (cf. VII.24f).

228. The agenda of plans and goals enables participants to rate discourse actions as appropriate to the situation, effective in supporting goals, and efficient in operating with moderate resources. Discourse strategies are set for the other processing domains about what to say and how, usually in respect to a style, text type, and discourse domain (or some combination of them), e.g., to sound self-confident in an institutional scenario like a board meeting. You may need extensive planning for costly goals and serious problems, but may also need to conceal your planning by appearing to talk spontaneously lest people get suspicious about being manipulated. When goals dominate over content, the discourse counts as political in a broad sense (I.8); the discourse of professional ‘politicians’ is an obtrusive special case many people rightly mistrust (cf. VIII.43).

229. Topicalizing (TO) would be the ‘next-deepest’ and more °semantic° domain that co-ordinates plans and goals with the activation of global content. A topic (in some sources also called ‘theme’, ‘main idea’, ‘macro-structure’, or ‘superordinate proposition’) acts as a control center to steer the evolution of the text-world model, i.e., the total array of knowledge entailed in processing the text (II.97; III.136). The topic flow controls the conceptual evolution and periodically undergoes a topic shift.

230. One major discourse strategy is to explicitly announce the topic, e.g., in the ‘topic sentence’ traditionally expected to come at or near the start of a written paragraph; another is to leave the topic to be inferred from thematic content of the individual clauses and sentences; still another is to planfully conceal the topic behind some other diversionary topic, e.g., in an advertisement that pretends merely to inform you about cooking (cf. I.21). A topic that seems too unfamiliar or complex to be foregrounded right away can be postponed until an adequate background has been presented — a helpful strategy in °discourse for special purposes° (‘DSP’). Also, in cultural settings where goals are handled indirectly, a close, obtrusive connection between goal and topic should be avoided as self-serving, so you ‘beat around the bush’.

231. Concept-elaborating (CE) would be the semantic domain for enriching and integrating the central concepts within the °text-world-model°. This domain could operate both in a more °global, top-down mode by expanding the connections among concepts associated with the topic, and in a more local, bottom-up mode by activating concepts from which the topic emerges via critical mass° (cf. III.142). Both modes of operation could run simultaneously in °parallel distributed networks that converge to produce the coherence° of the text (cf. II.102; III.221). 

232. Lexicalizing (LX) would be the °lexical° domain subsuming all operations for ‘lexicalizing meanings’ by interfacing the elaborated array of topics and concepts with the language expressions of words and collocations. Operationally, an ‘expression’ is a cue that instructs how to process conceptual content, whereas content is a request that instructs how to search for such a cue (cf. III.203). If the association is fairly strong and direct, expressions would °automatically come to mind with the concept and vice-versa via spreading activation°, a cheap mode of breadth-first search that activates adjacent connections without any controls (III.246, 249). Weaker and more remote associations might need a directed °problem-solving search° specified at first only by the general parameters that candidates should fit; the activated items could then be more specifically weighed and compared. Here too, the speed and efficiency of cognitive processes has both good and bad sides: operations are easy but they encourage you to rely on uncreative routines and banal expressions (cf. I.58; III.97) You are then less sensitive to the delicate demands of such factors as °style and text type° (cf. III.129, 228; Ch. V.A-B). 

233. Grammaticalizing (GR) would be the °grammatical° domain subsuming °morphology and syntax° and delegated to ‘grammaticalizing and linearizing meanings’ by phrasing the lexical items and shaping and arranging them within co-texts. The grammar would be a system of processing instructions (as envisioned, say, by Talmy Givón) interacting richly with the lexicon by forming words (e.g. choosing inflections), building phrases, and implementing any grammatical constraints that accompany lexical items or collocations, such as ‘warrant’ (cf. II.68-78). The unified lexicogrammar would be functional in assessing the forms of language as °means for ends°, and also cognitive and social in °integrating constraints of knowledge about world and society°, e.g., about how Processes are typically organized into ‘Agent’, ‘Action’, and ‘Target’ (cf. IV.B.1). This knowledge would be activated together with lexical items and grammatical patterns so as to °interface standing constraints with emergent constraints°. 

234. Sound/letter layout (SL) would be the domain with the °hardest coupling to the detailed phonetic-acoustic or graphic base° — the so-called ‘surface text’. Logically, processing a lexical item or a grammatical pattern should cover the sounds and letters it subsumes. But operationally, executing or recognizing the sounds or letters requires fine-grained automatic action packages stipulating articulatory, scribal, auditory, and visual actions for utterance or inscription in production and for recognition in reception. These packages would consist of °non-deterministic local motor programs whose targets are met within a reasonable goodness of fit°, e.g., when the phonemes are phonetically realized (cf. II.29). The targets are thus not just specific neural or muscular events but grids of spatial coordinates to control the local movements of the speech organs, hand, or eyes, and grids of temporal coordinates to control the order of the movements in real time. This mode of operation is efficien t and well-adapted to process events that fluctuate in their concrete realizations, e.g., in tone, quality, and volume of voice, or in size, shape, and style of print and handwriting, which would be demand great expenditures for deterministic operations (cf. III.58). Indeed, phonetics and optics tell us that any one realization differs from all others and cannot be exactly duplicated or repeated, even when all participants would agree that you are ‘saying exactly the same thing’ (cf. III.119; V.16.5). In return, a margin of error is entailed along both types of coordinates, either confusing similarly articulated or inscribed units or else getting units out of order (cf. III.204f, 237).

235. The °linear principles° proposed back in III.203-12 could apply to all six processing domains and not just to the ‘shallower’ ones. °Pacing, look-back, or look-ahead° could be deployed when handing plans or topics just as well as when uttering a sequence of sounds. An important and costly goal is likely to be paced slowly and cautiously as you look back and ahead to what has happened or will happen; on a smaller scale, you might adopt a slower pace when publicly reading aloud a passage with words you might have trouble pronouncing. The same generality should hold for the less obviously linear operations of °merging, loading, and assigning cores and adjuncts°. The cores versus the adjuncts in the respective domains would typically be, at least for English: in sound/letter layout, the vowels versus the consonants within the syllable; in grammaticalizing, Content Words versus Function Words; in lexicalizing, the more specific expressions that set and maintain the parameters of style and text type versus the more general expressions; in concept-elaborating, the Objects and Events versus Attributes, Values, Circumstances, and so on; in topicalizing, the topic concepts such as major Events or Actions in a story versus the minor ones; and in goal-planning, the higher goals versus the plan steps and intermediary subgoals. Finally, merging and loading could be done whenever alternatives compete and resources must be selectively distributed.

236. Interactive linearity would also support the discourse actions — actions done in and by means of discourse — that are jointly implemented across multiple processing domains. The utterance would be a stretch of text that is realized in real time under on-line constraints ranging from the ‘deep’ plan it serves over to the ‘shallow’ sounds it contains; the inscription would be the corresponding written unit in letters or characters at the ‘shallow end’. The tone group would be a sequence with its own intonation contour and bounded by pauses (IV.210-13). The conversational turn would be a person’s contribution at one point in a multi-party discourse (V.54-59). The discourse move would be a discoursal action that ‘moves’ the interaction a step forward, e.g., by invoking solidarity prior to asking for help. The discourse episode would be a cycle of such moves, e.g., dropping a whole series of hints about your problem until your friend finally offers to help you (cf. V.57ff). In written discourse, the organization of moves and episodes is often indicated by divisions into paragraphs, whose functions also include signalling steps in a plan or stages in a topic flow. Some languages have ‘discourse markers’ for signalling the bounds of moves and episodes in speaking, similar to ‘spoken paragraphs’, e.g., in the narration of a story (cf. IV.155).

237. Exactly when processing goes ‘linear’ in the familiar sense of word after word is hard to determine, even for lexicalizing and grammaticalizing. Experiments indicate that sometimes the lexical items are selected before they get placed inside a grammatical pattern; other times, the pattern is selected before the lexical items to fill it (IV.13). This flexibility, which again indicates the °functional unity of the lexicogrammar° (cf. II.32; III.130; IV.13ff; VII.145), frees either domain from waiting upon the other and allows choices and operations to be done at any opportune moment, depending on the current activations and associations and on the various resource limitations enumerated in III.213. In return, two complementary types of error may happen: the wrong words in the right order, and the right words in the wrong order (cf. III.204f).

238. The process model outlined here has been neutral regarding text production versus text reception. It might be conveniently assumed that text production begins with goal-planning and ends with sound-letter layout, whereas text reception does the reverse — or, in more popular terms, the speaker goes from purpose to meaning to utterance and the hearer goes the other way. But such an assumption only fits the serial relay models we rejected in III.221. In interactive parallel models, the speaker and hearer are each anticipating or reconstructing what the other is doing (cf. III.185, 229, 231). Here, production is partly co-reception, and reception is partly co-production. This sharing of operations would be optimal for °tuning the respective current language versions of speaker and hearer° (I.37; III.107, 140), whose total knowledge of the language necessarily differs.

239. The chief distinction between production versus reception would be in control. The speaker has more control not just over choosing everything from goals and topics to lexical items and phrases but also over steering the resource demands and navigating between °cooperation versus confrontation°. An ominous °trade-off impends: easy production can make for difficult reception, whereas efficien t reception can be costly to prepare for during production°. A common °power move° within social interaction is to weight this trade-off in favor of the producer, especially in bureaucratic or technical discourse, forcing the receivers to work very hard or else get lost (V.80f; VII.219). The producer may have no confrontational intention and may just be insensitive to the audience or unskilled in adapting. Yet whatever the intention, power moves to distribute the work of discourse unequally are at the heart of the °global communication crisis° diagnosed in II.133. If discourse practices are to °progress from power and confrontation toward solidarity and cooperation°, text producers plainly need a better sense than conventional socialization and education provide of how to share the work (cf. VII.265).  

240. A °post-classical process model° that respects the wide range and flexibility of potential operations cannot be expected to enumerate the exact operations that took place during the production and reception of a given text. At most, we can specify plausible processing types and results by taking into account some influential conditions, e.g.: whether the producer was knowledgeable on the topic, sensitive to the audience, creative in handling the style, etc.; and whether the receiver was also knowledgeable, or was attentive, interested, cooperative, etc. A given context of conditions, however detailed and specific, does not exert a °deterministic causality° to produce exactly this one discourse with just these features, but rather °regulates the probabilities, with some features being more predictable°, e.g., the constraints within collocations like ‘evidence + warrant + trial’, and other features being less so, e.g., the pacing of conversational turns. Small changes in the conditions might have unexpectedly large effects on the discourse, e.g., when the attention of the participants is so distracted by abrupt fluctuations in the intensity of noise that severe misunderstandings occur. Still, our model should allow us to assess how far the interaction was °efficient, effective, and appropriate  for producer or receiver°. We can compare the actual discourse with plausible alternatives, e.g., if the discoursal work had been distributed more equally or the willingness to cooperate had been greater. In some settings, we might actively control and alter the conditions and observe the different outcomes.  

241. Our model must also contend with the practical difficulties of °monitoring the highly skilled and rapid self-organization of discourse processing rather than just analyzing the text-artifacts°. We can gather some modes of data by observing writers as they write and revise or by making protocols as they ‘think aloud’ (III.193, 197). But we are limited in what clues we can detect, what people can or want to express, and what can be made conscious at all. Even highly skilled text producers may not succeed in reliably describing what they do; high skills might instead be the hardest to model if they work with a widened range of fluctuation, novelty, and complexity. Lower skills would be easier to model, but the producers might be more reluctant to have them investigated or might try to look better by offering unrepresentative or misleading data. Our model should distinguish among varying degrees of skill and indicate how people might progress by moving up (VII.102, 108). 

242. Our model must also consider that a significant portion of discourse processing may be ‘pre-organized’ in the standing °subcritical baseline of knowledge about language, world, and society prior to the discourse event° (cf. III.52, 80, 87, 95, 136, 175). The move toward the °critical mass of discourse can be activated° by ‘cues’ that are either internal (e.g., an idea coming to mind) or external (e.g., a salient event in the situation), or a combination of both (e.g., a perceived object reminding you of a long-planned goal or a favorite topic). When the discourse event winds down, processing goes subcritical again but continues to operate, and the transactions of the discourse can go on interacting with other drifts in the general knowledge network.  

243. These factors point to the key role of memory in discourse processing. Fig. III.38 back in III.224 shows the memory ‘spans’ generally assumed in cognitive psychology and psycholinguistics to have differing focuses (suggested by dot shadings in the Figure): (a) short-term sensory storage (STSS) lasting for only a second or two and focusing on sounds and letters; (b) short-term memory (STM) lasting some twenty seconds and focusing on lexicogrammar; and (c) long-term memory (LTM) having apparently unlimited duration and focusing on semantic or conceptual meaning and pragmatic goals. Yet these focuses are far from exclusive if the resources of the various domains, like the units of the linguistic levels, are °functionally interconnected as means and ends° (II.60). Typically, the ends would be ‘deeper’ and would persist longer in memory than the means, but the two might sometimes be about the same. Experimental research on reading has found long-term memory receiving some traces of the ‘surface’ text, e.g., verbatim recall of a passage, or a record of whether an item was presented visually or acoustically; reciprocally, immediate perception has been found to have some limited access to the ‘deeper’ domains of concepts, topics, and goals. The mere recognition and reading of words already can be affected by a seemingly instantaneous knowledge of what they mean, as if form and function can be activated together. So the storage of language in the human mind could indeed have a multiple design, operating as a lexicon, a grammar, a lexicogrammar and so on, as the occasion requires (II.79). 

244. Access to the text itself would be in three modes, also shown in Fig. III.39: (a) the retrospective representation of prior text; (b) the perception of current text; and (c) the prospective representation of later text. Since the perception of current text is limited to the brief span of short-term sensory storage holding the acoustic or visual surface text, most processing activities would apply not to the surface text but to its mental representation in short-term and long-term memory, whose format is not just °linear but richly interconnected and hierarchical°, and includes much associated data not expressed in the text. This factor points to hidden problems in discourse processing models which assume continual direct contact to extended stretches of text, especially written ones, and which draw a strict dichotomy between the sparser co-text of the actual wording versus the richer context within the mental representation (cf. II.108). The rapid transition between the co-text and context might have influential effects on the quality of the data, e.g., when the participants believe they heard or read something that wasn’t actually said (III.206; IV.25). 

245. Such effects are to be expected if knowing what a word or lexical item means in the richest sense entails knowing in what collocations and co-texts it typically appears (cf. II.67). Discourse processing would be continually collating co-texts encountered before with the co-text at hand by constructing and applying the various sets of constraints they have in common. Doing so could well foster fuzzy borders around what was actually said on this one occasion. 

246. In recent years, extensive °cognitive research on discourse processing° has yielded a more detailed view of the operations for using the data from co-texts to construct contexts. The individual cues such as sounds or words activate relevant data in storage and operation to organize and integrate patterns of meanings. Spreading activation moves out from an activated node to those connected to it within the active sub-network of data (III.232), exciting relevant connections while inhibiting irrelevant ones, e.g., crime report for ‘warrant’ => ‘police’, ‘issue’, ‘search’, versus ‘warrant’ => ‘warranty’, ‘appliance dealer’. As the network °self-organizes and stabilizes°, further nodes are contacted by inferencing about the context, e.g., ‘police’, ‘issue’, ‘search’ => ‘stolen goods’, ‘fugitive suspects’. Some data are updated when new data imply some changes, e.g., when we read that no ‘goods’ or ‘fugitives’ were found but a stockpile of explosives was. Or, data that have lost their activation can be reinstated for further processing, e.g., to recall why the ‘warrant’ had been issued.  

247. How could these operations exploit the organization of prior knowledge so as to reorganize the data into the current context? The answer suggested by °mainstream linguistics and classical semantics° would be that °complex classes of deterministic rules perform specific searches and formal operations°, e.g., to recover the ‘deep structure’ or the ‘logical form’ of a sentence (II.50). But it’s hard to see how such an expensive machinery of ‘rules’ could work with so little time and effort and could handle an indefinitely wide variety of contexts (cf. II.44ff; III.58, 87). The same problem upsets the notion that a language has a °complete and frozen set of purely linguistic constraints° (II.45), and the °category mistake that logical operations are the privileged model for all thought and language° (cf. II.13; V.124). 

248. A less familiar answer would be that °fairly simple classes of non-deterministic operations construct and regulate active networks through parallel distributed processing among many fairly simple local interactions°. This answer has been developed especially by °transdisciplinary groups of researchers in artificial intelligence and more recently in artificial life°, who faced the problem of writing computer programs to simulate complex processes ranging from the emergence of the basic metabolism up to the operation of human vision and memory (cf. II.100ff). At first, some research again sought to build systems of complex, expensive rules similar to those found in generative linguistics and formal logic, two domains that had not been not conceived for doing actual simulations. But such systems either had to operate in unnaturally sparse and simplified worlds with pre-selected, step-by-step tasks, or else incurred a combinatorial explosion of branches and loops that wouldn’t finish. Human beings, in contrast, easily perform an indefinitely wide variety of tasks, simultaneously using multiple sources of knowledge in rich and complicated worlds. Inspired by the physiology of the human brain and neural system, researchers began building °connectionist simulation models° wherein complex information processing is carried out in networks of highly interconnected elements sending each other very simple excitive or inhibitive ‘messages’ and updating their own activation status from the ‘messages’ they receive (cf. III.56). The °global  processes appear to be emergent properties of the local operations°. Intriguingly, the °convergence of multiple constraints° does not make processing slower and harder, but faster and easier. These models have since been developed for such phenomena as motor control, perception, memory retrieval, and learning, including models of language acquisition and sentence processing that do not use the usual linguistic ‘rules’ at all. So far, the simulations have had impressive success in achieving results quite similar to those that empirical research has discovered in human beings.

249. Another trend in the same direction has recently emerged from unexpected but robust experimental findings on priming in human text reception during reading. A probe item such as a word is held to be primed if its degree of activation in memory is raised above the inactive state, like standing at attention and waiting to be called. Primed items will be consistently recognized and responded to more rapidly than others, e.g., by pressing a key to signal that it either is or is not an English word (a ‘lexical decision task’). Surprisingly, the experiments indicated that when a word is recognized, all its meanings are initially activated, not just the relevant one. Yet after a short time the non-relevant ones are deactivated  , while the relevant ones raise their activation and spread it to further associations. Suppose you are a speaker of American English reading a text on a moving computer display containing this passage:

[58] The townspeople were amazed that all the buildings collapsed except the mint.

The text suddenly halts at ‘mint’, and the display gives you a target item to decide if it’s a real word. For a brief interval up to roughly half a second, your response would probably show priming for both the relevant ‘money’ and the non-relevant ‘candy’, but not for the inferrable ‘earthquake’ (what made the ‘buildings collapse’). Thereafter, the non-relevant item would lose its activation while the relevant and the inferrable items would gain. Evidently, the constraints of co-text and context exert their control during this tiny interval, and a series of cycles applies excitation and inhibition to regulate the strength whereby any one meaning is connected with the rest, e.g., though a shared topic.

250. This important finding led Walter Kintsch and his colleagues to formulate a new °construction-integration model° of human discourse processing. The construction phase runs on °local bottom-up operations that are rather non-deterministic° about what they do and look for, but in return are far faster and cheaper than deterministic rules. The operations need merely make it probable that the relevant word or meaning is among the ones that get activated. Then, the relative strengths of the connections in the pattern get adjusted to excite the relevant connections and inhibit the irrelevant ones, so that a coherent array of meanings emerges from the converging integration phase. This array is like a °schema° that it is not stored and activated as a whole but made to order expressly to suit the ongoing context. Computer simulations have shown that this mode of operation is computationally quite feasible.

251. °Classical science° might find it contradictory that initial memory access should be non-deterministic and non-selective and yet produce determinate and precise results. It seems more ‘logical’ for the precision to be there from start to finish. But we can now see some advantages for the flexibility and evolution needed to handle the enormous range of knowledge entailed in discourse, including ideas you never thought of communicating before. We also see an operational base whereby °transitory activation patterns° could support the continually evolving design of a discourse participant’s °currently active version of the language°. The non-selective initial activation would make these versions briefly carry along some irrelevant baggage, but without much expenditure or distraction; branching alternatives come and go so rapidly as to incur no heavy resource loads, and get merged quickly and cheaply.

252. The lexical, grammatical, semantic, or pragmatic constraints discovered in large data corpuses could help control the operations by raising the activation levels of the words and meanings that are likely to occur together. If you read, say, the collocation ‘enough evidence to warrant’, the items ‘investigation’ or ‘trial’ are likely to be primed (II.72, 76). An interesting question for future research is how far grammatical categories like ‘Noun’ and ‘Verb’ or semantic concepts like ‘knowledge-gathering activities’ can be primed independently of the lexical items that instantiate them, and at what degree of delicacy, e.g., Count Noun (like ‘trial’) versus Mass Noun (like ‘evidence’) for grammar, or ‘investigation’ versus ‘retaliation’ for concepts. We might find differing strengths of activation corres-ponding to the relative typicality and probability of the collocation, e.g., ‘warrant + trial’ being stronger than ‘warrant + expenditure’.[=]McKoon

253. The °construction-integration model° is a distinguished case of a major theoretical revision driven by empirical data to buck the scientific trends, namely the trends drawing a strict dichotomy between the °bottom-up language-centered view of psycholinguistics versus the top-down schema-centered or script-centered view in cognitive psychology and artificial intelligence° (cf. II.96f, 100f). Each view had claimed that the other type of process applies only when the type it emphasized runs into difficulties, as when bottom-up operations consult the topic to resolve indeterminacies, or top-down operations do special word-scans to revise schema-based expectations. Yet tracking of eye fixations indicated that even fluent readers densely sample the words in a text. And readers must often process content in contexts where no stored schema may be applicable because the topic is unfamiliar, mixed, or shifting, e.g., at the beginning of a discourse (II.97).

254. We now see this divergence of scientific trends to be a genuine ‘Hegelian dialectic’ of ‘thesis’ and ‘antithesis’, for which the findings just cited suggest a convergence into a welcome ‘synthesis’. Network evolution makes no sharp dichotomy between the more local and externally controlled bottom-up operations versus the more global and internally controlled top-down ones, but balances both types. Only special cases would tilt the balance, e.g., when you read a highly unfamiliar technical term and you have to work hard to activate some appropriate schema.

255. A next step in research might be to investigate how the processing of real discourse (and not just isolated invented sentences) might be modeled in terms of the °self-organizing systems discovered by complexity theory and artificial life°. Their general principles, such as the °amplification of data fields on a material base° and the °spontaneous emergence of order near the boundary of chaos°, are plausibly implicated in the emergence and evolution of life-systems and of natural languages. °Parallel distributed processing in a sufficiently amplified system° might supply the °operational implementation for cognition and communication in society to run in rapid timing and on small expenditures of matter and energy°. The text-event might be modeled as an °evolving system of cascading activated knowledge networks°, whose design takes on the degrees of complexity, determinacy, and so on, that suit the context. The °currently active version of the language plus the activated knowledge of world and society° would be richly supportive of both the ‘shallower’ and the ‘deeper’ processing domains described in III.222ff. The two poles of °virtual system versus actual system° — or ‘langue’ versus ‘parole’, or ‘competence’ versus ‘performance’ — are °dialectically reconciled as the ongoing co-text is steered by the constraints from previously encountered co-texts sharing comparable collocations°, such as those displayed in large corpuses (III.244ff). The more typical collocations can supply ‘prototypes’ for constructing less typical ones.

256. Shared collocations in co-texts would also support the °convergence between the discourse processes of speaker and hearer° (or writer and reader) and their °consensus about what it means and how it bears upon their scenario of the situation°. Two of the main criteria proposed in II.28 as °test scales for linguistic theories and descriptions° would thus be achieved within the domain of investigation itself. We now appreciate that discourse participants pass the tests because they do not operate under the postulate that °language by itself is disconnected apart from the constraints of world and society°. Our task is to provide theories and descriptions of how they do operate so successfully, and how high their convergence and consensus typically are or need to be (III.137). The findings that led to Kintsch’s construction-integration model indicate that the self-organizing operations performed in parallel by speaker and hearer deploy shared design principles with a reasonable probability of constructing similar networks which get tuned by means of the co-text itself. The same might hold for investigators of text and discourse in our roles as prospective participants insofar as we work under natural conditions and with authentic discourse data. But certain factors may get shifted when we work as theorists, describers, and model-builders; and our theories and models need to indicate how we can retain enough control to support our own convergence and consensus about models of text and discourse.

257. In this section, I have suggested how a °transdisciplinary science of text and discourse° might design an interactive model of discourse processing to be compatible both with extensive empirical evidence and with general principles of design, control, and operation in richly amplified systems. We do not start from the reassuring °classical assumptions that language is already given as a uniform, stable, and self-sufficient formal system, and that its main function is to make true statements about reality°. Instead, we attempt to derive language from °general principles of evolvability in self-organizing dynamical systems°, and we see its °main function in constructing and negotiating models of world and society°. This approach situates our °post-classical research program for a science of text and discourse° at a transdisciplinary crossroads. Its scientific validity will depend on its °goodness of fit with a rich range of empirical data°. Its ecological validity will depend on its ability to support an °ecologist program for enhancing human freedom of access to knowledge and society through discourse°. We naturally hope that these two modes of validity can be reconciled instead of, as in °classical methods, trading off the ecological to secure the scientific° (I.3; III.1, 192; V.47). But we also recognize that their reconciliation will require concentrated attention to the conditions of our own enterprise in a rapidly evolving modernized world and society whose design may well drift out of control unless people adopt more °progressive discourse strategies° (II.133).  

Commentary to III.A (III.1-47)

¶ III.1f On »classical« vs. »post-classical science«, see Note to III.11. »trans-disciplinary«: see Note to I.5. »ecologism«: see Note to I.8; »failure to connect«: see Note to I.4. »scientism«: see Note to I.3. ¶ III.3 »inclusive theory versus exclusive practice«: see Note to I.6. ¶ III.4 Unlike the term ‘action’, the term »move« usefully suggests changing things and places while following conventions and often strategically deliberating, as in a game. On science as a »cognitive enterprise«, compare Oeser (1972). »modeling style«: see also III.16, 24, 146, 149, 167, 185, 191; VII.104. For the purposes of my discussion, I leave religion aside. The main religions are non-Western in origin, and combine non-realism and introspection with a concept of ‘God’ being both universal observer (as Ken Pike would say) and ultimate determiner of causes and effects. The universe can thus be completely ordered even though humans cannot fathom its order — a prospect some people (not including me) find reassuring. On »bottom-up« vs. »top-down«: see Note to II.96. ¶ III.5 »realism«: cf. II.13; III.7, 10ff, 16ff, 23f, 28, 41, 69, 84f, 90, 98ff, 128, 144, 146, 158f, 165, 179, 191; IV.18, 72, 75, 96, 125, 140; V.56, 98, 124; VI.80; VII.63, 115, 141, 290ff, 298ff, 336ff; VIII.22, 46, 114(d), 134: on its role in the ‘authority of science’, see Feyerabend (1978); Oeser & Bonet (eds.) (1988). »idealism«: cf. II.13, 121; III.10, 17, 85, 87, 94, 128, 146, 165, 191, 219; VIII.16, 20, 43 114(d). »mechanism«: cf. III.6f, 11f, 159, 169, 179. ¶ III.6 The term »classical reality« was popularized as the counterpoint to ‘quantum reality’, but I shall use it the sense defined here; the same holds for ‘classical science’ (Note to III.11), popularized mainly by the ‘new physics’ (as in the quote in III.12). Contrasting terms are ‘quasi-classical’ (cf. Note to III.12), ‘non-classical’ (III.14, 46, 68, 142, 165, 174; IV.8, 117; VIII.135), and, programmatically, ‘post-classical’ (II.55f, 79; III.4f, 16, 19, 24f, 32f, 44ff, 69, 74, 145f, 148f, 152f, 174, 181, 185, 187, 191, 193, 216, 240, 257; IV.30, 77, 133; V.55, 90, 93, 101, 121, 123f; VII.104; VIII.21, 57, 99, 131, 140f, 144). The sense of ‘classical antiquity’ (ancient Greece and Rome) will be subsumed under the neologism ‘ancientism’ in Ch. VII (see Note to VII.58). ¶ III.7 true statements«: see Note to I.32. »paradoxically«: see Note to I.3. ¶ III.9 Hume (1962 [1777]:63) ¶ III.10 »empiricism«: see also III.90, 128; V.124; VIII.46; Feyerabend (1985); »physicalism«: see also II.119; III.17, 146, 179, 185; IV.125; VII.115; VIII.10, 114(d), 134. »positivism« and »verificationism«: see Note to II.13. »unified science«: Neurath, Bohr, Dewey, Russell, Carnap, & Morris (1938); cf. II.119; III.146, 179; IV.125; V.101, 123f; VII.76. [29]: Hume (1962 [1777]:163ff), speaking of ‘the havoc we must make’ in our ‘libraries’ if we limit ourselves to this sole type of ‘object’. [30] Carnap (1966 [1926]); [31] Berkeley (1949 [1734]: 42f); [32] Descartes (1911 [1647]: 443). »organs of sense«: compare Hume (1962 [1777]:62) ‘it seems a proposition, which will not admit of much dispute, that all our ideas are nothing but copies of our impressions’. ¶ III.11 »classical science«: see I.2, 4; III.1f, 4, 12f, 18f, 30, 41, 44, 51, 99, 146, 148f, 151f, 155f, 158, 163, 166, 179, 181, 184, 187, 191ff, 251; IV.55, 76ff, 98, 101, 103f, 121; VII.63f, 103f, 199; VIII.1, 10, 20, 24, 45ff, 104, 134f, 159, 167, and Notes to III.30, 55. The term ‘classical’ was already used in this sense in respect to physics by Niels Bohr (1934 [1925-31]), as a counterpoint to ‘quantum’. Bazerman (1988) explores scientific discourse to retrace the historical emergence of this »program« between 1650 and 1800. »natural sciences«: cf. III.148, 157, 179; »dictionaries«: e.g. Webster’s Seventh New Collegiate Dictionary (1963:563) (cf. Note to III.157). ¶ III.12 Prigogine & Stengers (1984:213). »new physics«: see Davies (ed.) (1989) for a comprehensive survey, and compare III.18f, 22, 51, 148f, 159, 193. »probabilistic ‘quantum-mechanical’ universe«, »correct«: Gell-Mann (1994:24f, 131ff,); on ‘quantum-mechanics, see also III.13ff, 18, 36, 41, 68 174, 225, and Note to III.6. »alternative histories», »summed over«: Gell-Mann (1994), drawing on Richard Feynman (1967); see also III.36f, 142, 225; IV.8. ¶ III.14 The ‘EPR paradox’ of Einstein, Podolsky, & N. Rosen (1935) was formulated for electrons (compare riposte in Bohr 1935); the standard version today was reformulated by David Bohm for photons, whence the term ‘EPRB paradox’; according to Gell-Mann (1994:172), the phenomenon has been widely misrepresented as a claim that ‘the measurement causes a physical effect to propagate from one photon to the other’. J.S. Bell’s (1964) version refuted theories which proposed to circumvent quantum mechanics by postulating ‘hidden variables’ to produce similar effects (see now Shimony 1992:382-89). Some physicists have even speculated that the photons may ‘process information’, and that ‘quantum mechanical processes’ may be ‘associated with consciousness’ (see discussion in Herbert 1985). It would be more apt to say that the photons and processes are interconnected within a field of material and data (III.16). Stapp cited in Zukav (1979:283ff). »faster than light«: Aspect, Dalibard, & Roger (1982); on the notion of ‘superluminal communication’, see Herbert (1982). Heisenberg (1960, 1971) took a lively interest in cognition. »Conjugate variables« allow some events which ought to be impossible, provided they are rare and rapid enough. A particle can briefly ‘borrow’ enough momentum from the ‘uncertainty relation’ to escape from a fixed position, as when alpha particles leave the nucleus of the atom. Moreover, when time and energy are conjugate variables, energy can be ‘borrowed’ to create particles that exist for a tiny instant and are then annihilated. »Ironically«: like paradoxes (Note to II.4), ironies also pervade issues in cognition and communication, and especially education (cf. IV.15; V.116, 124; VII.29, 56f, 86, 115, 135, 147, 192, 290, 298, 333; VIII.26, 57f, 75, 115, 117, 137, and Note to IV.52). ¶ III.15 »rope off the quantum world«: this move was has been attributed to the Copenhagen group led by Niels Bohr (1934#). »projection«: Gell-Mann (1994:25ff) citing (without references) the work of Todd Brun. ¶ III.16 »post-classical«: see Note to III.6. »modeling style«: see Note to III.5. Stapp cited in Zukav (1979:72). On »independent entities« versus »web of relationships«, compare: ‘a competent faculty for thinking’ must dismiss the ‘absurdity’ ‘that one body may act upon another at a distance without the mediation of anything else’ (Newton); and ‘all natural science’ rests upon ‘the belief in an external world independent of the perceiving subject’ (Einstein); versus: ‘the discovery of the universal quantum of action, which expresses a features of wholeness in atomic processes that prevents the distinction between observation of phenomena and independent behavior of the objects’ (Bohr 1958:98); and ‘we had this old idea, that there was a universe out there, and here is man, the observer, safely protected from the universe by a six-inch slab of plate glass; now we learn from the quantum world that even to observe so minuscule an object as an electron we have to shatter that plate glass’, ‘so the old word “observer” simply has to be crossed off the books and we must put in the new word “participator”’ (John Wheeler) (all but Bohr quoted in P. Buckley & Peat 1979). Each »substrate« constitutes a ‘layer’, the one subsuming the material aspects and the other the data aspects, however the two might be mixed or distributed. ¶ III.17 On problems with »dichotomies«, see Note to I.39. ¶ III.18 »built into«: compare III.23, 27, 29, 41, 45, 49, 86, 129, 184. On various »dialectics«: see Note to I.37. On »coarse-grained« vs. »fine-grained«, compare Gell-Mann (1994). ¶ III.19 On the interaction of »observed« and »observer«, see also Maturana (1970); Overton (1984); Peat (1987); W.I. Thompson (1987). ¶ III.20 »top-down« versus »bottom-up«: see Notes to II.96 and III.250, 253. »specifying versus manifesting constraints«: cf. I.17; II.27, 61f, 66, 112; III.71, 85, 135 159, 182, 191(b); V.3, 45. ¶ III.22 »approximation«: see also II.78; III.12f, 185, 188; IV.29; V.6, 30, 105; VII.119, 210, 324, 336; »goodness of fit«: see also I.41; III.40, 46, 151, 185, 192(a), 234. On »freedom« versus »constraints«, see III.32; V.40 and Note to V.40. Maturana & Varela (1987:135). ¶ III.23 »Actualizing«: see Note to III.101. »reification«: see Note to I.12. ¶ III.24 »bracketing«: cf. III.25, 42, 144, 148f, 191; V.90, 121f; VII.104, 217; VIII.141. On the dual moves of »integrating« and »diversifying«, see also III.38, 42.6, 46f, 181, 194, 209. The »discourse« of »classicalizers« has been documented in interesting fieldwork by N. Gilbert & Mulkay (1984). »means« for »ends«: see Note to II.60 ¶ III.27 »built-in«: see Note to III.18: »frozen«: see Notes to II.45 and III.53. Each design parameter has a paired ‘Process’ which can have either an ‘Agent’ (e.g. a perceiver or a scientist) or a ‘Medium’ (e.g. a model or a domain): »fluctuating«, »stabilizing«, »innovating«, »familiarizing«, »complicating«, »simplifying«. One odd pair is left over: ‘determining’ is for an Agent and ‘being deterministic’ for the Medium, whereas the converse could be circumscribed as ‘making indeterminate’ and ‘being non-deterministic’. For a »dynamic« view of the »stability« of »structures«, see Thom (1972). ¶ III.28 »simple matched with complex«: Lenat (1984); Wolfram (1984); Gleick (1987); Langton et al. (eds.) (1992); Waldrop (1992); and compare III.34, 39ff, 46, 49, 51, 58f, 64, 72, 79, 85, 93, 166, 172, 174, 180, 248. ¶ III.30 »reality will prove simple«: Prigogine & Stengers (1984:21) remark, a bit inconsistently with the rest of their book, ‘in the world of classical science, complexity merely veils a fundamental simplicity’. They cite Feynman’s (1967) well-known analogy between nature and chess: ‘the complexity is only apparent; each move follows simple rules’ (1984:44); but chess differs starkly from nature (and from language, pace Saussure) in precisely this respect (see III.41 for a different account). »classicalize«: see III.24, 34, 143f, 149, 167; V.16.1, 40, 46; VII.297f. ¶ III.31 »sparse« vs. »rich«: see Note to I.42. and III.54. »paradox«: see Note to I.3; »enrich«: see Note to III.54. ¶ III.32 »constraints reintroduced«: cf. II.51, 54, 83. ¶ III.34 »projecting to be simpler«: see Spiro, Feltovich, & Coulson (1988). »relying on stereotypes«: compare Schaff (1980). ¶ III.36 »surmised« see Beaugrande (1987a). On »trade-offs«, see Note to I.1. »quasi-classical« and »quantum-mechanical«, »alternative histories«, »summed over«: see Notes to III.6 and III.12. ¶ III.37 On »the Good« as defined in Aristotle’s Art of Rhetoric, see the Lawson-Tancred version, pp. 91ff. ¶ III.38 »diversify and integrate«: see Note to III.24. »multiple intelligences«: see VII.70, 109, 116. ¶ III.40 On »goodness of fit«, see Note to III.22. ¶ III.41 »built into«: see Note to III.18. »lexicogrammar«: see Note to I.19; »delicacy«: see Note to II.63. ¶ III.42 »bracket«: see Note to III.24; »supportive« versus »resistive« are more precise terms than ‘positive’ versus ‘negative’ (in my earlier work) and do not invite confusion with grammatical Polarity (in the sense of IV.46); ¶ III.43 The concepts of »progress« and »regress« will be »enriched« again later on in terms of design (III.49) and of social interaction (III.227); see also III.52. ¶ III.44 »diminishing« vs. »increasing returns«: cf. II.98; III.53, 117, 120, 151. The motto »order out of chaos« was the title of Prigogine & Stengers’ (1984) influential book and has captured the public imagination; compare Schuster (1984) on ‘deterministic chaos’, and I. Stewart (1989) on the ‘mathematics of chaos’; and see also I.39; III.53, 56, 105, 117, 152ff, 174, 181, 255; VII.7f, 86f, 217, 228; Note to VIII.24 ¶ III.45 »formal semantics«: cf. cf. II.48-51 and Note to II.48. »epiphenomena«: see also II.170, 192; V.16.3; VI.63. »grammar of dictionary definitions.«: Beaugrande (1984b). ¶ III.46 On »integrating« the »diverse«, see Note to III.24. »Superstring theory«: see Davies & Brown (eds.) (1988); Peat (1988); Beaugrande (1995b); and compare II.94, 159; V.75; »big bang«: Gribbin (1986); Overbye (1992); Gell-Mann (1994).

Commentary to III.B (III.48-61)

¶ III.49 »enrich our definition of progress«: compare Notes to III.43 and III.54. On »freezing the whole system« and »frozen islands«, see Notes to II.47. ¶ III.50 The term »critical mass«, which in biochemistry designates the conditions for a ‘phase transition’ where a reaction takes off, has been popularized in a broader sense to designate any state or quantity needed to set a process going. The biochemical sense offers a way to interpret the dualism of material and data such that critical mass generates new data that is not the sum of its parts (cf. II.102). Probably because biochemical reactions are often irreversible, the converse term »critical dispersion« has not been popularized, but seems quite useful as a model of data loss under analysis. The term »reduction« has been used in such diverse and disputatious senses that it does not seem preferable here. If a substance is »frozen« (‘quenched’) too rapidly to maintain thermal equilibrium, the result is a class of ‘structurally disordered systems’, called ‘glasses’, of which ordinary window glass is an anomalous case, and which have been actively investigated in complexity theory (cf. Stein 1989; Stein & Palmer 1989). ¶ III.51 »DNA«: cf. III.60, 64, 66f, 74. ¶ III.52 »equilibrium«: cf. II.98; III.44, 98, 151ff, 155, 163; V.109ff, 120; VII.85, and Note to III.50. »evolvability«: due to Kauffman (1990a, 1990b); compare Dawkins (1988); see also III.57, 102, 145, 257. »baseline«: see also II.93; III.80, 87, 95f, 99, 136, 242; »subcritical state«: see also III.80, 87f, 153, 242. ¶ III.53 »self-organizing«: see Ashby (1962); Demetrius (1984); F.E. Yates, Garfinkel, Walter, & G. Yates (eds.) (1987); Langton (ed.) (1989); Kauffman (1990b); Langton et al. (eds.) (1992); and compare the notion of ‘autopoiesis’ in Maturana & Varela (eds.) (1980). »motto«: Waldrop (1992:17) »frozen accidents create regularities«: Gell-Mann (1994:134); see also I.38; III.61, 115f, 152, 162; IV.143; VI.3, 5, 53. ¶ III.54 On »enriching a sparse model«, cf. III.48ff, 69, 102f, 105, 135ff, 157-64, 180, 257, and Notes to III.43 and III.71. »Cellular automaton«: the idea came from Stanis½av Ulam and John von Neumann, but did not receive wide attention until high-powered computers appeared; compare early sources like Codd (1968) and Burks (1970) with recent ones like Wolfram (1986) or Langton (1992b); summary in Waldrop (1992:87, 219ff). »complex behavior«, »mimic«: Wolfram (1984: 194, 198). ¶ III.55 »classes«: I give partly adapted user-friendly names to the ‘rules’ in Table III.2, instead of just calling them ‘I-IV’. »dynamical systems theory« is ‘the study of set of numerical variables (e.g. activation levels) that evolve in time in parallel and interact through differential equations; the classical theory of dynamical systems includes the study of natural physical systems (e.g. mathematical physics) and artificially designed systems (e.g. control theory)’ (Smolensky 1989:195). ¶ III.56 »connectionism«: see Note to II.102. »simulated animals«: Dawkins (1988); Packard (1988); R. Collins & D. Jefferson (1992). ¶ III.57 »complex adaptive system«: see now Gell-Mann (1994). ¶ III.58 »cheap order«: Toffoli (1990); cf. III.52, 80, 132, 152, 232, 250f. ¶ III.59 »primordial soup«: Artificial life explores ‘systems that spontaneously generate life-like self-reproducing processes in a simulated chemical soup’ (Taylor 1992:28) compare Farmer, Kauffman, Packard & Perelson 1987). »catalysis«: cf. III.160, 176. »autocatalytic set«: Kauffman (1990a, 1990b; Bagley & Farmer 1992). ¶ III.60 »Self-reproducing«: even without computers (they were just beginning, thanks largely to his own work), von Neumann (1966) managed to prove mathematically that there exists at least one automaton capable of reproducing itself (cf. Waldrop 1992:220).

Commentary to III.C (III.62-103)

¶ III.62 The »maxim«, attributed to Karl Ernst von Baer and Ernst Haeckel, has remained problematic, chiefly because biologists focused much more on evolution of the species than on the development of the individual (Bonner 1988:15; cf. Garstang 1922? in UF: 1897 Walter. Natural History. NY: Appleton.# Waddington 1940, in UF 1939 Introduction to Modern Genetics London: Allen & Unwin1957; de Beer, gavin 1958 first 1940. Embryos and Ancerstors. Oxford: Clarendon). »deliberate adaptive changes«: Lamarck’s (1809) notion that adaptation might function this way (e.g., giraffes developing long necks in order to graze on tall trees) is still widely held by common sense (cf. Note to III.76 and Gell-Mann 1994:118). ¶ III.63 »dialectic between virtual versus actual«: see Note to I.17. »genetic engineering«: see Bishop & Waldholz (1990); Langton et al. (eds.) (1992). »genome«: cf. III.66f, 77, 161, 176f. ¶ III.64 »DNA«: see Note to III.51. »double helix«: see Gribbin’s (1987) popular account; I follow here Waldholz & Bishop (1990:23ff). »blanks«: Weinberg (1987) compared a gene to ‘a small archi-pelago of information amid a vast sea of drivel’: ‘much of this material seems to represent evolutionary detritus, dead ends, side alleys, and parasitic sequences’ (quoted in Waldholz & Bishop 1990:221). ¶ III.65 »geniculate cortex«: Peat (1987:59f). ¶ III.66 »analogies to language«: quoted from Bishop & Waldholz (1990:16, 74, 23f); compare also Hoffmeyer & Emmeche (1991). On »amino acids« as ‘chemical signals’, see Iverson 1984). »proofreading«: Hopfield (1974). ¶ III.67 The term »nanotechnology« refers to the small sizes of cells (review in Schneiker 1988), a ‘nanometer’ being one billionth of a meter; »retroviruses«: see Ettinger (1972); »death«: Ettinger (1964); G.R. Taylor (1968). »quasi-classical« and »quantum mechanics«: see Notes to III.6 and III.12. ¶ III.69 On »enriching a sparse model«, see Note to III.54. The concepts of »state«, »event«, »process«, and »participants« will be central in the lexicogrammar proposed in Ch. IV. »structure«: but compare the sparser sense in structuralism (II.114); both ‘structure’ and ‘function’ have been unstable terms (II.81). »means« for »ends«: see Note to II.60 . ¶ III.70 »critical mass«: see Note to III.50. »emergent properties«: see Note to II.102. ¶ III.71 »system«: this core definition covers the uses of the term in such fields as sociology, cybernetics, factor analysis, information theory, game theory, thermodynamics, and mathematical topology (Prewo et al. 1973:12f.) (cf. II.98 and Note to II.98). »function«: the concept of a relation between means versus ends (II.59ff) is an enrichment of this and will be retained. »top-down« versus »bottom-up«: see Notes to II.96 and III.250, 253. ¶ III.72 »division of labor«: compare Bonner (1988:98f): ‘complexity gives a division of labor’, and this ‘interaction of parts is most characteristic of complex biological systems’; or: ‘one can equate the subdivision of the innards of the body with a division of labor’. »multi-system«: see Note to I.18. ¶ III.73 »storage« versus »operations«: compare the requirement of ‘rate-independent’ representation in Pattee (1977). The division between »token versus type« has been variously used in linguistics to construct ratios between language by itself versus language use, or else between category/class versus member. ¶ III.74 »DNA«: see Note to III.51. »neuron groups«: see Changeux (1985). ¶ III.75 »informational closure«: e.g. in Varela (1979); Maturana & Varela (eds.) (1980). »solipsism«: Efran et al. (1990:76). ¶ III.76 »internal model«: compare Pattee (1977). »sum of parts«: see Notes to II.102 and III.50. ¶ III.78 The term »emergent model« has been expounded by John Holland for default hierarchies with weak general rules (see Holland, Holyoak, Nisbett, & Thagard 1986). ¶ III.79 »subcritical«: see Note to III.52. »receptor cells«: Bonner (1988:193). »retina«: Moravec (1988:187ff). »simple entities« versus »complex data«: see Notes to III.28 and III.30. ¶ III.80 »baseline«: see Note to III.52. ¶ III.81 »decisive advance«: compare Pattee’s (1977, 1979a, b) contrast between ‘dynamic’ versus ‘linguistic’ domains, where a ‘language’ is ‘any system containing a repertory of signs, syntactic-semantic rules, and a read-write mechanism’. »adaptive meanings«: see also I.59f; II.4f, 48; III.83; VII.18; VIII.4, 22, 43, 48, 140 146, 156, 159, 162. »adaptive action space«: see also II.91, 121; V.40; VII.5f, 111, 119; VIII.22, 42, 131, 141, 146. M.A.K. Halliday (1985, 1994a). On ‘the notion of »adaptive value« in macromolecular systems’, see Demetrius (1984). ¶ III.83 »misunderstanding«: contrast Schegloff (1987a); Blum-Kulka & Weizman (1988) suggest that it is ‘inevitable’. »maladaptive meaning«: see also I.59f; II.5; VII.18; VIII.4, 31, 48, 129, 132, 143, 146, 147, 156, 159, 162. ¶ III.85 »realism« vs. »idealism«: see Notes to III.5; »cognitive revolution«: see Note to II.97. »receptor and signal molecules«: Bonner (1988:195ff). On the role of ‘transmitter release’ in the ‘molecular biology of memory’, see Kandel & Schwartz (1982). ¶ III.86 »automatic« versus »attention«: see also III.213.2, 215, 217ff, 232, 234, 240; VI.16, 22; VII.326, 331, 333, and Note to III.212ff. ¶ III.87 »frozen system«: see Notes to II.45, 47. »baseline«: see Note to III.52. »script«: Schank & Abelson (1977). ¶ III.91 »means« for »ends«: see Note to II.60. ¶ III.92 On parallel »distributed processing« see Note to III.102. ¶ III.93 Processing language on the plane of »subsymbols« has been impressively simulated by the Parallel Distributed Processing Research Group (cf. Smolensky 1989:260f); compare III.222f, 231, 248, 255. A major advantage is that learning can generate schemas very gradually, which suggests an explanation for the origin of schemas (cf. III.250, 253f); compare also Bartlett (1932); Kintsch (1977); Rumelhart (1980); Denhière (1982); Alba & Hasher (1983); R. Anderson & Pearson (1984); Brewer & Nakamura (1984). ¶ III.94 »electron«: cf. Gell-Mann (1994:123f). ¶ III.95 Crick & Mitchison (1983). ¶ III.97 »ecologist«: see Note to I.8. ¶ III.98f »beliefs« and »attitudes« have both been overloaded terms, and the boundary between them unsettled, the former investigated more by philosophers (e.g. Sticht 1983; L.R. Baker 1988, 1993; R. Warner 1993) and the latter more by behaviorist psychologists (e.g. papers in Himmelfarb & Eagley [eds.] 1974); but contrast Barwise & Perry (1983). Both sides have been troubled by the prospect that a given content may be somewhere in between believed or disbelieved, valued or devalued. On »problems« with »input«: see also III.75, 81, 84, 90.

Commentary to III.D (III.104-45)

¶ III.104 »origin of language«: Firth’s (1964 [1930-37]:25f, 141f) jovial sketch of the ‘theories of the origin of speech’ with quaint ‘reduplicative nicknames’ (‘bow-wow’, ‘pooh-pooh’, ‘ding-dong’, ‘yo-he-ho’, ‘ta-ta’) is symptomatic of the disinterest of linguists in questions of origins, witness their fateful resolve to study language as a static (‘synchronic’) system (cf. II.24). ¶ III.106 »product of social interaction«: the social origins of language and cognition established outside the individual before they are internalized has been a key theme in Soviet research, e.g. Volo inov (1973 [1929]), Vygotsky (1930, 1962 [ca. 1930]; cf. Wertsch 1985), Leon'tev (1971 [1969]), and Luria (1973, 1981) and offers a useful counterpart to the atomistic viewpoint of the West exalting the individual (I.7, 10ff, 39). ¶ III.107 »low energy and rich data«: cf. I.42; III.58, 92, 107. »dichotomy«: see Note to I.39. ¶ III.108 »Talking about things«: see Pike (1993). On the ‘linguistic coding of epistemology’, see Chafe & Nichols (eds.) (1986). ¶ III.110 »vocal tract«: Lieberman (1992); »dental«: Greenberg, Turner, & Zegura (1986). »stones«: Schmandt-Besserat (1978). ¶ III.111 »proto-language«: cf. III.117, 130-33, 145; M.A.K. Halliday (1985:xvii; 1973:97) notes that the proto-language used by human children suggests how ‘language evolved in the human species’ from ‘an early stage’ ‘without any grammar’, the ‘meanings’ being ‘expressed through rather simple structures whose elements derive directly from the functions’. »natural language«: see Note to II.12; »lexicogrammar«: see Note to I.19. »inflation« coined by Alan Guth (1988); compare references in Note to III.46. ¶ III.112 »commands«: see IV.38. ¶ III.113 Saussure (1966 [1916]:90, 81, 99f, 87, 152ff, 161, 89, 169); compare II.24. ¶ III.114 »specifying versus manifesting constraints«: see Note to III.20. ¶ III.115 »arbitrary«: see Notes to II.48 and IV.14. »limit«, »in spite of evolution«: Saussure (1966:133ff, 87). The ‘symbol’, the most ‘arbitrary’ type of ‘sign’, has recently been critiqued in view of broader notions of the ‘iconic’ and ‘indexical’ (cf. II.90; (Dressler #  Widdowson 1984b). »frozen accidents«: see Note to III.53 ¶ III.116 »frozen islands«: see Note to II.47. »written standard«: see VII.62, 81, 162, 224, 232. ¶ III.117 »increasing returns«: see Note to III.44. Compare Sapir (1921:119): ‘no language wholly fails to distinguish noun and verb«’, whereas ‘no other parts of speech’ are ‘imperatively required for the life of language’. »ends« and »means«: see Note to II.60. ¶ III.120 »target«: see II.62; III.234. ¶ III.121 »finicky distinctions«: see IV.52, 92; VII.162. ¶ III.122 Evidently, »whom«: survives best right after a preposition, e.g., ‘Firth, of whom we all have heard’ vs. ‘Firth, who/whom we all have heard of’. »Northumbrian«: I am indebted to Nikolaus Ritt for this example. ¶ III.124 »French-based names«: Wandruszka (1976:19f). ¶ III.125 »maternity/paternity suit«: I am indebted to Charles Fillmore for this example. ¶ III.128 »acquire«: For mentalists, ‘acquisition’ is spontaneously learning one’s native language, whereas ‘learning’ is a deliberate process, e.g. in a classroom (cf. VI.130-34, 326-31). »instantaneous model«: Chomsky (1965:36). ¶ III.129 »linguistic universals«: cf. Greenberg, Ferguson, & Moravcsik (eds.) (1963); Bach & Harms (eds.) (1968); Hoenigswald (1978); Hopper & Thompson (1984); Comrie (1989); Greenberg (1992); Haberlandt & Heltoft (1993). ¶ III.130 M.A.K. Halliday (1994a:xviii). ¶ III.141 »statistical analysis«: e.g. Miles (1967). »spoken usage«: M.A.K. Halliday (1995) cites The Chinese Secret History of the Mongols, a rare bilingual text intended to help the Chinese learn Mongolian, and therefore recording ordinary Chinese usage of the period (cf. Halliday 1959).

Commentary to III.E (III.146-82)

¶ III.147 »scientific revolution«: see Note to II.2. ¶ III.149 modeling style«: see Note to III.5. »critical mass/dispersion«: see Note to III.50. »emergent properties«: see Note to II.102. ¶ III.151 »classical reality«: see Note to III.6; »goodness of fit«: see Note to III.22; »diminishing« vs. »increasing returns«: see Notes to III.44. ¶ III.152 »boundary between order and chaos«: see also III.153, 174, 255; VII.7f, 86f, 228; Note to VIII.24; compare Note on »order out of chaos« in III.44; »patterns in chance«: I. Stewart (1987:131). »resists being frozen«: see Note to II.45. ¶ III.153 On ‘»creative« processes in scientific discovery’, see especially Langley, Bradshaw, Simon, & Zytkow (1987). »renormalized«: my use is informal, versus the use in physics for the cancellation of coordinated infinities. ¶ III.155 On the »conservatism« of science, see M. Harris (1980:284f). ¶ III.156 On »bottom up vs. top down« see Notes to II.96 and III.250, 253. »dialectic«: see Note to I.37. »paradox«: see Note to I.3; »accidental«: compare Note to III.53 on ‘frozen accidents’. ¶ III.157 »global movement between sparser versus richer«: I am deeply indebted to Gene Yates and Peter Kugler and their colleagues at the UCLA Institute of Medical Engineering for lively discussions of the prospects (compare the ideas in F.E. Yates 1982; F.E. Yates & Kugler 1984; Reed, Kugler, & Shaw 1985; F.E. Yates, Garfinkel, Walter, & G. Yates [eds.] 1987); and see already Lotka (1925), Bohr (1934), Pepper (1942), Van der Steen (1970), and Pattee’s (1979a, b) advocacy of ‘complementarity’ instead of ‘reduction’. »equated«: one definition of ‘science’ in Webster’s Dictionary (p. 771) is simply ‘natural science’ (cf. Note to III.11). ¶ III.158 »mathematics exploits«: see P. Davis & Hersh (1981); on ‘mathematical models’, see Edmundson (1967); on ‘mathematics in a ‘cognitive’ perspective, see Nesher (1986). »reducing«: compare B.F. Skinner’s boast that ‘behaviorism’ yields ‘results that may be formulated in centimetres, grams, and seconds’ (in Psychology Today, 16/5, 1982, p. 48). In contrast, Pattee (1982:335f) remarks that ‘measurement is a very restricted form of perception’ for ‘mapping from a physical system to a symbol’; ‘the essence of this mapping is the high selectivity or simplification of the system to the one attribute we have chosen to observe’. ¶ III.162 On factors correlating with »size«, see Bonner (1988). ¶ III.163 »Notable instances«: cf. III. 163, 179; V.116; VII.76ff. ¶ III.164 »trade-offs«: cf. R. Rosen (1986) and Note to I.1. ¶ III.166 Hjelmslev (1969 [1943]:79f, 100f), citing Hilbert, Carnap, and Tarski. Bloomfield (1933:33). Chom-sky (1957:48). Firth (1957 [1934-51]:225). ¶ III.167 »exploited«: historically, this tactic emerged in philosophy before linguistics, e.g. Frege (1892), following a venerable heritage in projects for universal systems of knowledge, e.g. Descartes and Leibnitz. The most radical proposal I know of was advanced by Hjelmslev (1975 [1941-42]) in the 1940s but not put to use (cf. IV.167; V.96). On logical analysis versus discourse analysis, see Charolles (1990). ¶ III.169 »arbitrary«: see Note to II.48. ¶ III.170 »rewriting«: for Chomsky (1957:26ff), the term covers both the conversion of a word or phrase into symbols and vice versa, and the conversion of one symbol or symbol string into another ¶ III.174 Because of their predominance in nature, assigning a small domain to »non-linear systems« in physics has been compared with assigning a small domain to ‘non-elephant animals’ in zoology. »boundary«: see Note to III.152. »Mandelbrot set«: see Mandelbrot (1982). »iteration«: I. Stewart (1987). ¶ III.175 »semantic feature ‘±Human’«: see II.49, 73, and Note to II.49. ¶ III.176 »metabolism«: Bagley & Farmer (1992); Bagley, Farmer, & Fontana (1992); see also III.162, 191(g), 248. »receptor molecules«: Bonner (1988:147). ¶ III.179 Skinner (1957:348ff); see Beaugrande (1984a:57ff). ¶ III.180 »study of man«: C. Geertz (1973:33), citing La Pensée Sauvage (Lévi-Strauss 1966). »chaos producing order«: see Note to III.44. ¶ III.182 »self-similar«: see Mandelbrot (1967) and compare III.68, 174, 190. . On ‘discourse analysis of discourse studies’ and on ‘framing discourse research’ for ‘communicative practice’, see Tracy (1988, 1990).

Commentary to III.F (III.183-201)

¶ III.183 »model«: pioneering theoretical works on models and systems in a fairly sparse perspective include Bertalanffy (1968); Boulding (1968); W. Buckley (ed.) (1968); Stegmüller (1969-73); Prewo et al. (1973) (eds.); compare Note to II.98. »axioms«: I. Stewart (1987:50, 73). On the notion of »schemas« as frameworks for doing science, see Gell-Mann (1994). ¶ III.184 »formal notations«: see Note to II.27. »refuted by test scales«: see II.28, 41, 49, 54; III.168, 256. ¶ III.185 »cognitive revo-lution«: see Note to II.97. »goodness of fit« and »approximation«: see Notes to III.22; »modeling styles«: see Note to III.5. ¶ III.186 These »ecological goals« would correspond to the German term ‘Erkenntnisinteressen’ (cf. Hartmann 1963a; Schmidt 1978) harking back to sources like Cassirer (1906-07). »ecologist«: see ­Note to I.10. ¶ III.188 On these »scientific moves«, see Hartmann (1963a); »linguists dispute«: e.g. Chomsky (1957, 1965). ¶ III.190 »build models of our own activities«: cf. II.61, 131; III.11, 30, 33, 68, 99, 174, 182, 188. On ‘coherence in discourse’ see Tomlin (ed.) (1987), and in conversation see Tracy & Craig (eds.) (1983), and in »spontaneous« text see now Gernsbacher & Givón (eds.) (1995). ¶ III.191 On ‘natural explanation’ in ‘contemporary science’, see Hilton (ed.) (1988). ¶ III.192 »ecological validity«: see Note to I.33. »ecologist program«: see Note to I.8. ¶ III.193 »fact-finding interviewer«: compare Ecker, Landwehr, Settekorn & Walther (1977); Spradley (1979). ¶ III.194 »diversifying and integrating«: see Note to III.24. »interviews in public media«: see Schwitalla (1979); R.R. Hoffmann (1982); Street (1984); Greatbatch (1988). ¶ III.195 »explanatory work«: cf. II.24, 29, 34, 38f. The speaker of [33] was Charles E. Osgood at a conference (Maclay & Osgood 1959:25). Compare Goldman-Eisler (1958:99): ‘even with highly educated speakers, spontaneous speech is such that well constructed, grammatically correct sentences spoken without repetition or midway changing of grammatical construction, etc., are few and far between’. [34]: R. Watson (preprint dated 1990); compare Linell & Jönsson (1991). »trade-off«: see Note to I.1. ¶ III.197 »think-aloud protocols«: Ericsson & Simon (1980, 1984); Ericsson (1988). Hirsch (1978) proposed a ‘cyber-netic analysis’ of the »conditions of the scientific experiment«. Bronfenbrenner (1977:516) notes that ‘the strangeness of the laboratory situation tends to increase anxiety’ and ‘decrease manifestations of social competence’, especially for ‘lower-class families’.

Commentary to III.G (III.202-12)

¶ III.202 An utterance is of course not at all »linear« in the computational sense of mathematics and physics, because it does not exhibit predictable behavior, nor arise from the sum of its parts, nor evolve in smooth and gradual ways (III.176). ¶ III.204 »linear principles«: I have modified my earlier terms (e.g. Beaugrande 1984a:153ff) to make them clearer and more general: »pacing« from ‘pause’, »merging« from ‘disambiguation’, and »loading« from ‘heaviness’. ¶ III.205 »evidence«: see Beaugrande (1984a:161ff) for more detailed references, especially for »speech pauses«; surveys are also given in Rochester (1973); Matsuhashi (1982). ¶ III.206 [36], [38-42], and [47] were produced by my students in Florida; [35] and [37] are cited from Fromkin (1971:30, 39). Intriguingly, experiments show that even in reading, where the surface text remains available, the eye does not look very far back or very often, and when it does, it tends to fixate strategic prior words likely to assist current processing, e.g., plausible prior nouns to resolve a Pro-Noun (cf. McConkie & Rayner 1976; Just & Carpenter 1980). »miscues«: see Y. Goodman et al. (1987); Allan & D. Watson (eds.) (1976). »collocative«: see Note to IV.39. ¶ III.209 »summary«: cf. Seidlhofer (1996); »salient«: cf. Kintsch (1977:397ff). ¶ III.211 »ego«: W. Cooper & Ross (1975:67); Ertel (1977:156f). ¶ III.212 »aphasics«: Luria (1979:285).

Commentary to III.H (III.213-18)

¶ III.213ff »major factors«: compare the account and references in Beaugrande (1984a: Ch. III). »automatic and attentional«: see Note to III.86; for standard research on these concepts, see Keele (1973). The well-known study by Shiffrin and Schneider (1977; also Schneider & Shiffrin 1977) raises ecological problems, involving huge numbers of trials on simple recognition or recall tasks with sparsely contextualized materials, such as numeric or non-meaningful visual arrays. »speed-accuracy trade-off«: Kintsch, Welsch, Schmalhofer, & Zimny (1990). III.214 »overload«: see also I.53; II.99; III.215f, 281, 222; VII.22f, 66, 81, 84, 87f, 92, 102, 105, 109, 114, 117, 194f, 210, 214, 232, 235, 263, 268; »degradation«: cf. I.99; III.212, 215; VII.23, 84, 214, 268. »critical dispersion«: see Note to III.50.

Commentary to III.I (III.219-57)

¶ III.220 »multi-system«: see Note to I.18; »processing domains«: an earlier version (Beaugrande 1984a: 105-20) used the term ‘phases’, which suggests a fixed temporal grouping of operations, utterly inappropriate in a massively parallel distributed processing system (cf. III.238 and Notes to II.102 and III.93). The actual temporal order is set by scheduling (III.221ff). ¶ III.221 »Black boxes« came from an engineering term for components (e.g. electric circuits) whose input-output function was specified but whose internal construction was not. The behaviorist model of ‘stimulus and response’ was a metaphorical analogue, representing only input and output but not the intervening processes (III.163). The cognitive revolution opened the way for ‘glass box models’. Now, models of »parallel distributed processing« do without any ‘boxes’. »emergent properties«: see Note to II.102. These »partially fitting constraints« are not ‘rules’ in the sense of mainstream linguistics (Hinton, McClelland, & Rumelhart 1986:78); see III.247ff. ¶ III.222 »process each word«: this mode of operation is favored by researchers who, like Keith Rayner (1977) and Marcel Just and Patricia Carpenter (1980), have special instruments to track the fixations of the eye during reading; cf. Notes to III.206 and III.253. »science conceals«: Bazerman (1988); compare Banks (1995). ¶ III.224 »processing depth« was a term current in psycholinguistics (e.g. Mistler-Lachman 1974; Craik & Tulving 1975); survey in Beaugrande (1980-81). The terms »shallower« and »deeper« should not get confused with models where local versus global units are called ‘low level’ versus ‘high level’ (see Note to II.102). On »micro« versus »macro«, see Schelling (1978); Vipond (1980); Alexander, Giesen, Münch, & Smelser (eds.) (1987); Schegloff (1987b). ¶ III.225 »computational research«: Newell & Simon (1972); Winston (1977). ¶ III.226 In a programmatic contrast to psycholinguists, van Dijk & Kintsch (1983) adopted the term »situation model«, but did not address it in the experimental work they reported. »Scenario« is common in ethnographic work and suggests roles and staging. ¶ III.229 »text-world model«: compare Deidre Gentner & Stevens (eds.) (1983); Denhière & Baudet (1989); Garnham (1989). ¶ III.232f On the notion of »instruction« in the linguistics of text and discourse, compare Schmidt (1973); Weinrich (1976); Givón (1989). »spreading activation«: Collins & Loftus (1975); see also III.246, 249. ¶ III.234 On »programs« for »utterance« and »inscription«, which I adopted from Peter MacNeilage (1964, 1970, 1980), see references in Beaugrande (1984a:116f). ¶ III.236 »discourse markers« and »spoken paragraphs«: see Longacre (1970). ¶ III.237 On »lexical items« first versus »grammatical pattern« first, see references in Beaugrande (1984a:116). ¶ III.240 »compare with alternatives«: see Note to I.19. ¶ III.241 »degrees of skill«, »progress«: see Snow (1990). ¶ III.243 On »memory spans« and their respective »focuses«, see references in Beaugrande (1984a:125ff). ¶ III.246 On »exciting« and »inhibiting« in networks, see Rumelhart, McClelland et al. (1986), who use the odd term ‘relaxing’ where »stabilizing« would seem more apt; on the ‘strength’ of connections in text processing, see Abbott & Black (1982). »inferencing«, »updating«, »reinstating«: references in Beaugrande (1980:26); and see now van de Velde (1984); Garrod, O’Brien, Morris, & Rayner (1990); Sanford (1990). ¶ III.248 »parallel distributed processing«: cf. Notes to II.102; III.93, and III.220. »combinatorial explosion«: cf. Woods (1970); »connectionist«: see Note to II.102; this description follows McClelland, Rumelhart & Hinton (1986:10ff). »not use linguistic rules«: see Rumelhart & McClelland (1986); and compare Bereiter (1991). ¶ III.249 »priming«: see Swinney (1979); Kintsch & Mross (1985); Kintsch (1988, 1989); and Till, Mross, & Kintsch (1988), who supplied [58]. »half a second«: the typical time breakdown is: feature detection 0-50 milliseconds; sense activation 50-350 ms.; sense selection 350-500 ms.; and sense elaboration after 500 ms. (Kintsch 1989:199; cf. Kintsch, Welsch, Schmalhofer & Zimny 1990). ¶ III.250 Kintsch (1988:163; 1992:276) emphasises the »bottom-up« aspects of his model (see also Mannes & Kintsch 1991:310), but the top-down aspect is retained via the ‘general knowledge network’ (e.g. Kintsch 1988:180), which both supplies the raw material and steers the ‘integrating’. For the same reason, his emphasis on ‘perception-like aspects as distinct from controlled conscious problem-solving processes’ (1992:276) does not imply a fresh dichotomy between perception versus cognition (cf. III.79). »faster and cheaper«: Kintsch (1988:163f), contrasted this to the ‘smart system’ of Kintsch & Greeno (1985), simulated by Fletcher (1985) and Dellarosa (1986); the new model is »simulated« by Mannes (1989). ¶ III.251 »Non-selective memory access« fits the currently leading ‘random-access model of memory of Raaijmakers & Shiffrin (1976); compare Golden, Rumelhart, Strickland, & Ting (1993) on ‘random fields for text comprehension’. »currently active version«: see Note to I.47. ¶ III.253 »revision driven by empirical data«: compare Morton (1979) changing his ‘logogen model’ of activation. »difficulties«: e.g. Schank (1978:94): ‘in natural language understanding’, ‘analysis proceeds in a top-down predictive manner; ‘it is only when expectations are useless or wrong that bottom-up processing begins’. »eye fixations«: Just and Carpenter (1980); cf. Notes to III.206 and III.222; no stored ‘schema’: Kintsch (1988:164); cf. II.97. ¶ III.255 »discourse modeled« via »self-organizing systems«: compare II.101f; III.117, 135f, 241, 246. »Parallel distributed processing«: see Notes to II.102, III.93, and III.220; »cascading«: compare McClelland (1979) on ‘systems of processes in cascade’.

Click here to go to Ch. IV

Click here to go to Foundations Main Page