Basic Issues systems and models

Basic Issues

1. SYSTEMS AND MODELS

1.1 In many countries, the increasingly prominent role of text linguistics in the discipline of language study appears to signal a “paradigm shift” in the sense of Thomas Kuhn (1970). The older preoccupation with demonstration sentences isolated from communicative contexts is yielding to a new concern for the naturally occurring manifestation of language: the TEXT. Language occurrences may have the surface format of single words or sentences, but they occur as texts: meaningful configurations of language intended to communicate. The implications of this shift of investigation are far-reaching indeed. We are not simply moving from the exploration of shorter toward longer language samples. We are also replacing an emphasis on abstract forms with an interest in the UTILIZATION PROCESSES of language (cf. Hörmann 1976; H. Clark & E. Clark 1977).

1.3 In this wide framework, text linguistics would constitute the verbal domain of semiotics, dealing with the entire range from one-word texts (e.g., ‘Fire!’)¹ [Throughout the book, I enclose linguistic samples cited within the running text in single quotes; punctuation is included inside only if part of the sample to texts as vast as The Divine Comedy. The decisive trait of the text is its OCCURRENCE IN COMMUNICATION (Hartmann 1964), where it is produced by a single participant within some temporal limits (cf. Weinrich 1976: 187). A set of mutually relevant texts can be said to constitute DISCOURSE,² [2. The term “discourse analysis’ has been variously used for beyond-the-sentence linguistics in general (Harris 1952) and for the study of conversation in particular (Coulthard 1977). In my scheme, both of those domains are only parts of a science of texts as actual communicative occurrences.] a progression of occurrences that may be continued at a later time (cf. van Dijk 1977a). The total constellation of mutually relevant discourses in a group or society can be called a UNIVERSE OF DISCOURSE (cf. Coseriu 1955-56; Pike 1967: 596; van Dijk 1977a: 127).

1.4 Human language is so complex in its organization and so diverse in its manifestations that the science of linguistics naturally remains in continual evolution. The linguist faces a formidable overabundance of data ranging from observable, face-to-face speaking to abstruse mathematical and philosophical speculations on language. In its early stages, linguistics was compelled to be highly selective and reductive in its treatment (Uhlenbeck 1973: 107; Grimes 1975: 3). Continued progress has rendered much of this attitude superfluous, though debates over admissible issues still rage. As our scope expands, we approach the time when linguistics may be able to meet the demands that society imposes upon a science of language (cf. Hartmann 1970).

1.5 However much data a researcher may gather and evaluate, data can be significant only with respect to the COGNITIVE INTERESTS of a discipline (cf. Kuhn 1970; Schmidt 1975): commitments to seeking certain kinds of knowledge. Especially in linguistics, what constitutes worthwhile data, or how data should be treated, is by no means self-evident. The cognitive interests expressed in this volume are devoted to texts as vehicles of HUMAN ACTIVITIES — a notion already envisioned by many founders of linguistics (e.g. Malinowski 1923; Jespersen 1924; Bühler 1934).

1.6 Using a term made current by Carl Hempel (quoted in Stegmuller 1969: 205), we can define the scientific treatment of data as SYSTEMIZATION: the imposition of a system upon obtainable evidence. A SYSTEM is considered to be a unity of mutually relevant ELEMENTS³ [3.I use the term ‘element” for any item whose occurrence or use is governed by SYSTEMIC (pertaining to a system) principles.] whose FUNCTIONS are determined by their respective contributions to the workings of the whole. To account for data, researchers build a SYSTEMIC MODEL whose operations might yield such data (on the notion of the model, see Hartmann 1965; Gülich & Raible 1977). The correlation of the model with an empirical domain is regulated by BRIDGE PRINCIPLES (Hempel 1966) that state the degree of APPROXIMATION between the model and the domain (cf. Apostel 1961). Ideally, scientific progress continually reduces the degree of approximation and makes the model a more exact representation. However, the object of study may itself be FUZZY: characterized by structures and operations that are probabilistic and not exhaustively delimited. Such is the case with natural language communication.

1.7 The view of language as a system is well established (see for instance Saussure 1916; Weinreich 1954; Firth 1957; Halliday 1967a, 1969; Heger 1971; Laboy 1971; Winograd 1972; Berry 1977; Clark & Clark 1977; van Dijk 1977a). The systems approach implies cognitive interest in the DYNAMICS of an entity (Hartmann 1963a, 1963b), such as CONTROL and INTERACTION among elements. Systems theory, which has become a discipline in its own right (cf. Boulding 1956; Bertalanffy 1962; Buckley [ed.] 1968), has found acceptance in research areas as diverse as psychology, sociology, behavioral science, design engineering, information science, computer science, factor analysis, thermodynamics, mathematical topology, and many more. The prevalence of the systems approach is heuristically enriching in allowing models to be shared and borrowed among disciplines. To be sure, borrowed models have crucial effects upon research methods, and must be applied with caution.

1.8 Language is initially given as a MANIFESTATION: an occurrence or set of occurrences at least partly accessible to apperception (cf. Stegmüller 1969: 93). The observable aspects interact with non-observable ones in intricate and diverse ways. The total picture of a language must be gradually assembled via a sequence of systemizing tasks, for example:

1.8.1        IDENTIFYING a manifestation, its constituents, or its environment;

1.8.2        GENERALIZING about related or relevant manifestations;

1.8.3        DESCRIBING a set of manifestations methodically;

1.8.4        EXPLAINING the existence or occurrence of manifestations;

1.8.5        PREDICTING manifestations under statable conditions;

1.8.6        RECONSTRUCTING artificial correlates of the manifestations;

1.8.7        MANAGING the occurrence of manifestations.

1.9 This list of tasks is arranged in what 1 consider the order of increasing difficulty. Accomplishment of a given task presupposes prior achievement of those above it. In practice, however, we must often work provisionally without these prerequisites. For example, public education may require the management of language in absence of any thorough explanatory or predictive account.

1.10 In order to establish itself as a discipline, linguistics initially dissociated itself from the prescriptive ambitions of traditional grammar and sought to develop some objective, reliable tools for the more basic tasks of identification, generalization, and description. This early phase, usually entitled “descriptive linguistics,” attained a sufficiently rigorous methodology to uncover the grammars of numerous hitherto unrecorded languages, even with no prior knowledge of their structure (especially via the “tagmemics” developed by Kenneth Pike [1967] and Robert E. Longacre [ 1964]). Emphasis was understandably placed upon the language aspects most accessible to observation: sounds, forms, and arrangements of utterances. The treatment of non-observable aspects, such as communicative strategies or psychological processes,⁴ was informal and intuitive. Communication as a human activity was not viewed as a major object of study in its own right.

1.11 The “generative” approach to language study embarked on the more arduous tasks of explanation, prediction, and reconstruction (beginning with Hjelmslev 1943; compare Chomsky 1957). Generativists borrowed heavily on formal logic to build an idealized model of human language with stringent restrictions on the object of study. The divergences between their model and the real data were sufficiently strong in some domains to cause an apparent antagonism between “model-oriented” and “data-oriented” studies (Liefrink 1973). However, the expansion of linguistics to new tasks remains an enduring contribution of the generative approach (cf. Dingwall 1971).

1.12 If models mediate between what we can apperceive and what we want to explain (motto attributed to Anaxagoras by Gulich & Raible 1977: 14; cf. Wagner 1974: 150), then the following situations would justify reliance on models (adapted from Apostel 1961):

1.12.1 when no known theory exists for a domain;

1.12.2 when a theory is known, but too complex to allow solving problems with currently available techniques;

1.12.3 when a theory is known and partly confirmed, but still incomplete;

1.12.4 when new research permits the correlation or integration of two or more known theories;

1.12.5 when the objects of investigation are too large, too small, too remote, or too arduous to allow direct observation and experimentation.

1.13 All of these situations obtain to some degree in linguistics:

1.13.1 Some areas are still without a workable theory, such as the interface of language with emotional states of communicative participants. Peter Hartmann (personal communication) comments that the notion of ‘observation’ in these early discussions was somewhat out of place. The actualisation of communicative strategies guarantees the potential for objective study, whether or not we are looking directly at the data in its most detailed form.

1.13.2 Some theories are available, such as Montague’s (1974) grammar, but are too intricate to be useful in solving empirical language problems, at least with current technology.

1.13.3 The “generative’ approach to language remains incomplete until it can explain how texts are actually produced and understood by humans.

1.13.4 Research on texts demands the integration of theories from many areas: sentence grammar, philosophy, computation, cognition, planning, and action. 1.13.5 Some aspects of language are too large (e.g. the totality of discourse in a whole society), too small (micro-impulses of nerve cells during language processing), too remote (storage of knowledge in the mind) or too arduous (relating every minimal feature of utterances to its social, psychological, and historical evolution) to be pursued in direct experimentation.

1. 14 To sort out the regularities of language from the accidental details, one can distinguish between the SYSTEMIC aspects of language, styled “langue” or “competence,” and the seemingly accidental or irrelevant aspects, designated “parole’ or “performance.” During the evolution of the discipline, the borderline between the two aspects shifts as domains once thought to be fortuitous are discovered to have a systemic nature after all. For instance, the sentence was assigned to “parole” by Saussure (1916: 172), only to become the primary entity of “competence” in transformational grammar since Chomsky (1957). I shall argue in 1.5 that “textual competence” encompasses a markedly different domain from sentence competence (cf. also IV. 1.24).

1.15 The respective limits of “langue” or “competence” influence and are influenced by the models and methods in use. Descriptivists broke down their sample evidence by extracting LEVELS of MINIMAL UNITS to be classified into a TAXONOMY: a scheme for sorting elements by distinctive features. If each level of minimal units was thought to be a system of mutual oppositions, the entire repertory of each system had to be exhaustively assembled, for example sounds (phonology) and forms (morphology); exhaustive treatment of meanings or situations was deemed impossible, and those domains were set aside. Later, the generativists preferred to begin at the other end with a GRAMMAR as a set of rules stipulating what does or does not belong to the language. The problem of exhaustiveness was suspended by postulating that all complex entities (however many there might be) could be derived from a limited set of simple ones (kernels) by using the proper rules. The rules were designed so as to produce an infinite set of sentences.

1.16 The generative approach is vastly more ambitious than the descriptive one, since it must not only systemize all occurrences: it must also preclude all non-occurrences (McCawley 1976). Strictly speaking, it is not a grammar of occurrences at all, because it professes to deal only with abstract potential. The empirical verification of such a grammar can be a major difficulty. Outside the relatively small range of obvious, uncontroversial cases, people have trouble judging which utterances their native language should allow (cf. Labov 1966; Lakoff 1969; Carden 1970; Heringer 1970; Wedge & Ingemann 1970; Ringen 1975). Grammaticalness judgments are supposed to apply only to structure, not to context. But structures never occur naturally without context, and language users therefore do not possess the expertise needed to make consistent judgments. Informants are, in reality, trying to imagine possible contexts for each sample (Uhlenbeck 1973: 42; McCawley 1976: 155; van Dijk 1977c; Snow & Meijer 1977). The designation of “grammatical” is awarded to banal sentences whose occurrence is easy to imagine for everyone (Householder 1960: 340). For less banal cases, opinions are unstable and inconsistent. Heringer (1970) found that providing a context for such a simple sentence as ‘John left until 6 P.M.’ elicited a change in people’s grammaticalness judgments of no less than 40%! People are still less able to decide about elaborate utterances, such as those which Robinson (1975: 141ff.) contests against Chomsky (1972).

1.17 The lack of empirical verification procedures for large sections of a language theory can lead to disquieting tactics (cf. Beaugrande 1979k):

1.17.1 Circularity of proof: the correct rules are those which generate only grammatical sentences; grammatical sentences are those generated by the rules (cf. critique in Dik 1967; Uhlenbeek 1973).

1.17.2 Intuitions: the linguist presents demonstration sentences which are (in his or her own intuition) grammatical or ungrammatical (criticism in Dik 1967: 372; J. Anderson 1976 :69; Schiesinger 1977: 210). The linguist becomes the informant and is free to exclude unfavorable examples at will (cf. Rieser 1978: 8), thus sacrificing both objectivity and generality.

1.17.3 Appeals to the competence/performance distinction: all data that the theory cannot treat are shunted off into the domain of “performance” and excluded from consideration as non-issues (criticism in van Dijk 1972a: 314). An egregious and anachronistic illustration is Dresher and Hornstein’s (1976: 328) claim that “a study of competence abstracts away from the whole question of performance, which deals with problems of how language is processed in real time, why speakers say what they say, how language is used in various social groups, how it is used in communication, etc.” A language user without all of that knowledge could be called “competent” only by a bizarre perversion of the whole notion in its commonsense usage.

1.18 If a language model can do no more than assign structural descriptions to sentences, then it deserves the old designation “structural descriptive linguistics” rather than “generative.” To earn the latter designation, the model should suggest how utterances might actually evolve (J. Anderson 1976: 118; Simmons 1978: 2). Only then can we reasonably conduct empirical tests and agree on rational standards for evaluating, verifying, and accepting one account of language over any other.

 

2. LEVELS IN MODELS OF LANGUAGE

2.1 A spoken or written text in English could appear to be or to consist of various things. One observer might notice a stream of sounds or forms following each other in the real time of speaking or of moving from left to right on a page. Another might notice that the text is intended to embody knowledge and meaning. Still another might notice that the text could be a vehicle for someone to get something done or to reach a goal. Each of these observers would be apperceiving a single, simultaneous aspect of the text: one of its LEVELS.⁵ [5. The term ‘level” has been used indiscriminately in the past, often being conflated with such notions as ‘rank.” I consider a “level” to be the total aspect of a participating language system; a ‘rank” is a unit of a given dimension in a hierarchy of size (e.g. word, sentence, etc.).] It seems reasonable that language science should attempt to extract and systemize these levels as a proper domain of investigation.

2.2 In its early phases, linguistics proceeded on the assumption that levels should be systemized independently of each other (e.g., Trager 1950). That outlook seemed to be successful for the description of sounds, though, as Kenneth Pike (1967: 362f.) notes, it was not fully upheld even by its defenders. Later, acute problems emerged when the borderline between morphology and syntax came into view. A distinction was drawn between the paradigmatic aspect that determined what items might fill a slot, and the syntagmatic aspect that determined the sequence of slots themselves.

2.3 The independence of syntax from meaning was maintained with considerable vigor (cf. Harris 1951; Chomsky 1957). Harris himself allowed for an expedient to use meaning as a short-cut for analysing language, provided that a purely formal analysis of the distribution of language items would arrive at the same result. In effect, Harris postulated that an item’s meaning is the sum of all positions it can occupy in usage. This postulate is not in itself unreasonable, but for a model of either linguistic analysis or human language activities, it is unworkable. Meaning would remain undiscoverable until we had catalogued all distributions of a given item.

2.4 Rather than upholding the separation of syntax from meaning, Harris’ postulate tends to suspend it. Meaning and syntax must interact in order for language items to have a given distribution. I shall argue, however, that we should go further and investigate the PROBABILITY of occurrences in systematic environments. The “well-formedness” (i.e., conformity with the grammar) of language sequences is not, taken by itself, a sufficient principle (cf. 11. 2.36ff.; IV.1.24).

2.5 Although reaffirming the independence of syntax from meaning, Chomsky’s (1957) “transformational grammar” sought to escape the unworkability of Harris’ postulate of distribution. Instead of analysing the distributions of language items as such, Chomsky undertook to set up an abstract rule system which could produce all allowable distributions of a language. Attention was transferred from the analysis of extensive samples toward the construction of rules. In essence, this transfer did not simplify linguistic research. Every counter-example to the previous rules called for new rules — a factor making the transformational model immune to falsification as a theory.

2.6 In semiotics, it is traditional to subsume all aspects of formal arrangement under the notion of SYNTAX, and all aspects of meaning under that of SEMANTICS; the use of language was subsumed under PRAGMATICS. To deal with an entire language, the “transformational” approach began with an autonomous set of syntactic rules; semantics was treated as an after-the-fact ‘interpretation’ of syntactically produced strings. In some models, pragmatics was simply added as a further phase of “interpretation” (e.g. van Dijk 1976). Such an approach was obliged either to ignore the interaction of these three factors in the actual production and comprehension of utterances, or to reconstruct them all in terms of arbitrary syntactic rules. An alternative account in which meaning was given the key role from the outset was introduced as ‘generative semantics” (see. 11. 1.6). Quite aside from the detailed issues of rule constructing, this controversy pointed up a basic question concerning the building of language models. I shall examine the question here from a systems-oriented viewpoint.

2.7 In systems theory, we can distinguish the approaches of MODULARITY and INTERACTION (cf. Sussman 1973: 12f.; Winograd 1975: 192). The reliance upon formal logic and mathematics in generative grammar fosters modularity, in which system components are substantially independent, and operations are cumbersome (cf. Levesque & Mylopoulos 1978: 2). My outlook here will be directed toward interaction, without which the utilization of text would simply not be operational (cf. 11. 1; Walker [ed.] 1978).

2.8 Imagine for a moment a different kind of language model. We might start out with these two well-known levels:

2.8.1 SYNTAX PROPER is concerned with the abstract patterns and sequences which the grammar of a language stipulates independently of context.

2.8.2 SEMANTICS PROPER is concerned with “the relations between signs or symbols and what they denote or mean” (Woods 1975: 41). The repertory of signs and symbols with a statement of their meanings is contained in the LEXICON. If lexical items are defined according to their content, we have INTENSIONAL MEANING (e.g., ‘blue’ is the color lying in the spectrum between green and violet); if items are defined by their REFERENCE to entities, we have EXTENSIONAL MEANING (e.g.’blue’ is the property shared by all blue things in the world). The standards for judging the correctness of statements about some world and for combining statements in that perspective are set forth as TRUTH CONDITIONS. The extent to which a reference encompasses an object or class of objects is the issue of QUANTIFICATION (e.g. ‘every person’ or ‘all persons’). The probability or necessity of a statement regarding some world is its MODALITY.

2.9 In the above definitions, syntax and semantics proper are indeed independent of each other. Formal sequences can be envisioned before deciding what specific lexical items might fit into them; and lexical meaning need not commit an expression to appearing in a given slot of a sequence. Yet no utterance could ever be produced without making these decisions and commitments, and none could be understood without recovering them. It follows that syntax proper and semantics proper as previously set forth are components of logical languages, but not of natural languages in use. Instead, let us envision two different levels of language use:

2.9.1 The SEMANTICS OF SYNTAX is concerned with how people utilize formal patterns and sequences to apply, convey, and recover knowledge and meaning. For example, noticing a noun-verb sequence might give rise to an expectation that an agent and an action are being expressed (cf. III.4.16.1).

2.9.2 The SYNTAX OF SEMANTICS is concerned with how concepts like agent, action, state, attribute, etc. are connected to yield the total meaning of a text. Semantics of syntax has a more predominant linear or sequential organization than does syntax of semantics. For instance, an action might be linked to an agent, a time, a location, a cause, and so on, while various linear arrangements would be possible for expression (cf. VII.2) (on grammars without fixed linearity, cf. Petöfi 1972).

2.10 These interfaced levels are not wholly novel here (compare Ihwe 1972: 339; Schank 1975b: 14f; Rieser 1976: 13). Their function is to co-ordinate operations people perform when utilizing meaning in connected utterances. I shall develop this direction by pursuing the notions of SEQUENTIAL CONNECTIVITY: how elements are arranged in the surface text; and CONCEPTUAL CONNECTIVITY: how underlying concepts and relations are put together.⁶ [6. Compare the notion of ‘sense constancy’ in Hörmann 1976, ch. 7.] The interaction of these two is controlled by MAPPING PROCEDURES (cf. Goldman, Balzer, & Wile 1977). The characteristic mapping procedures selected to produce a text yield the STYLE of the text.

2.11 When each language level is systemized, the entire language appears as an INTERSYSTEM the workings of which depend on the interaction of participating systems (cf. Halliday 1969; Berry 1977; Dressier 1979). Each system has INTERNAL CONTROLS that regulate the availability of options and the allowability of combinations; and EXTERNAL CONTROLS which regulate that system’s interaction with other SYSTEMS.⁷ [7. In his general systems theory, Luhmann (1970) stipulates that every system must have a “differentiation of internal and external,” i.e. be distinguishable from its environment. Linguistics has been concerned with differentiating sets of system elements at the expense of operations, functions, and controls. As J. Anderson (1976: 80) notes, control has hardly been studied in cognitive theories.] Both kinds are indispensable to the production and utilization of texts, but external controls have received little attention in conventional linguistics.

2.12 Issues of control were largely assigned to the poorly explored domain of pragmatics, dealing with language use. In that line of reasoning, pragmatics becomes “meta-syntax” and “meta-semanties,” rather like a self-awareness of decisions about arrangement and meaning. However, in order to attain a workable design, each system should have at least its own essential controls built right into it. Pragmatics is properly the domain of the human activities of PLANNING texts as vehicles of PURPOSIVE ACTIONS directed toward GOALS (Beaugrande 1979b). The theory of texts accordingly requires a different triad of domains than the old semiotic scheme:

SYNTAX => SEQUENTIAL CONNECTIVITY

PRAGMATICS => ACTIONS, PLANS, GOALS

Each domain is subject to relevant controls during communication. The discrete items occur within a CONTINUITY which arises from the DIRECTIONALITY of CONTROL FLOW. Accordingly, we need a dynamic outlook for investigating not only the presence of structures in texts, but also the operations that can create, build, and utilize structures (cf. Hartmann 1963a; Mukarovsky 1967: 11; Woods 1970; Winograd 1972; Koch 1976). If we define STRUCTURE as a relation between at least two systemic elements in occurrence, it is clear that a theory of language use should be centered upon the notion of CONNECTIVITY.

3. TEXT VERSUS SENTENCE

3.1 Nearly all accounts of language structure set forth since classical antiquity have relied decisively on the notion of the SENTENCE. It is disquieting that this basic entity has been vaguely and inconsistently defined, even up to the present (O’Connell 1977; Glinz 1979; Beaugrande 1999). Different criteria for ‘sentencehood’ have persisted without being explicitly recognized as divisive rather than unifying standards, for example: (1) the expression of a “complete thought” (see Ivic 1965: 20); (2) a sequence of speech units followed by a pause (see Gardiner 1932: 207; Goldman-Eisler 1972); (3) a structural pattern with specified formal constituents (cf. Harris 1951; C. Fries 1952; Chomsky 1957). The functional implications of each of these criteria are radically distinct from the others. Empirical research makes it plain that people disagree with each other’s judgments about what constitutes a sentence. When speech pauses were consulted, “many segments identified by this study as sentences would not be considered separate sentences by other criteria” (Broen 1971: 30). A still graver problem is that the boundaries of utterances are often marked by non-linguistic signals (Hörmann 1976: 329).

3.3 It appears that linguists tended to confuse optional mapping with obligatory mapping. The sentence provides no more than a grammatical format into which semantic and pragmatic unities can be mapped; some linguists treated the sentence as the format into which these unities must be mapped. This practice undermined the proclaimed autonomy of syntax, because many attributes assigned to the sentence actually belong to semantic and pragmatic unities. Consequently, the question of how humans decide what to map onto what could hardly be raised. We can readily observe that people can make a wide range of decisions about syntactic formatting (cf. VII.2). As long as linguistics presupposes the sentence at the outset, such facts are difficult to treat; the linguist is compelled to retreat into a remote, reductive version of “ideal competence.” An impasse has arisen beyond which linguistic theory cannot advance, because the most basic concepts are short- circuited across each other, rendering many vital realities of communication inadmissible issues.

3.4 I assert that the multi-level entity of language must be the TEXT, composed of STRETCHES OF TEXT which may or may not be formatted as sentences. I would cite the following essential distinctions between text and sentence:

3.4.1 The text is an ACTUAL SYSTEM, while sentences are elements of a VIRTUAL SYSTEM, as I shall explain in 1.4.