II.  

Scientific Precursors

1. STRUCTURALISM AND LINGUISTICS

    1.1 Many diverse trends have carried the label "structuralism," sometimes against their will. Different structuralist approaches shared some consistent, stable tendencies, yet often disagreed vigorously about the nature of their enterprise. Their investigations of language or language-related domains were dominated by the search for WELL-STRUCTUREDNESS — a common ambition of scientists (1.1.3). If a "structure" is two or more elements with a relation obtaining between them (1.4.5), structuralism is the exploration of formal units and their patterns. In some domains, the nature of those units and patterns, and the methods for uncovering them, appear comparatively straightforward. But language offers considerable latitude for controversy.

 1.2 Ferdinand de Saussure's pioneering "course in general linguistics" (1916) proved enormously influential and conformed to the conventional ambitions of scientists depicted in 1.1. Because it was published by his pupils after his death, the basic tenets of his theory remained open to differing interpretations (cf. Wells, 1947a; Godel, 1966). Saussure defined what he called "langue" "as a purely relational structure, as a pattern, as opposed to the usage […] in which this pattern is accidentally manifested" (Hjelmslev, 1947: 73);¹ [I translate into English any quotations that require it.] the "usage" was termed "parole." In effect, Saussure was selecting the virtual system over its actualizations (I.4.5) and inaugurating an abstraction that became a fundamental methodological step for many schools of modern linguistics. Saussure also proposed several further abstractions:

 1.2.1 Language should be studied as synchronic system: a configuration of elements captured at a single point in time; the diachronic perspective {44} of the system evolving and changing in historical time was relegated to a separate field (termed “philology” or “historical linguistics”).¹[1 “Philology” usually denotes language inquiry (e.g. Delbrück, 1904) as developed before the rise of modern linguistics. “Historical linguistics” is usually the extension of modern methods to language history.]

 1.2.2 Language was to be considered a system of signs, i.e., perceivable traces with a conventional meaning and purpose. Each sign would arbitrarily relate an “acoustic image,” later called a “signifier,” to a “concept,” later called a “signified.” Saussure (1916: 98ff) hoped this approach would supplant the popular view (going back at least to the account in the Book of Genesis) that a language item is a “name” for a “real thing” in the world. However, as Helbig (1974: 39f) remarks, Saussure himself fell back into this view by proving the arbitrariness of sound-meaning relations on the grounds that the same real object has different names in different languages (cf. Gardiner, 1932: 29ff; Benveniste, 1939: 24). The alternative would have been to recognize that meaning can be constituted during usage (Schmidt, 1969; cf. 1.4.6f; III.2.26); and Saussure had already excluded usage from linguistic inquiry (II.1.2).

 1.2.3 As a system, language was to be described via the notion of opposition: each sign or element would be defined by how it differs from all others signs. To be sure, exact recordings and measurements reveal that every utterance is physically different from all others, even when the same person repeats the same words. However, linguistics should be concerned with only those differences that are SYSTEMIC, i.e., essential to the definition of the system. (For example, Arabic has two “glottal stops” one pronounced in the back of the throat (transcribed ‘q’) and one in the front (transcribed ‘k’), whereas English treats both as the same sound: the difference is systemic in Arabic, but not in English.) Systemic differences rest upon distinctive features. (For the Arabic example, the two sounds differ in the distinctive features of “back” vs. “front”, yet share other features, e.g., the “stop” of the air flow.) The sounds of a language form a system based on these opposed features (I.1.5).

 1.3 These abstractions — synchronic systems of signs defined by mutual oppositions and distinctive features — became the cornerstone of structuralist approaches, both inside and outside linguistics. Such abstractions lend considerable importance to the reconstructive actions of the investigator: not just to uncover units and features, but to decide what constitutes a feature and which features should be postulated. Martinet (1962: 59) surmised that the structures were “nothing but a frame invented by the linguist in order to help him classify the data. In other words, a structuralist is not one who discovers structures, but one who makes them” (emphasis added). As Hartmann (1963a, 1963b, 1964) has emphatically demonstrated, signs and sign patterns have reality only as actual occurrences. To have an abstract system, the speakers of a language (including a linguistic investigator) {45} must first synthesize real experiences and invent their underlying principles (I.4.12). This factor was overlooked in a “science” whose “empiricism” “suppresses the constructive role of operations”; yet “in ‘operating’ on the objects, the subject alone, by his action, elaborates structures” (Piaget, 1976: 132). The “objectivity” of the human sciences is the result, rather than the exclusion, of cognitive (subjective) activities among the scientists themselves. Cognition is controlled by one’s prior model of the world.

  1.4 It is perhaps fair to say that the central enterprise of modern linguistics has consisted of assigning to language theoretical substrates of structure (Beaugrande, 1982d, 1983a). This trend brought a steady increase in the theoretical well-structuredess of language and thus rendered it a more convincing object of scientific investigation (cf. I.1.3). However, the scientific climate led many structuralists to act as if the structures were situated in the object of inquiry. As long as the human processes of discovering and utilizing structures were carried out without acknowledgement by the structuralists themselves (cf. I.1.3), the most interesting aspect of the enterprise remained hidden from view. As structuralism moved toward more complex domains of language, unity and consensus steadily diminished, because the investigators had to supply more diffuse, elaborate contexts.

 1.5 The history of structural linguistics supports this contention. The most successful project was phonology, the study of sound systems. The positions of the vocal organs during the production of language sounds were used for classifying the sound system into units called phonemes: the smallest sound units capable of determining meaning. Phonemes were “systemized” according to whether the vocal cords vibrate and how the tongue, lips, teeth, gums, palate, etc. participate. The catalog of “distinctive features” included “voiced,” “unvoiced,” “dental,” “palatal,” “glottal,” and so on (cf. Trubetzkoy, 1939). Jakobson, Fant, and Halle (1952) reaffirmed Saussure’s outlook (II. 1.2.3) by situating all features in binary oppositions such as “ + /- voiced,” “ + /- dental,” etc. The binary opposition became a further hallmark of structuralism,’ and obscured the recognition of gradations (cf. I.4.15).¹ [The binary opposition is vacuous unless both sides are relevant alternatives. Otherwise, we could put “ + /-” in front of anything because its presence is trivially the opposite of its absence (cf. Werth, 1976). Linguists tended to use the “ + /-” symbol as a trademark for making vague concepts look “scientific.”]

 1.6 Phonology proved to be the most enduring success of structuralist linguistics precisely because its categories are closely allied to observable sound events. The problem of discovering structures vs. creating them was not acute as long as researchers agreed on units and features-easy enough in regard to sounds. This success brought enough credibility to structuralism that researchers widely accepted and applied its basic assumptions; later structuralist explorations of language attempted to generalize the principles {46} that had worked for phononogy (Pike, 1967: 95; Koch, 1973-74: xvi; Quirk, 1976: v). Yet the correspondence between sound systems and the “deeper levels” of language, such as meaning and purpose, is far from obvious. The processing actions that investigators had to perform in order to discover structures were much more complex than what was needed for sound events (II. 1.4). By asserting arbitrary ties between meanings and words, structuralism precluded a study of meaning in terms of conceptual events (II. 1.2.2). They failed to appreciate the difference between the total units and features that analysis can abstract vs. the relevant features which are actually consulted and utilized in language processing. The latter are most essential for theories of communication and cognition, even in phonology (Beaugrande, 1978b).

 1.7 The structuralist offshoot known as “Russian Formalism” can illustrate my point. Around 1916, a circle of Russian scholars, including at various time Roman Jakobson, Lev Jakubinskij, Viktor Sklovskij, Boris Eichenbaum, Boris Tomasevskij, Viktor Vinogradov, Jurij Tynjanov, and Viadimir Propp, merged into a “society for the study of poetic language.” These “Formalists,” as they soon came to be known (cf. Eichenbaum, 1965 [1927]),² I place in square brackets original reference dates that bear on the discussion. The Formalists wrote mostly in Russian, so that their impact on Western methods was long delayed until translations were made (e.g. Todoroy [Ed.], 1965; Matejka & Pomorska [Eds.], 1971).They analyzed units and features in such literary text types as poems, short stories, folktales, and novels. Literature was defined by its additional “structuration” (cf. I.4.5) most obvious in poetry that “de-automatizes” language processing and hence intensifies and diversifies literary communication (cf. VI. I. If, 7). Jakobson’s detailed analyses (e.g. Jakobson & Levi-Strauss, 1962; Jakobson & Jones, 1970) “surprise the examiner himself by unexpected, striking symmetries and asymmetries, balanced structures, efficient accumulation of equivalent forms and salient contrasts” (Jabobson, 1968: 602f). However, it is hard to tell how many of these structures are relevant to, or experienced in, the act of producing or receiving the poem, and how many are derived from the language system as such (cf. discussions in Posner, 1971; Lotman, 1976; Werth, 1976; Beaugrande, 1978c, 1978d; Dillon, 1978; Koch, 1978). We need to know how representative the refined methods of the Formalists in finding structures might be for literary communication at large. Some literary scholars have at least expressed a self-awareness of their own role as processors (e.g. Riffaterre, 1966; Fish, 1970, 1976, 1980; Iser, 1980). Ultimately, only general empirical studies among authors and audiences can uncover how people make structures (cf. Wienold, 1972; Groeben, 1980; Schmidt, 1982). The same demand holds for ordinary communication as much as for literature.

 1.8 Soon after the Formalists went to work, a “Linguistic Circle” assembled in Prague to investigate language. {47} At the First International Congress of Slavists in 1929, the structuralist approach was affirmed (1929 report cited in Jakobson, 1973: 11f):  

 Were we to comprise the leading idea of present-day science in its most various manifestations, we could hardly find a more appropriate designation than structuralism. Any set of phenomena examined by contemporary science is treated not as a mechanical agglomeration but as a structural whole, and the basic task is to reveal the inner, whether static or developmental, laws of this system. […] Therefore, immanent structural considerations of language were predestined to take a prominent place in the debates of the Congress.  

The Prague School, dating from 1926, developed a variant of structuralism that sought to relate structures to their functions (cf. surveys in Mathesius, 1947; Trnka, 1957; Garvin [Ed.], 1964). However, an operational definition of “functions” requires a greater consensus about the nature of the system and its elements than the Saussurian abstractions allowed. Research was mostly limited to phonology and word order (survey in Firbas & Golkova, 1976; cf. IV.2.63).

 1.9 A rival school, convening in Copenhagen in 1933, preferred the designation “glossematics” (cf. Hjelmslev, 1943, 1953; Siertsema, 1955; Uldall, 1957). The Danish scholars claimed to follow Saussure’s theories most faithfully.¹ [A letter to Hjelmslev is extant from Charles Bally, Saussure’s successor in Geneva, certifying that the Copenhagen school had indeed most fully understood Saussure’s injunction to study “language in and of itself” (cf. Helbig, 1974: 62).] Like Formalism, Glossematics suffered from a language barrier-Hjelmslev’s fundamental book was in Danish.] They introduced a still greater abstraction by electing to study only “form” and to disregard “substance” (whereas phonology was allied to substantive sound events). Language was seen as a system composed entirely of abstract relationships (“functives”) among elements (“terminals”) that could be either “constants” or “variables” (cf. Uldall, 1957: 9f). This austerity made glossematics the most radical version of structuralism and placed extraordinary demands on the linguist’s powers of structural analysis. Not surprisingly, glossematics offered few attempts to describe real languages (e.g. Togeby, 1951). Nonetheless, the glossematists Hjelmslev (1943) and Uldall (1957) recognized that the text is systemic (cf. Gülich & Raible, 1977: 92ff; 1.4.5).

 1.10 In the United States, structuralism came under the dominance of physicalism in science (see II.2.25). American research, far from working through difficult philosophical abstractions, sought language in observable events in order to describe it in “physical terms” (Bloomfield, 1936; II.2.1ff). Again, the dilemma arose of knowing what “observable events” mean in communication. Though once a pupil of Sapir’s (1921) “mentalist” approach, Bloomfield (1933: 162f) averred that linguistics must “always start from the form and not from the meaning,” because “meanings cannot be defined in terms of our science.” Meaning resulted from the “speaker’s {48} situation” and the “hearer’s response”; hence, “a scientifically accurate definition of meaning for every form in a language” would require “a scientifically accurate knowledge of everything in a speaker’s world” (1933: 139; cf. IV2.65). So Bloomfield (1933: 77) elected to use only “differential meaning” — a tenet fully in line with the structuralist notion of opposition (II.1.2.3). The investigator would proceed by asking “informants” (naive native speakers) whether or not two linguistic forms mean the same thing. Bloomfield’s successors (e.g. Bloch, 1948; Harris, 1951; Fries, 1954; even Chomsky, 1957) felt that if meaning could not be exactly defined, then it should not even be consulted in the study of language. Harris (1951: 3) proposed instead to found linguistic inquiry wholly on “the distribution or arrangement within the flow of speech of some parts or features relative to each other.” Meaning was to be used only as a “short-cut,” under the condition that a purely formal analysis of distributions would eventually arrive at the same results.

   1.11 This reduction of language to physical events and formal patterns could not be upheld in practice. Bloomfield’s followers “bypassed the mention of meaning without bypassing the use of meaning” (Pike, 1967: 61). Linguists must consult meaning in order to decide which “parts or features” in the “flow of speech” should be studied in their distribution; which pairs of items should be presented to informants; and so on. (For example, linguists would isolate ‘imp-’ as a segment in ‘impish’, but not in ‘impious’, although ‘-ish’ and ‘-ious’ are both common adjective endings; and ‘impart’ and ‘implore’ would never be construed as the ‘art’ and the ‘lore’ produced by some ‘imp’-precisely because linguists know what the words reasonably mean. No purely formal analysis, however involved, not even listing every known use of all English words beginning with ‘imp-’, forces us to the conclusion we arrive at so effortlessly by relying on meaning; the “short-cut” (I.1.10) is the only road. Hence, the interesting object for study would be the cognitive processes of linguists or informants who discuss, analyse, or give opinions about language. These processes determine the value, consistency, and reliability of the whole enterprise and would reveal whether agreement or disagreement is systemic (in the sense of II. 1. 2.3) or purely accidental.

  1.12 Morphology closely resembled phonology, but dealt with the forms, rather than the sounds, of words (cf. de Courtenay, 1972 [1888]; Sapir, 1921; Nida, 1947) (cf. II. 1.6).¹ [The first account of “morphology” in linguistics appears to be that of de Courtenay in the 1880’s. For a time, this area was eclipsed by the great success of phonology, until its importance for fieldwork brought it back into prominence, thanks to Sapir, Nida, Pike, and their co-workers. To limit the cataloguing, “morphemes” were eventually restricted to function words and affixes, while the content words were called “lexemes” (cf. Weinrich, 1976). An unresolved difficulty in the “synchronic” approach (II. 1.2. 1) is that many morphemes in English are recognizable as word-parts only from a historical standpoint, e.g., ‘en-’ in ‘enchant’, or ‘-fy’ in ‘satisfy’. Again, the structure varies according to who is doing the analysis, a professional linguist or an everyday language user.] In analogy to a “phoneme” (II. 1.5), {49} a morpheme was defined as the smallest unit of form that could determine meaning. Every word should contain at least one morpheme, plus any morphemes made of affixes: parts added either before (prefixes), inside (infixes), or after (suffixes) the word stem. Hence, a word like ‘unbeliever’ would contain three “morphemes.” An utterance would be a sequence of morphemes, and hence a framework of slots that could be occupied by given classes of fillers. For example, the subject slot of the sentence takes the filler class “noun phrase.” The distribution of a language element, i.e., the sum of all the slots it can fill (Harris, 1951; Gleason, 1955; cf. II. 1. I 0), should give an exhaustive account of its use and meaning. This postulate is essentially potent (Beaugrande, 1980a: 7), but it is hardly practicable to assemble all the occurrences of a word or word-part. Linguists had to be content with a few occurrences which they (or their informants) considered representative by virtue of knowing the language.

    1.13 A popular structuralist approach was immediate constituent analysis (cf. Wells, 1947b; Fries, 1952; Gleason, 1955; Hockett, 1958). A sentence was dissected into steadily smaller pieces until every unit was classified as a part (I.e. an “immediate constituent”) of a larger unit. To start, the sentence would be cut into clauses, and each clause cut into subject vs. predicate, which raises porblems That all-time linguists’ evergreen in (4) is easy to dissect, whereas in (4a) the predicate is partly before and partly after the subject:

             NP             VP

 (4) The man | hit the ball.

       Art    n        v  art    n

                   VP                     NP            VP

 (4a) In the ninth inning | the man | hit the ball

      prep art mod     n        art   n       v  art    n

KEY art: article; mod: modifier; n: noun; NP: Noun Phrase; prep: preposition; v: verb; VP: Verb Phrase

As we see, the method works fine when grammatical units are continuous, i.e., if their constituents are directly adjacent (IV 1.3), but leads to difficulties if they are discontinuous. Though most linguists asserted that speech was primary over writing (e.g. Saussure, 1916; Bloomfield, 1933; V.3.1), {50} and the success of phonology (II. 1.6) firmly anchored speech sounds in the theoretical layout of the discipline, constituent analysis usually took the short-cut of using written sentences already segmented by the investigator into written words. This discrepancy was later far more pronounced in the terminology of transformational grammar (e.g., “right-branching” vs. “left-branching” is oddly metaphorical unless sentences are written down) (cf. II.3.13). Indeed, the written modality is generally better than speech for defining the “sentence” (cf. II.3.13; IV.2.17; V2.18, 30, 33, 45; V.3.7ff; VI.1.10).

    1.14 Despite these difficulties in both theory and practice (I.1.11-13), American structuralism made important contributions to language exploration by conducting fieldwork on remote languages about which very little was known in wider circles. In order to record and describe a language without a grammar or dictionary (and sometimes without even a bilingual interpreter) the linguist had to take account of communicative and cultural settings. The tagmemics developed by Pike and his associates (cf. Pike, 1967; Pike & Pike, 1975; also Longacre, 1964, 1976) expressly went beyond sentence patterns to apply the slot-and-filler approach (I. 1. 12) to human behavior at large (II.2.25). The “tagmeme” itself was defined as a slot plus the class of items that could fill the slot. It made no difference if the slot appeared in a sentence or in a football game. The linguist was to be a “human observer” who “demands the right of multiple perspective, the recognition of parallel, simultaneous reaction to crisscrossing, intersecting vectors of experience, mental tools, values, and psychological pre-sets” (Pike, 1967: 10). ¹[Unfortunately, the slot-and-filler approach is especially unsuited to representing the “simultaneous” processing envisioned here (cf. IV. 1.7).] This broad scope made tagmemics the most successful method for linguistic fieldwork. Under the auspices of the Summer Institute of Linguistics, the Pikes and their students described literally hundreds of little-known languages, many of them already disappearing. However, the slot-and-filler method was too narrow a basis for a “unified theory of the structure of human behavior.” It fails to emphasize communicative participants’ intentions, plans, and cognitive processes responsible for patterns of behavior. In practice, the meaning and purpose of discourse were informally supplied by the fieldworkers themselves (cf. I.1.3; II.1.3), who relied on commonsense knowledge about social interaction. (I.15). Over the years, linguistic structuralism covered word sounds (phonology), word forms (morphology), and word order (syntax). These domains were generally called LEVELS (cf. I.4.2). Some early researchers (e.g. Hockett, 1942: 20; Moulton, 1947: 225; Trager & Smith, 1951: 81; Hall & Trager, 1953: 57) insisted that the levels be independently studied and described. A complete account of language would be the sum of the descriptions of all its levels. However, the actual result was an extreme fragmentation {51} (cf. I.1.3) — language taken to pieces until its multi-level unity in communication faded from view altogether. Too much after-the-fact analysis obscured the synthesis that creates language samples in the first place (cf. 1.4.3, 14). Predictably, the ban on level-mixing soon evaporated, at least in practice. Yet a preference for "modular" research on language has only recently yielded to an "interactive" approach (cf. III.1.27; III.2.3ff).

  1.16 Syntax became the decisive testing grounds for structuralism. To see why this level resisted a transfer of methods from phonology and morphology, we must bear in mind that a taxonomy is expected to offer an exhaustive catalog of elements in a system (1.4.3). This requirement posed no obstacle to phonology, because the sound systems of languages are fairly compact. In morphology, most languages have, besides their basic vocabulary of content words, a limited number of function words and/or affixes (cf. I.4.6) But in syntax, the feasibility of an exhaustive catalog of phrases and sentences seemed very doubtful. They might well be too numerous for a linguist to list.

  1.17 This critical impasse was resolved with Chomsky's (1957) notion of "transformation" borrowed from Zellig S. Harris (1952). Harris's earlier (and barely noticed) paper had suggested using transformations for distilling out "equivalent distributions of items in a discourse.¹ [For example, Harris (1952: 27) "transformed" 'satisfied customers' into 'customers are satisfied' to make the phrase "equivalent" to 'You will be satisfied’] Chomsky's theory postulated a small set of rules to make basic sentences, plus a larger set of rules to "transform" basic patterns into any other possible sentence of the language. This solution eliminated the need for an exhaustive catalog of sentence patterns while preserving the characteristic scientific ambitions of linguistics: well-structuredness, formality, and abstraction. Any sample sentence could be described by stating the patterns and rules which would “generate" it, that is, assign to it a structural description (cf. II.3.5, 9).

   1.18 It was probably inevitable that this new direction, though temporarily upholding structuralism, would eventually transcend the confines of the latter. Research was shifting from the structures themselves over to structure-building operations. Naturally, the question of whether the operations of the grammar were also those carried out by speakers and hearers could not remain unnoticed indefinitely. The field of "psycholinguistics" for testing the correspondence between human processes and linguistic theories was already well on its way (cf. Osgood & Sebeok [Eds.], 1954). The contact of linguistic structuralism with other aspects of human behavior and mind profoundly influenced the direction of language research. The next sections of this chapter retrace that evolution.

   1. 19 Structuralism had its weaknesses, and its critics have been numerous. Yet we should not overlook its substantial contributions. The rigorous definition of theoretical language units, and of procedures for discovering {52} those units, helped to render language research consistent, reliable, and cumulative. The systemic nature of language came into new focus, along with the role of differentiation within systems. If structuralism failed to recognize the contribution of human processing to the structures of language, it was only following the spirit of the times (I.1.3f). Placed in a proper context, the structuralist heritage can continue to serve us in future explorations.  

2. PHYSICALISM AND BEHAVIORISM

  2.1 The prestige of physics as a science (I.I.Iff) understandably encouraged researchers to treat language as an agglomerate of physical mechanisms that can be externally observed and measured — an outlook we can call PHYSICALISM. The philosophical direction known as “positivism” (cf. John Stuart Mill, 1843) had in part prefigured this outlook. Karl Spencer Lashley, at a 1948 symposium fittingly titled “Cerebral Mechanisms Of Behavior” (proceedings in Jeffress [Ed.], 1951), stated the credo: “Our common meeting ground is the faith to which we all subscribe, I believe, that the phenomena of behavior and mind are ultimately describable in the concepts of the mathematical and physical sciences.” Willard van Ormen Quine (1971: 142) asserted in a 1968 lecture “that knowledge, mind, and meaning are part of the same world that they have to do with, and that they are to be studied in the same empirical spirit that animates the natural sciences.” The commonsense dictionary definitions cited in I.1.1 imply that "science” itself is pre-empted, at least in America, by the “natural sciences” that “deal with objectively measurable phenomena.”

   2.2 It was thus not surprising that American linguistics and psychology sought to gain credibility by looking for physical events. Margaret Floy Washburn ( 916: 11f) depicted speech as a “successive movement system,” i.e., as “a combination of movements so linked together that the stimulus furnished by the actual performance of certain movements is required to bring about other movements.” John Broadus Watson (1920) viewed thought as inaudible movements of the vocal organs, and Louis Max (1937) saw consciousness as “motor activity.” Agnes Thorsen (1925) concluded that thought processes could be detected by studying tongue movements (cf. VI. 31). B[urrhus[ F[rederick] Skinner has, for almost 50 years, advertised his “conditioning” theory as scientific because its “results may be formulated in centimeters, grams and seconds” (1982: 48) — as iof science were a part of the Olympic Games.

   2.3 Physicalist approaches tend in two directions. The first direction is to reduce all human actions, including those of cognition and communication {53}  to detectable body functions, notably those of the nervous system. We might strive to record which parts of the brain perform given operations, consulting (a) patterns of excitation measured with an electroencephalograph (a brain-wave recording device); (b) concentrations of blood flow measured with positron emission tomagraphy (a technique for displaying proportions of radioactively labeled substances in the body); (c) dysfunctions resulting from brain damage, e.g. aphasia (language impairment caused by a lesion or tumor); and so forth. These directions have developed along with recent advances in neurology and neuroscience, but fundamental obstacles persist:

 2.3.1 The correspondences between patterns of neural excitation and language events are not clearly defined. A recording from an electroencephalograph does not by itself suffice to reveal the organization of a specific language event. The location of the neural activity is also inconclusive. The finding that language is dominantly centered in the left hemisphere of most people’s brains is too simple to yield a theory of discourse. Glassner (1981: 152) found from electroencephalographs that creative (“exploratory”) text production causes higher activity in the right hemisphere, while the reporting of familiar material has the same effect on the left. Yet this finding does not provide us with a detailed theoretical model of writing.

 2.3.2 Recorded activities of the nervous system could hardly be matched up with the many cognitive and communicative events that are not observable. The recordings themselves are the only evidence. We cannot watch, for example, a writer’s brain considering several language options and eventually selecting one of them. We couldn’t use the pattern of brain waves or blood flow to identify the options and the selective processes. We have only the final word choice, plus the writer’s introspective self-commentary evidence after the fact, not of the fact.

 2.3.3 The effects of brain damage are typically gross, yielding dysfunctions mostly in large complexes of language behavior. The location of the damage accounts for some differences. Damage to Broca’s area (in the inferior frontal gyrus of the language-dominant brain hemisphere) impairs text production, while damage to Wernicke’s area (in the posterior part of the temporal lobe of the language-dominant brain hemisphere) impairs reception (cf. Broca, 1865; Wernicke, 1874). But text production and reception are themselves constituted by elaborate complexes of activity that could be sorted out only if impairments were much more specific.

    2.3.4 Finally, the human nervous system is enormously adaptive and flexible, not just for different persons, but also for the same person at different times. Early researchers (e.g. Eberstaller, 1890; Cunningham, 1892) already found that the organization of brain surfaces is far from identical {54} for everybody — a finding borne out by recent studies of language localization (Whitaker & Selnes, 1975).

  2.4 For all these reasons, it seems implausible that a theory of cognition and communication can be strictly derived from neurological observations. Already proposed theoretical models of language operations can be tested, but hardly created, by observing the activities of the nerves. The differences between brain hemispheres might account for aptitudes of language learning during maturity (cf. Lenneberg, 1967; Scovel, 1969); for inclinations toward induction vs. deduction (Winterowd, 1979); for creative vs. stereotyped expression (Glassner, 1981, 1982; cf. II.2.3.1); or for sound/letter correspondences in spelling (V1.26). But only a theoretical model of text processing can effectually suggest why any of this should be so.

   2.5 The other direction for a physicalist approach to cognition and communication would be to consider the human a simple mechanism determined by the environment and its visible objects. Manipulating these things should then provide all needful insights about the human. This outlook was the basis for the research paradigm known as BEHAVIORISM (due especially to Watson, 1920; Pavlov, 1927; Skinner, 1938). Its central project was to observe an "organism" receiving an external STIMULUS and making a RESPONSE. By REINFORCING the response with either a positive or a negative reaction, the investigator could CONDITION the organism to behave in desired patterns. The conditioning could make the organism select a given action and perform it with FREQUENCY (number of acts and rate of occurrence) and FLUENCY (the smoothness of transitions among stages of the act). in accord with the ambitions of the time, the stimulus-response mechanism was claimed to follow a strict "law" of effect (Skinner, 1938; Keller & Schoenfeld, 1950), like the "laws" of the natural sciences, notably physics (cf. I.1.1; II.2.1).

   2.6 LEARNING takes place if the conditioned organism increases the frequency and fluency of the desired action. TRANSFER occurs if the activities of the test session carry over to typical behavior (cf. Greeno, James, DaPolito, & Poison, 1978). A DISTRACTOR is a task that interferes with actions and learning. In classical conditioning, the experimenter takes the initiative by stimulating the organism to do something, and replaces the organism's natural response with an arbitrarily induced response (cf. Pavlov, 1927, and his famous dogs). In operant conditioning, on the other {55} hand, the experimenter waits for the organism to display some bit of naive behavior and then uses reinforcement to shape and control (cf. Skinner, 1938, 1953, 1957; Bachrach [Ed.], 1962; Wolpe, Salter, & Reyna [Eds.], 1964; Krasner & Ulimann [Eds.], 1965).

   2.7 The clearest successes of behavioral conditioning were obtained with animals of limited reasoning abilities, such as rats, cats, dogs, and pigeons. With enough training, these animals could be taught to carry out tasks like activating levers or running mazes (see Bachrach, 1964, for one spectacular case). The animal had to be periodically rewarded (usually with food) or punished (usually with an electric shock) to show what sorts of behavior were or were not desired. The studies of “visceral learning” by Neal Miller and co-workers (cf. Jonas, 1973) suggested that even involuntary acts such as heartbeat could be shaped by expert conditioning.

   2.8 This reasonably secure body of research was then extrapolated (in the sense of I.1.12) to the study of human language. Some experimental techniques derived from conditioning became standard fixtures of research. In paired-associate learning, test persons are given a list of word-pairs to remember, then “stimulated” with a word, and required to “respond” with its paired word (cf. Murdock, 1963). The pairing can be varied according to whether the two words are similar or different in sound, word-class, meaning, and so forth (cf. Deese, 1965; Kintsch, 1968; Kintsch & Buschke, 1969). Today, the learning is attributed not to a simple strengthening of an undifferentiated association, but to the selection of some aspect of the stimulus that sets it apart from others (Greeno et al., 1978). Hence, learning word lists without context may be harder and shallower (in the sense III.1/11) of than learning from contextualized discourse (cf. Kircher, 1971; Brown, Campione, & Barclay, 1978), because the latter provides much richer means of differentiation (cf. IV.1.8f). In principle, the correlation between natural and experimental conditions is far from clear (cf. III.1.1, 3; III.2.40; IV.I.9; IV.2.16, 21; V1.13f, 19).

   2.9 The stimulus-response scheme (II.2.5) became the basis for behaviorist theories of language. Bloomfield (1933) depicted discourse as a response to an event or situation, with visible objects in the main focus (cf. II.2.23). In his twee example, Jill is stimulated to hunger upon seeing an apple tree, but alas, she is not sufficiently athletic. Instead of struggling with the fence and tree, she made a few small movements in her throat and mouth, which produced a little noise. At once, Jack began to make her reactions for her; he performed actions that were beyond Jill’s strength, and in the end, Jill got the apple (Bloomfield, 1933: 24). This reductionism converts discourse into a “few small movements” in the “throat and mouth” and “a little noise.” The cognitive processes whereby Jill formed a plan to attain her goal and constructed an appropriate utterance action, and the contexts that make the action meaningful, are ignored.  {56} The food reward reinforcing her act is the classic enticement of the animal conditioner (II.2.7).¹ [1 The intrusion of the apple tree in place of the hill-and-water-pail scheme may be a vague Biblical allusion (Genesis 3:6), though Bloomfield’s sexism impels him to give the initiative to the man rather than to the woman.] Jack’s agency is limited to “enacting Jill’s response” when stimulated by “movements” and “noise” to obtain food at the risk of breaking his crown. He could be replaced by a chimpanzee, a superior tree climber, had that animal not been eschewed by American behaviorists as too intelligent.²  [Kintsch (1977: 440f) remarks on the ominous consequences of selecting particular animals for learning experiments. Thorndike (1911) put cats in cages with levers and concluded from observations that rote repetition and trial-and-error should become the foundations for American learning theory-which they did, for decades (cf. Das et al., 1979: 3ff). Kohler’s (1925) chimpanzee showed substantial reasoning powers, but these tests were mostly ignored by American educators.

    2.10 Like most reductionist theories, behaviorism presented as general cases only those special cases it could deal with (I.1.3). It also obscured the difference between external and internal actions, between performance and knowledge (cf. I.2.8.2; I.3.3, 11; III.3.2.2; IV.1.12, 48; V1.14f; VI.3.22). Some situations do stimulate people to perform discourse as a response — a special case I call situation monitoring (I.4.11.5). But this monitoring is normally mediated by the cognitive organization of the text producer and by the exigencies of text production (what to mention, how to describe it, what motives to follow, etc.) (cf. I.4.6). If discourse is mediated by the intent to steer the situation toward one’s goals, we have situation management (I.4.11.5; cf. Beaugrande & Dressier, 1981: 163ff). By trying to stretch monitoring to include management, behaviorist language theories like Bloomfield’s deny the importance of plans, goals, and world-models in human communication. No explanation can be found for how humans decide what environmental events are relevant, and what response might be appropriate.

 2.11 Hobart Mowrer (1954) thought of the sentence as a conditioning device: the subject is the conditioned stimulus and the predicate is the unconditioned one.  Understanding the sentence means “pairing” the subject and predicate in a new “association” (cf. II.2.8). If a text producer sees Tom stealing money, the utterance

 (5) Tom is a thief.

 takes “Tom” as a known point of departure and impels the hearer to pair “Tom” with “thief.” The effort required to learn this new association varies according to the text receiver’s situational role, e.g. Tom’s mother vs. a policeman (Osgood, 1963: 748). With only the stealing event in view, Mowrer’s example is pure situation monitoring. Still, Mowrer recognized the distinction between given vs. new (here, context vs. observed event) in a sentence or utterance (IV.2.63), whereas Bloomfield did not. However, having visible {57} objects be the “stimulus” for the utterance-the favorite example for behaviorists (cf. II.2.5, 9, 23; IV2.56) is just a special case. It’s hard to see any stimulus-response scheme in a text such as:

 (6) What a piece of work is a man! How noble in reason! How infinite in faculties! In form and moving how admirable! In action how like an angel! In apprehension how like a god! (Hamlet, II, ii)

 A strict behaviorist would predict that Hamlet should say:

 (6a) What a piece of conditioned behavior is a man! How observable in response! How law-governed in faculties! In form and moving how statistical! In action how like a mechanism! In apprehension how like a rat!

 And even (6a) is still not a direct “response” to a situation, but a statement mediated by the goal of propagating a reductive theory.

 2.12 The most ambitious behaviorist theory of language was Skinner’s (1957) account of “verbal behavior” derived from his work with operant conditioning (II.2.6) of animals (Skinner, 1938, 1953). He classified utterances as types of “verbal operants.” A “mand” was “a verbal operant in which the response is reinforced by a characteristic consequence,” e.g., the “mand” ‘Pass the salt’ being reinforced by getting the salt (Skinner, 1957: 350. A “tact” was “a verbal operant in which a response of a given form is evoked (or at least strengthened) by a particular object or event or a property of an object or event,” e.g., ‘How do you do?’ for a greeting and ‘Thank you’ for a service (1957: 81f). “Mands” and “tacts” (derived by dismantling the words “commands” and “contacts”) were central to Skinner’s approach because they are special cases where discourse actions fit the stimulus-response scheme most closely.

 2.13 To appreciate how such a bias abrogates communicative relevance (II.2.10), consider Skinner’s (1957: 253f) droll list of “all the available techniques” for “evoking verbal behavior,” e.g., for eliciting the word ‘pencil’. We can: (a) “mand the response by saying to the subject ‘Please say “pencil’”; (b) “make sure that no pencil or writing instrument is available,” and offer the subject “a handsome reward for a recognizable picture of a cat”; (c) “provide echoic stimuli (a phonograph in the background occasionally says ‘pencil’) and textual stimuli (signs on the wall read ‘PENCIL’)”; (d) “produce interverbal responses: the phonograph occasionally says ‘pen and —; and there are other signs reading ‘PEN AND ____; or puts very large or unusual pencil in an unusual place clearly in sight-say, half-submerged in a large aquarium or floating freely in the air near the ceiling.” The word ‘pencil’ seems to be a “mand” in (a) and (b), and a “tact” in the other cases, especially (e). Only (b) has any motivational relevance to the speaker’s goals. “Echoic stimulation” and “interverbal responses” are special cases with the least resemblance to creative communication. Though methods (a) through (c) should elicit the word ‘pencil’ from the most rat-brained test-person (unless the latter, not surprisingly, runs away, believing that the experimenter has lost his mind and may be dangerous), they are far from hardly represent the natural conditions for saying ‘pencil’, let alone for composing utterances like Hamlet’s in (6). A composition teacher who tried to “evoke” student essays with such methods could reach the age of retirement before one paper got finished.

   2.14 Most of Skinner’s examples were either one-word utterances like ‘pencil’ or else standing phrases in “mands” and “tacts” (cf II.2.12). He barely touched on the complex activities of “textual behavior.” Since physicalists stress the environment over the human being (II.2.5), Skinner (1957: 31 1) saw the text producer as a mere “interested bystander” impelled not by creativity, but by the stimulus-response mechanism. In Skinner’s (1957: 346) view, a text might be (a) made of “memorized sentences emitted as purely interverbal sequences”; (b) “reproduced as echoic of textual behavior”; or (c) “blended” from “a few key responses with stock patterns.” To engage in “composition,” the text producer could use: (a) “the essential operants,” (b) “the interverbals possibly arising from these operants in the course of emission (often composing thematic groups of responses),” and (c) “the autoclitic framework” (1957: 353f). Thus, the formatting of texts depended “autoclities”: “responses within behavior which is based upon or depends upon other verbal behavior,” such as: “informing the listener of the kind of verbal operant it accompanies”; “describing the state of strength of a response”; “indicating the emotional or motivational condition of the speaker”; and so on (1957: 315f). Grammar and syntax would be “autoclitic processes” that signal the relationships among the “manded” responses to “a particular object or event or a property of an object or event” (II.2.12). This account implied that the “key” content words (nouns, verbs, and modifiers) are selected before function words, and that the main vocabulary is chosen before the phrasing to put it in (cf. III.2.30; IV.2.10, 15, 26ff).

   2.15 Still more unsatisfying and reductive was Skinner’s (1957: 438fn) treatment of “thinking,” “verbal fantasy,” and “rationalization.” Here, he could only feebly assert that text producers are “automatically reinforcing themselves.” Skinner thus abandoned his only firm ground: the externalized stimulus-response scheme. Chomsky (1959: 37f) objected here “that a person can be reinforced though he emits no response at all and the reinforcing ‘stimulus’ need not impinge upon the ‘reinforced person’ (it is sufficient that it be imagined or hoped for.” Skinner had finally tossed out the evidence on which the behaviorist approach had been originally founded. The extrapolation from the animal-conditioning laboratory to human behavior became a leap of faith, a palpable metaphor, or an “outrageous hyperbole” (Lazarus, 1982: 43) (cf. I.1.3, 12). Without a unifying theory of communication as such, Skinner could only transpose the notions of operant conditioning wherever he saw a vague resemblance between discourse events and stimulus-response chains. Chomsky (1959: 430) concluded: {59}

 Skinner […] utilizes experimental results as evidence for the scientific character of his system of behavior, and analogic guesses (formulated in terms of a metaphoric extension of the technical vocabulary of the laboratory) as evidence for its scope. This creates the illusion of a rigorous scientific theory with a very broad scope, although in fact the terms used in the description of real-life and of laboratory behavior may be mere homonyms, with at most a vague similarity of meaning.

 2.16 In the behaviorist view, the proper place to look for meaning was not the human mind, but the environment, (II.2.5). Bloomfield (1933: 144) equated meaning with the “speaker’s situation” and the “hearer’s response” (II.1.10; II.2.9). Skinner’s (1957) “tacts and “mands” also discount meaning itself in favor of its external consequences. “Tacts,” being evoked by objects, events, and their properties (II.2.12), fall under the general notion of REFERENCE: the capacity of language to point to objects and situations in a (real or envisioned) world (I.4.11.2; III.2.22). This notion was central to the philosophical trend of positivism, with its credo that natural science was the only legitimate vehicle for knowing about the world (II.2. 1). Thus, the meaning of a statement became equivalent to the state of the world that makes it true or false.¹ [In philosophy, this outlook has been called “truth-conditional semantics,”  and variously applied in “model-theoretical”, “extensional,” and “truth-functional semantics” (cf. Lyons, 1976, for discussion). The analogy between world/statement vs. stimulus/response made the same fixation on “truth” appears like a common shibboleth of behaviorism and mentalism.]  A unification of behaviorism with the theory of reference was indeed undertaken by Willard van Ormen Quine (1960, 1968, 1971). His approach depicted “knowing a word” as having “a phonetic” part” “achieved by observing and imitating other people’s behavior,¹ [The “phonetic part” would seem to be the simple word-sounds, in contrast to the “phonemic” value of those sounds in a language system. Later in the same paper, Quine switches to the “phonemic” standpoint (1971: 152).] and a “semantic part” whose “paradigm case” is the word’s reference “to some visual object” (Quine, 1971: 143). A comparison with Saussure’s “acoustic image” and “concept” (II. 1.2.2) at first suggests that Quine has regressed to the pre-structuralist standpoint in which words are names for things.

    2.17 However, Quine mistrusted this approach to reference because the relationship between a word and the thing it stands for is a mental construct (cf. II.3.1). He wanted to replace the things themselves with people’s responses to things. He asserted that “there are no meanings, nor likenesses nor distinctions of meaning, beyond what are implicit in people’s dispositions to overt behavior” (1971: 143). We best understand speakers by “seeing what is stimulating” them (ibid.). The clearest case would be “direct ostension”: “pointing your finger” at “an opaque surface”; if you point in order to illustrate not the object itself, but its abstract properties (e.g. pointing at grass to illustrate ‘green’), you engage in “deferred ostension” (1971: 148f). Reference (surprise!) becomes “darker” when we move away from ostension  {60} toward “denotation” (the capacity of words to mean things). Quine (1971: 153) hinted at the hope that all other uses of language are a “grand and ingenious permutation” of the basic “denotation” of “familiar terms.” If we explain the meaning of one term by using another term, and the meaning of the latter by using still another term, we are threatened by an infinite regress “ we could eventually break by pointing to an object” (1971: 153).

    2.18 Ultimately, Quine conceded that reference is (for his approach, at least) “inscrutable,” because “reference is nonsense except relative to a coordinate system” (1971: 153). What a word means depends on its place in a system of meaning (cf. I.4.5; III.2.22). Quine (1968) therefore had to be content with an “ontological relativity” that permitted suspiciously many “ingenious permutations” and “compensatory adjustments in the interpretations” of words. Here, Quine seemed to unintentionally swing into line with the structuralist conception of “systems of opposition” that make language meaningful (II.1.2.3). In fact, however, he demanded not a theory of abstract systems, but “a relational theory of objects” to explore “not what the objects of a theory are, absolutely speaking, but how one theory of objects is interpretable or reinterpretable in another” (1971: 153f). The meanings of words would be explained by a, theory of how objects are relatable in pace,” “position,” “velocity,” etc. in a “co-ordinate system” (ibid.), such that people can translate object systems into language systems, and, on that basis, from one person’s language system to another person’s (cf. Osgood’s scheme in II.2.23).

    2.19 Quine’s demand clearly transcended the narrow framework of orthodox behaviorism set down by Skinner (1957). In effect, Quine admitted that people’s “dispositions to overt behavior” do not arise from a simple stimulus-response chain, but from the position of any stimulus event or object in a relational system. Experiments have indeed shown that the language used to refer to visible objects depends on what the text producer considers necessary and relevant in the current context, for example: how many objects are similar or different, and in what ways (Krauss & Weinheimer, 1967; Olson, 1970; Osgood, 1971); how an object is located in a system of classes subsuming, or subsumed by, other classes (Collins & Quillian, 1972; Smith, Shoben, & Rips, 1974; Osgood, May, & Miron, 1975); what traits of an object are thought more or less essential to its nature (Rosch & Mervis, 1975; Rosch, 1977); and so on. Incorporating such factors makes behaviorist approaches to communication seem less narrow, though still incomplete.

    2.20 Charles E. Osgood refined his model with the notion of MEDIATION — in my definition, the degree to which a processor imposes its own {61} outlook (organization, priorities, beliefs) upon a situation (cf. I.1.7; I.4.6; I.4.11.5; II.2.10). Osgood’s approach interposed mediation between the stimulus and its response, for instance, to explain why a person doesn’t react to a word the same way as to the thing the word refers to: “the mediating reaction to the sign is not the same as the reaction to the significate, but rather consists of those most readily conditionable, least effortful, and least interfering components of the total original reaction” (Osgood, 1963: 740). Osgood comes close to saying that word meanings can be dynamically constituted on each occasion (I.4.6); but he would attribute the effect to behavioral impulses, whereas cognitive decision-making gives to human processing at least as much initiative as to the environment.

   2.21 Osgood’s mediation theory sought to resolve the matter of initiative by introducing PROBABILITIES into the stimulus-response mechanism in place of the natural “laws” envisioned by orthodox behaviorists (II.2.5). He argued that “decision in behavior is simply selection of the momentarily most probable alternative”; and that “control over behavior is simply the way in which combination and patterning” modify “the momentary probabilities” of the alternatives (1963: 741). Lounsbury (1954) extended Hull’s (1943) work on habits and suggested that utterances are controlled by “habit family hierarchies” determining what actions or reactions are probable. Probabilities can be “reshuffled” in context, so that the same stimulus can elicit different responses. If this view holds, each successive word in an utterance should be more probable than the one before it, as context continually narrows down the likely choices (cf. Goldman-Eisler, 1958a; cf. IV.1.1, IV2.15). However, this strict correlation was not found (Tannenbaum, Williams, & Hillier, 1965). In fact, the end stretches of utterances are more often devoted to new or unexpected material (IV,2.63f).

    2.22 Still, Osgood’s model allowed a more refined outlook on discourse than orthodox behaviorist ones. He broke down the simple stimulus-response concept into three “stages” that were carried out during both “decoding” and “encoding” (Osgood, 1963: 740). A presented text would be first registered as a “pattern of sensory signals,” then “integrated” (or “chunked”, I.4.5), and finally assigned a “representation” (or meaning); to produce a reply text, an “intention” to be “meaningful” would be relayed to the “skills” of “integration” and from there to “execution” (1963: 741). Hence, the whole process is both psychologically and physiologically reversible: production and reception are mirror images of each other (but cf. III.1.9). The person’s vocabulary is stored in a huge “word form pool,” a catalogue of “surface expressions” (the forms of words, not their meanings).¹ [On how word forms are accessed, see V.1.34ff.]These forms are passed to a “semantic key sort” that decides what meaning is most probable, where “meaning” is defined by Osgood as {62} a “bundle of simultaneous semantic features.² [“Semantic features,” an ingenious extrapolation from sound features (II.1.5f), enjoyed some Vogue, especially as a way to make static objects out of word meanings. But nobody could devise a catalog of them for more than a handful of contrived examples (usually objects like chairs, whose parts or traits made obvious “features”), and the “ +/-mania soon reached bizarre extremes (cf. II 1).] “In a sentence (or text), these feature bundles are integrated by a cognitive mixer” that establishes “congruity” of meaning and eventually leads to the “total response.” For example, to understand Mowrer’s sentence (II.2. I 1) (5) ‘Tom is a thief’ a “semantic fusion” puts ‘Tom’ together with ‘thief’; if we had: (5a) ‘Tom is not a thief’, a “semantic dissolution” would be done instead (Osgood, 1963: 748). The more complex a text, the more “mixing” is needed for “fusion” and “dissolution.” The “features” in these “bundles” were explored by an experimental technique called “semantic differential,” where test persons rated words on scales like “light vs. dark,” or “pleasant vs. unpleasant” (Osgood, Suci, & Tannenbaum, 1957).³ [Though interesting, the “semantic differential” approach is not a theory of meanings so much as a theory of responses to meanings (connotations, attitudes, emotions).]

   2.23 In his early work, Osgood (1963: 741) resolutely denied the need for such notions as “cognition,” “purpose,” and “volition,” and stretched his concept of “probabilities” to cover them all (cf. II.2.21). He carefully reduced all meanings to “past experience with behavioral outcomes” (ibid.). Later, Osgood (1971: 498ff) propounded a “non-linguistic cognitive system” “shared” both by “perceptual signs” and by “linguistic signs.” This system would be capable of the “representational mediation” that allows “perceptuomotor behavior” to engender “symbolic processes.” However, the treatment of meaning was still in terms of “patterns of stimulation, including previously learned signs, which regularly and reliably produce distinctive patterns of behavior” (1971: 523; cf. Quine’s scheme in II.2.18). He sought to prove this view with his well-known experiment where students watched him put or move objects on a table and “simply described” the events to a fictitious “six year old boy just outside the door” (Osgood, 1971: 497). His students produced very similar sentences to describe a given event, depending on the visibility, salience, and similarity of the objects (e.g. moving rather than stationary) (cf. II.2.19; IV.2.56ff).

   2.24 By seeking the basis of language in perceptual systems, Osgood in effect adopted the empirical correlate of Quine’s philosophical account. Though both scholars strove to found meaning upon “overt behavior” and on responses to visible objects (Quine, 1971: 143; Osgood, 1971: 523, 497), they {63} both had to expand the behaviorist-physicalist framework by postulating complex systems that determined the real significance of external events or objects. Ineluctably, the “stimulus/response” mechanism became a commonplace (and hence unnoticed) metaphor for causality. For the sake of orthodoxy, researchers would say “stimulate” instead of “present,” “make happen,” “cause,” “enable,” and so on. Such usage often had little relation to the original context of animal-conditioning (cf. II.2.7, 12, 15).

    2.25 Among structuralist trends, American descriptivism and tagmemics proved the most congenial to behaviorism, e.g., in Bloomfield’s account of meaning (II.1.10; II.2.9, 16). Pike’s tagmemics leaned toward behaviorism as a vehicle for fieldwork (II.1.14). The fieldworker immediately sees “the interdependence of language and non-language behavior” and must “correlate particular sequences of sounds” with the “particular environment in which they are uttered and with the particular non-language activity which is going on” (Pike, 1967: 29). However, this “unified” approach was rejected by most linguists in both European structuralism and American mentalism, mainly because they hoped to separate “pure” language from everything else. Thus, language as behavior sustained no common ground with language as an abstract system, because the two reductions of language went off to the opposite extremes. Unfortunately, this divergence prevented a dialogue or integration that would have benefited both sides.

   2.26 Behaviorism had a profound impact on education and child development programs.¹ [All behaviorist educational programs inherit the contradiction between making learners self-reliant and creative vs. applying a theory best suited for making them pliable and “conditioned”. The tactics are well-known: rote repetition, punishment (“negative reinforcement”), and a simplistic equation of observed behavior, especially errors, with the child’s true abilities (cf. I.3.3, 9).]  Behaviorists found in children’s language good support for their views. Skinner’s (1957) “mands” and “tacts” (II.2.12), Quine’s (1960) progression from pointing to objects over to using language to refer (II.2.17), and Osgood’s (1971) “conditions for sign learning,” were all illustrated with children’s behavior. The motive is clear: children fit the behaviorist world-view much better than adults, allowing this extrapolation: animals => children => all humans. For instance, the pairing of words with visible objects (a type of “paired-associate learning,” II.2.8) is really feasible only for a young child with a minimal vocabulary (Gough & Hillinger, 1980: 183).

   2.27 It follows that behaviorist learning techniques work best for training young children (or the mentally retarded) in rudimentary, emergent skills, when complexity has not yet imposed severe demands on cognitive organization. Later, the techniques become steadily less helpful, because they ignore more elaborated processing capacities in favor of wasteful, eventually unmanageable methods of rote repetition and reinforcement. Whether anything useful is learned and retained from behaviorist schooling {64} is mainly  pure chance, as long as the educators don’t understand the processes of cognitive reorganization that, according to Mandler (1979), are the necessary preconditions for learning from repeated experiences (cf. II.3.31; III.1.28; IV.2.38). Thus, behaviorist education, like Darwinism, weeds out children for whom it doesn’t happen to work; and any inequities are rationalized as results of “genetic intelligence” and “aptitude” (I.3.Iff).

    2.28 Nonetheless, education is concerned with behavior in the project of improving human skills, including writing. It is unwise to lose sight of the fact that writing is an action. Porter (1962) criticized composition classes if they violate these basic conditions for improving behavior: (a) the behavior has consequences, (b) learning doesn’t have to be done in a single leap, and (c) overzealous punishment does not suppress behavior altogether. Similarly, Zoellner (1969) advocated his “behavioral pedagogy of composition” because writing instruction so often flouts learning theory by being intransigent toward errors, disregarding students’ “here-and-now behavior,” failing to appreciate and work with approximations, and attempting to impose conformism (cf. also Zoeliner, 1971; I.2.17ff; 1.3.7ff). Zoellner (1969) extrapolated from “operant conditioning” (II.2.6) to advise that the instructor should wait for students to “emit” a desired “bit of behavior” and should then “shape” and “reinforce” it with frequent reactions. Since behaviorism offers no standards of value or communicative relevance, Zoellner focused on fluency: the rate and ease at which students write (cf. II.2.5; II.3.21, 38). Students were to “blurt” something out”¹ [Zoellner uses the term “visceral blurt” as a facetious metaphor alluding to Miller’s “visceral learning” (II.2.7).] and instantly record it on a blackboard or drawing pad. This record serves as an approximation to be used in later stages of the writing. The student must not be asked to “think,” lest there ensue “an immediate drop in the density and variety of external behavior”: the student’s “body tenses up, he [sic] becomes motionless, his face takes on a wooden expression, and his vocalization drops to zero” (1969: 298).

    2.29 This emphasis on fluency suggested that fast, easy writing equals good writing (compare Zoellner, 1969: 270ff). Yet good writers are often not rapid or fluent (Boyle, 1969; Elbow, 1975; Lowenthal & Wason, 1977; Murray, 1978; Atlas, 1979; Odell, 1980; Wason, 1980), mostly because they expend intense effort. Zoellner extended over to writing the popular view (I.2.8.3) that fluent speech is a hallmark of literacy. However, due to differences in rates and processing conditions of the two modalities, a transfer of the rhythm and tempo of speech over to writing can be an impediment to composition as much as an aid (V.3.10-37). Student writers rush along, inserting fillers, trivialities, needless words, and restatements, just to keep moving. And since speech is not preserved, little use is made of the opportunity {65} writing offers to look back and build directly upon what was already said.

    2.30 Zoellner contended that students can talk out their message in a sustained, rapid-fire segment of sound stream” that “communicates quickly and effectively” (1969: 273), though their writing may be muddled. His appraisal of the need to transfer speech skills over to writing was sound, but he failed to spell out the steps in the transition. He had to presuppose that “the student has already thought the rhetorical problem out, and that it is only necessary to get him to say what he already knows” (1969: 299). As in operant conditioning, the experimenter/teacher begins the “shaping” only after the animal/student emits some “naive behavior” (II.2.6). For Zoellner (1969: 303f), a failure to write what one wants to say is like a pathological deviation between a brain impulse vs. a motor action, sometimes called “schizokinesis” (Gantt, 1953) — as if writers’ problems are comparable to mental illness. By decrying “thinking” (II.2.28) and taking for granted the resolving of “rhetorical problems,” Zoeliner neglected the cognitive processes that control organize, and evaluate the content of writing.

    2.31 Because the stimulus-response scheme abridges relevance and creative initiative (II.2.10, 13, 28), educational behaviorism at large deals only with gross externalizations of crucial cognitive developments (II.2.27). The status of any observed event as an indicator of underlying skills is drastically over-simplified. No doubt Zoellner himself evaluates student papers by standards wholly unrelated to operant conditioning. He publicly opposed the “behavioral objectives” that clutter up education as disguised demands for “middle-class” uniformity and “cultural coercion” (Zoellner, 1971: 231f), e.g., Lois Caffyn’s (1969: 72) goals of habitually using “preferred” language, listening eagerly and courteously, and attending community meetings and clubs. (I know I don’t do these things all that much, yet I can’t see how my writing is any the worse for it.) Yet the individualism Zoeliner (1971) favored is irreconcilable with his extrapolation from operant conditioning which, if anything, massively steers each individual toward predetermined ends -to be “conditioned” means to conform. The “shaping” can be tailored to the student, but the standards for judging what to write and how to react to it cannot be derived from the rat-training sessions Zoeliner extolled.

   2.32 Even Caffyn’s more convincing “objectives” of discussing issues, sharing experiences, and responding to fine distinctions, are mere labels for undefined cognitive and social developments. Writing is a genuine mode of human behavior, involving manifest actions, and it is organized in a real time progression (cf. Matsuhashi, 1979, 1981, 1982; Bereiter, 1980; Flower & Hayes, 1980, 1981). But conventional behaviorism is too simple and reductive to be more than a precursor for a science of composition. In future work, we must find new approaches, but we should also bear in mind some behaviorist tenets: {66}

   2.32.1 A theory of human language should be an integral part of a general theory of human behavior (Skinner, 1957; Osgood, 1963; Pike, 1967). We must keep in view the “broader field” of “the behavior of a most complex creature in contact with a world of endless variety” (Skinner, 1957: 456).

   2.32.2 “What needs to be explained is the total speech episode,” including “all relevant events in the behavior of both speaker and listener in their proper temporal order” (Skinner, 1957: 36). In this “total event,” “language behavior and non-language behavior are fused” (Pike, 1967: 26).

   2.32.3 “The listener” (or audience) is “an essential part of the situations in which verbal behavior is observed”; “the change which is thus brought about in the behavior of the listener is appropriately called ‘instruction’” (Skinner, 1957: 172, 362; cf. 1.4.6; III.2.26).

   2.32.4 Language education should be based upon “the actual here-and-now behavior of a given student or a given classroom of students” (Zoeliner, 1971: 232).

   2.32.5 Learning works best as a series of successive approximations moving steadily and confidently toward an end result (Sidman, 1962).

   2.33 Cari Frederiksen (personal communication) remarks that experimental psychology is inherently behaviorist to the extent that it observes and measures the responses of test persons to induced or manipulated events. Teaching has the same orientation to judge performance in terms of tasks and results. Thus, teaching is research (I.4.17; III.2.40; VI.3.25) by virtue of this analogy. However, we need a comprehensive model of writing that shows how behavioral evidence relates to the larger context of cognition and communication. Not until then ‘can we co-ordinate daily practices with the totality of our enterprise.

  3. MENTALISM AND TRANSFORMATIONAL GRAMMAR

  3.1 Whereas PHYSICALISM seeks explanations in observable mechanisms, MENTALISM looks to the mind and its capacities. Mentalists explore the ideas, idealizations, abstractions, intuitions, and so on, that underlie external events and behavioral dispositions. In American language research, mentalism had been represented only in a few studies (e.g. Bloomfield 1914; Sapir, 1921; Whorf 1965; [19441]. Bloomfield’s (1933, 1936) vigorous conversion to physicalism determined the course of structuralist linguistics for decades (II.1.10). Even in the study of meaning, mentalism was lamented as “vitiating” and “pernicious” (Quine, 1971: 142). Though physicalism and behaviorism were eventually recognized as inadequate, they were difficult to revise, as we saw from the efforts of Quine and Osgood (II.2.16-23). The terminology lived on, but more as a set of orthodox metaphors than as a description of human communication (II.2.15, 24). {67}

   3.2 The rehabilitation of mentalism in American linguistics and psychology came when formal logic was conspicuously introduced into theory and method. One of the early projects of positivism (II.2.1, 16) was to build a perfectly consistent and complete logical basis for mathematics and science (Russell & Whitehead, 1903). Gödel’s (1934) proof that every logical system allows true statements you can’t prove and false ones you can’t disprove put an end to that project, but not to the longings of scientists for such a system. In effect, the argumentative procedures of science continued to resemble those of axiomatic logic (survey in Stegmüller, 1976). Moreover, logic commands the respect of the general public who only vaguely understands its intents and limits (cf. Beaugrande, 1980a, 1981b). Logical languages fall far short of the flexibility, complexity, and subtlety of “natural” (i.e. non-artificial) languages like English, and fail to account for the participation of producer and receiver, and the informativity and situationality of the text. Logic presupposes well-defined classes, subclasses, and instances. Its conclusions are often so trivial and predictable that real people wouldn’t utter them in everyday communication. Hence, the use of logic as a model for human thought and language entails a different, though hardly less radical, reduction, than that of physicalism (II.2.9f), and is thus open to principled objections. A more interesting issue is how people (including philosophers) perform processes they consider “logical,” e.g., defining classes and drawing conclusions.

    3.3 In common parlance, “logic” designates reasoning procedures that emphasize truth, consistency, completeness, and causality. This kind of logic is popularly thought to be the ideal mode of thinking and communicating (cf. Olson, 1977a, 1977b). For example, composition teachers watch for violations of logic in students’ papers (I 317). However, too much stress upon logic can be inhibiting to creative learning processes whose approximations and diversities are wrongly looked upon as embarrassing. Seymour Papert (cited in McCorduck, 1979: 290ff; cf. Papert, 1971) remarks:

We all know that we have all these horrible moments of confusion when we begin a new project, that nothing looks clear and everything looks awful, that we work our way out using all sorts of odd little rules of thumb, by going down blind alleys and coming back again, and so on, but since everyone else seems to be thinking logically, or at least they claim they do, then we figure we must be the only ones in the world with such murky thought processes. We disclaim them, and make believe we think in logical, orderly ways.

The logical ideal thus plays down the intuitions, inferences, defaults, and approximations of everyday reasoning (cf. Piaget, 1976; Collins, 1978; Papert, 1980), and may mislead learners to simply give up trying. The real role of logic in cognition and communication is more likely that of a community of specialized processes (I.4.8; III.1.23); this “logic specialist” can be consulted as needed, but doesn’t dictate what a person can think, say, or mean. {68}

    3.4 The popular wishful thinking about logic helped to case the entry of mentalism into linguistics and psychology. Chomsky’s (1957) “grammar” promised to relieve a crisis in taxonomic structuralism: the unwieldy complexity and variety of language syntax (II.1.16f). He suggested that a natural language like English is composed of an infinite set of sentences, in the same way as logic permits unlimited embedding (inserting one instance of a structure inside another) and recursion (repeating a formative operation over and over) (cf. IV.2.13, 70; IV.3.37). Thus, no exhaustive catalog of surface patterns could or should be undertaken. His own system was an extrapolation from “axiomatic logic” (also called a “calculus”): a logic founded on a set of “strings” (sequences of symbols or words) called “axioms” because they are accepted a priori. The axioms are created by means of “formation rules.” All the strings of the logic can then be made by applying “derivation rules” to the axioms. Only the strings so derived count as “well-formed.” The meaning of a string is determined by applying “interpretation rules.” Only well-formed strings can be interpreted — ill-formed ones are considered meaningless. Two or more strings can be linked with “connectives”: ‘and’, ‘or’, ‘if […] then’, and ‘if and only if’ (also written ‘iff’). Whether a string so connected is as a whole “true” or “false” is decided by consulting “truth value tables.”

    3.5 At first glance, all these notions suited the purposes of logic for building formal proofs. The analogy between a logical language and a natural language also seemed reasonable. The “strings” were the sentences. The “axioms” were kernels (or “kernel sentences”). The “formation rules” were called “phrase-structure rules”, and the “derivation rules” were “transformational rules”. A “well-formed string” would be a “grammatical sentence” (or “well-formed sentence”). The logic itself was the grammar for the language in question, e.g. English. The “transformational grammar” propagated by Chomsky is one type of generative grammar: one capable of assigning a structural description to all the strings it allows (cf. II. 1. 17). Other grammars, both transformational and generative, were proposed once the trend started, but Chomsky stipulated most of the “standard” assumptions. For example, he rejected the notion of probability in linguistics on the grounds that every sentence has close to zero probability — logically true, but operationally false. Though the detailed surface structure of a complete sentence may often be unique, participants in communication have elaborated expectations about what concepts, goals, and actions are more or less probable in a given context. (II.3.14; IV.1.2; IV.2.15, 54).

    3.6 Despite common talk of a “Chomskyan revolution,” the new grammar actually upheld many contemporary attitudes in science and linguistic theory. The place of logic in positivism and popular opinion (II.3.2f) made it a good candidate for extrapolation. The already central notion of the “sentence” attained an even higher theoretical status when it became axiomatic, i.e. postulated a priori (II.3.4f). The notion of “deep structure” lent {69} new impetus to the established striving for well-structuredness by adding theoretical substrates of structure (cf. I.1.3; II.1.4), e.g., to distinguish between superficially similar sentences like (Chomsky’s, 1965: 22):

    (7) I persuaded John to leave.

(8) I expected John to leave.

 Thus, the overall scientific climate favored the new grammar despite its innovative terminology.

   3.7 This success determined the priorities of language study. First, the primacy of syntax was established not only over meaning and purpose, but also over phonology and morphology (cf. Householder, 1965; Dik, 1967). Researchers would blithely affirm things like:  

The fluent speaker’s ability to use and understand speech involves a basic mechanism that enables him to construct formal syntactic structures […] and two subsidiary mechanisms: one that associates a set of meanings with each formal syntactic structure and another that maps such structures into phonetic representations. [emphasis added] (Katz and Postal, 1964: 2)

Second, the independence of syntax from the rest of language was implied by the design of the logical model. Third, the independence of language from sociological and cognitive contexts had to be postulated if the grammar alone was to stipulate which sentences were well-formed or grammatical. The abstraction of the language system from language use — prefigured in Saussure’s (1916) dichotomy of “langue vs. parole” (II.1.2) — was renewed as “competence vs. performance.” Fourth, language was defined as a means of predication, i.e., the assigning of “arguments” to “predicates” to form “propositions” (e.g. Jarvella, 1977). The meaning of “propositions” depended on the conditions under which they are true (Chomsky, 1965: 22).

    3.8 Chomsky soon reached out beyond the domain of formal grammar. According to his own testimony (1975), he felt provoked by the “intellectual scandal” of physicalist and behaviorist theories, especially Skinner’s. Transformational grammar appeared suitable for the role of a rival theory of human processing. Admittedly, there was no hard evidence that humans produce and understand utterances the same way that the grammar handled sentences. Yet psycholinguistics, a field striving to merge linguistics with psychology (II.1.18), readily took this correlation as a working hypothesis (e.g. Miller, 1962; Mehler, 1963). Soon after, Chomsky’s (1965) version of the grammar ventured into extensive psychological implications. Even in his early Skinner review, Chomsky (1959: 56) vacillated: on the one hand, he warned that “the construction of a grammar that enumerates sentences in such a way that a meaningful structural description can be determined for each sentence does not in itself provide an account of this actual behavior”; on the other hand, he considered the “grammar” to be “a statement of the {70} integrative processes and generalized patterns imposed upon the specific acts that constitute an utterance,” and the “rules of the grammar” to be “the selective mechanisms involved in the production of a particular utterance.”

    3.9 This ambivalence persisted throughout transformational grammar and engendered a double-faced terminology whose formal definitions in the theory were at variance with their commonsense meanings: “competence,” “generate,” “speaker-hearer,” “language acquisition device,” and so forth. Ironically, Chomsky fell prey to the same predicament he diagnosed in Skinner (II.2.15): “the terms used in the description of real life” and those of the abstract grammar were often “mere homonyms.” The grammatical terms became pervasive, hardly noticed metaphors (cf. I.1.3).¹ [Chomsky (1965: 9) acknowledged that his terminology had created a “continuing misunderstanding,” but saw “no reason for a revision.” The “misunderstanding” was probably an advantage.] The correlation between the grammar and the human mind, like that between logic and language, was simply asserted and accepted as an act of faith. The finding that more complex sentences are harder to process would be predicted by almost any grammar, not just by Chomsky’s.² [ Specifically, Chomsky’s model predicts that processing times reflect the number of transformations people are doing in their heads to work down to, or up from, “deep structures”; this “derivational complexity hypothesis” would appear in some form for virtually any model (cf. Kintsch, 1974; Levelt, 1978).]

    3.10 Chomsky himself apparently considered empirical discovery via observation wholly inessential. He averred (1965: 15, 19f) that “the description of speech habits” and the discovery of “patterns” in “a corpus of observed speech” could “preclude the development of a theory.” “Reliable operational criteria for the deeper and more important notions of linguistics” would never “be forthcoming”; the “sharpening of the data by more objective tests” was “a matter of small importance,” since “objectivity can be pursued with little consequent gain in insight and understanding.” A better source of evidence was the linguist’s “intuitions” as a “native speaker”—an “enormous mass of unquestionable data.”

    3. 11 Though intuitions play some role in most sciences (cf. I.18), linguistics was the only discipline to proclaim openly that the “intultions” of the investigator are its “unquestionable data.” The supplying of contexts, previously unacknowledged in scientific procedure (I.1.3; II.1.3, 7, 11, 18), came out in the open, but with important restrictions. The bias toward mentalism and formal logic prevented an examination of the really broad contexts of communication in social interaction. On the contrary, Chomsky’s approach obscured rather than clarified the correlation between theory vs. empirical evidence observed in everyday language (cf. Beaugrande, 1982d). First, such evidence was “performance,” and the theory was directed to “competence”; “psycholinguistics” was expected to explain away discrepancies between performance and competence (Slobin, 1971). {71} Second, the theory focused on “deep structure,” whereas real language events provided no evidence except “surface structure,” whose nature was repeatedly redefined in “extensions” of the “standard” theory (cf. Chomsky, 1965 vs. 1972 and 1976).¹ [In the “standard theory” (Chomsky 1965), surface structure was given no role in the interpretation of a sentence. In the “extended standard theory” (Chomsky, 1972), both deep and surface structure were admitted to the interpretation. More recently still, in “trace theory” (Chomsky, 1976), the representation is a single level of surface structure annotated by “traces” that in effect preserve the predicate/argument structure of the sentence as a proposition.] Third, contrary evidence could be accounted for by adding new rules and leaving the theory itself untouched. Fourth, the role of context in defining grammatical categories was disregarded (I.4.6; V.2.27; Lenneberg, 1975); the more “abstract” the grammar, the less secure its constructions.

    3.12 It would have been necessary to test the more fundamental claims implied by transformational grammar: (a) that the structures of various sentences are derivable from each other by applying formal rules; and (b) that simple structures are more basic to language (and to knowledge of a language) than complex ones. Though seemingly reasonable, these claims may be either trivial or misleading, depending on how we interpret them:

   3.12.1 The two claims are trivial in the sense that the grammar itself stipulates the same categories for all structural descriptions. Hence, any sentence structure can by definition be derived from any other by manipulating shared categories. The number and elaboration of categories and rules also decide what counts as simple or complex. The two claims are thus inherent in the grammar itself, though so far, no complete grammar of this kind has been designed for any natural language.

   3.12.2 The two claims are misleading in the sense that the decision to create one sentence structure in preference to any other, e.g. an active vs. a passive, is guided chiefly by meaning and purpose, e.g. the desire to express an event with a particular focus (cf. IV2.56, 61; VI.I.22); syntactic rules only specify the options. Hence, the operational relationships among different sentence structures are semantic and pragmatic, not just syntactic. Consequently, purely syntactic rules yield no realistic measure of processing complexity (cf. II.3.2Iff). What is experienced as simple or complex depends on the total processing demands within the current context, including a person’s language skills. “The job of syntax is not central but rather peripheral in ordinary language-merely accommodating lexical decisions made on the basis of the fleeting interests and motivations entertained by speakers” (Osgood, 1971: 529). No one has yet proven that the categories and structures posited by the grammar play a prominent part in normal communication.

   3.13 The problem of empirical reality can be illustrated with the common assumption that the sentence is the primary unit of speech production and comprehension (Ohmann, 1964a; Bever, Lackner, & Kirk, 1969; cf. survey {72} in Levelt, 1978). This assumption is not borne out by everyday speech (cf. I.4.11.2; ]V2.17; V.2.18, 30, 33, 45; V.3.7ff; VI.I.10), where the boundaries of phrases and clauses often reflect intonational and conceptual units, not just syntactic ones. This discrepancy obliges the proponents of abstract grammars to manipulate evidence, e.g., to (a) convert it themselves into well-formed sentences; (b) assert that well-formed sentences “underlie” the evidence; (c) exclude any evidence that doesn’t happen to be a recognizable sentence; (d) invent their own examples as needed; (e) rely only on a written sampling, where sentences are usually complete and their boundaries well-marked (what Abercrombic, 1965, calls “spoken prose”; cf. II.1.13);” or (f) use the term “sentence” inconsistently to mean not just a grammatical structure, but also a predication, a proposition, an utterance action, a speech act, and so forth (cf. Beaugrande, 1980a: 11ff).

 3.14 All these recourses can be found in mentalist linguistics. As a result, evidence was subordinated to theory in unprecedented ways. Real discourse came to be regarded as deformed and degenerate (e.g. by Ohmann, 1975: 298; Stockwell, 1977: 168). Paradoxically, this attitude is just as evaluative as that of traditional prescriptive grammar, though for different reasons. The old grammarians unabashedly defended their own usage as “good” and “proper,” as meeting the demands of social prestige, elegant diction, and rhetorical effectiveness. The new grammarians propagated their own usage by allowing their language attitudes and social commitments to influence their judgements of grammaticality. People most readily recognize a banal sentence as grammatical, not because they have “internalised” a grammar of “well-formedness,” but because they can easily imagine a context where the sentence could be used (Bolinger, 1968; McCawley, 1976; Snow & Meijer, 1977). Therefore, the old grammarians’ samples taken from outstanding writers and orators of the day could in fact seem deviant, marginal, or ill-formed to the new grammarians cf. VI.1.If), who presented English as a flurry of inane sentences about‘John’ and ‘Mary’ — the star-crossed pair whose ideal, abstract exploits provided the content of linguistic samples for years to come. English was converted from a vehicle of memorable style into “ infinite set of banalities”.

 3.15 With so many empircal limitations, standard transformational grammar can be no more than a partial theory of literal paraphrase. Humans would create sentences in ways comparable to the grammar if they had to recast a sentence into a close paraphrase using the same word stems. This occasion is are in comparison to looser paraphrasings, {73} reports, gists, and so on; and even rarer among the totality of language activities. Chomskyan linguistics was thus forced to exclude major aspects of language and insist on a version of “competence” so narrow as to be irrelevant.

   3.16 However, “transformations” could be useful in a broader science of text and discourse not centered on abstract syntax. Configurations mapping from one language level or processing phase to another are presumably “transformed,” e.g., when an array of goals of concepts is projected onto a linear surface text, or vice-versa (cf. IVI.7-9). According to Wundt (1970 [1900]), text production takes a mental configuration (an “idea” in the sense of III.2.14) of simultaneously activated components, organizes it into grammatically or logically related segments within an orderly hierarchy, and finally transforms it into the surface sequence in the text. Other types of transformations have been envisioned that convert: (a) a spatial representation into linear syntax (Degerman & Mather, 1972); (b) a “logical form” into a sentence, as in “generative semantics,” a counter-movement inside transformational grammar (cf. papers in Steinberg & Jakobovits [Eds.], 1971); (c) one gestalt-like configuration into another (Piaget, 1976: 130f); (d) the order in which concepts are activated into the order in which those concepts are expressed (Osgood & Bock, 1977: 91); (e) an underlying content hierarchy into a story, as in the “story grammars” developed by cognitive psychologists (see especially Johnson & Mandler, 1980; survey in Beaugrande, 1982d); or (f) a text into an elaborate network of logical representations, as in János Petöfi’s (1980) “text-structure/world-structure theory.” Some of these transformations also go in reverse. The proposals all concur that transformations need not convert one syntactic pattern into another, but may rather a non-linear pattern into a linear one (or vice-versa). This approach to “transformation” deserves more research.

   3.17 In language education, mentalism has played a decidedly smaller role than behaviorism. Speculations about an innate “language acquisition device” (cf. Chomsky, 1965: 32, 38) that extracts a “grammar” from linguistic experience implied that language education isn’t needed at all. In that case, language teachers have nothing to do; and whatever successes the child attains with language would not go to the credit of the schools. In order to yield an educational tactic, mentalist linguistics had to be combined with behaviorist methods: “transforming” sentences for a specified purpose (cf. V.2.20ff). As Francis Christensen (1963: 155) observed, “the chapters on the sentence” in traditional composition handbooks “all appear to assume that we think naturally in primer sentences, progress naturally to compound sentences, and must be taught to combine the primer sentences into complex sentences-and that complex sentences are the mark of maturity.” Accordingly, Chomsky’s grammar became a new classroom tool for this old enterprise. Bateman and Zidonis (1963: 2f) experimented with the precept that “generative grammar […] is in essence a representation of the {74} psychological processes of producing sentences”; hence, “pupils must be taught a system that accounts for well-formed sentences before they can be expected to produce more of them.” Soon after, SENTENCE-COMBINING was launched by Hunt (1965): the application of transformational grammar to make longer, more elaborate sentences out of shorter, simpler ones.

 3.18 In the following years, sentence-combining developed into a large domain of writing research (cf. surveys in Daiker, Kerek, & Morenberg, 1978; Daiker, Kerek, & Morenberg [Eds.], 1979; Morenberg & Kerek, 1979; Kerek, Daiker, & Morenberg, 1980). This domain promised the well-structuredness that makes research problems widely attractive (I.11.3). The method worked with simple statistical tabulations: numbers of words in clauses or in sentences, number of clauses per sentence, and so on. The central entity was the “T-unit,” i.e., an independent clause with all materials that depend on it.¹ [1. The “T-unit,” a “terminable” sequence that could be a “terminal string” (sentence) in transformational grammar (cf. Hunt, 1965) suppresses the empirical fact that the writer did put periods in some places and not in others. The distinction between the period vs. the means for linking independent clauses (semicolon, co-ordinative junctive, etc) is sacrificed along with the conceptual and pragmatic motives that influence sentence boundaries. At least, a new theoretical justification for this alteration of the data is needed, now that the original theory has declined (II.3.44).] A “T-unit” could thus be punctuated as a sentence, whether or not the writer did so (compare Hunt, 1964, 1965). Such units are usually easy to recognize in prose passages, and difficult cases can be excluded from the data.

   3.19 Hunt’s (1964) project “analyzed by transformational methods” the “differences in grammatical structures written at three grade levels: fourth, eighth, and twelfth grades. He found that “older students increase the length of their T-units by using more subordinate clauses as well as non-clauses”; he concluded that “T-unit length is a more valid index of maturity in writing — at least in the early grades — than is (a) sentence length or (b) clause length [or] (c) the subordination ratio” (1964: 50). Note that “the word ‘maturity’ is intended to designate nothing more than ‘the observed characteristics of writers in an older grade.’ It has nothing to do with whether older students write ‘better’ in any general stylistic sense” (Hunt, 1965: 5). Hunt’s special use of the term “maturity” set a trend still beset by controversy today (cf. survey in Faigley, 1980).² [Faigley (1980: 293f) found, for example, that cookbooks yield ratios below Hunt’s figures for fourth graders.] Mellon (1969: 15f) “stipulatively defines” the “maturity of sentence structure” in “a strictly statistical sense, in terms of the range of sentence types observed in representative samples of student writing.” O’Hare (1973: 35) avers that his method “would increase the syntactic maturity in students’ free writing.” The group at Miami University (in Oxford, Ohio, of all places) holds that, of the three measures offered {75} by Hunt (cited above), “the single most accurate index of syntactic maturity for writers beyond the high school level is clause length, the mean number of words per clause,” while “T-unit length” is “the second most accurate index,” and “subordination ratio” is not “reliable” “at the post high school level” (Daiker et al., 1978: 39).

    3.20 Like many terms of transformational grammar (cf. II.3.9), the term “maturity” in these studies is given a theoretical definition at variance with its everyday meaning. “Maturity” normally designates a furthering of the processes traversed between childhood and adulthood. Sentence-combining researchers have taken one common, but not necessary, symptom of these processes and interpreted it as a cause or at least as a defining feature. Their thesis is then: “if children in lower grade level intensively practice that skill which enables older-more ‘mature’-students to produce writing characteristic of their own level, then such practice will help accelerate the younger children’s maturation as writers” (Kerek et al., 1980: 1068). Though educational methodology typically hopes to force “growth” via behavioral drills, the theoretical justification is by no means obvious. Older students typically smoke more tobacco, drink more alcohol, and commit more vandalism than younger students do; but no one proposes teaching the younger ones to act out these symptoms in a classroom in order to “accelerate their maturation.” External differences at various age levels should figure in education only if we can explain the relevance of the observed or rehearsed behavior to cognitive and social advancement.

   3.21 By themselves, the tabulations in sentence-combining research (II.3.19) only indicate syntactic complexity. Counting the number of different structures would measure syntactic variety. The rate and smoothness with which students create sentence structures would yield measures of frequency and fluency of syntax (cf. II.2.5). But these measures are all we can determine simply by counting, and whether they are relevant contributors to “mature” writing” is still an unproven hypothesis. Several disputes bear on that hypothesis.

   3.22 First, the particular tabulations are open to challenge. Christensen (1968) found his notion of a “mature style” was reflected not in the dependent clauses, but in free modifiers (adjuncts appended to entire clauses), especially in sentence-final position (compare Johnson, 1969; Wolk, 1970; Faigley, 1979a, 1979b). Shaughnessy (1977: 187-225) defined writing maturity by the size and variety of vocabulary: the types, not the numbers, of words in the text (cf. Grobe, 1981; Nelson & Piché, 1981; 1); word-frequency statistics (e.g. Kucera & Francis, 1967) could be used to measure the rareness of a writer’s choices. The word counts also ignore how length and complexity adapt to a particular mode of discourse (cf. Crowhurst & Piché, 1979; Faigley, 1980; Witte & Davis, 1980). Finally, the role of the writer must be kept in view. Syntactic proportions do not make Ernest Hemingway less “mature” than Edward Gibbon (cf. Hayes, 1969). Short, {76} simple sentences can be the sign of the mature “tough talker,” while the immature “sweet talker” and the dehumanized “stuffy talker” tend toward long, complex ones (Gibson, 1966). All these considerations, suggest that the usual tabulations do not equal maturity.

    3.23 Second, complexity shifts are not the only differences linked with age level, nor are they necessarily so. Loban’s (1976) twelve-year longitudinal study observed a general rise in complexity with increasing age, but specific constructions proved to be unreliable signals. Higher numbers of dependent clauses appeared as time passed and correlated with the higher abilities of students as gauged by other tests — but only for spoken language (cf. II.3.24). In written language, these increases leveled off by eighth grade and disappeared by tenth grade, long before the college level. Thus, specific constructions may indicate maturity for some writers, but not for others. For instance, a writer might prefer the “free modifiers” counted by Christensen over the “dependent clauses” counted by Hunt and his successors (II.3.19); the additive construction may have the same implications for maturity as the hierarchical one (cf. Faigley, 1980: 296f): Frequency count indicators of growth become equivocal when applied to specific usages, because linguistic devices are often substitutable for one another, so that the increase in the use of one may result in the decrease in the use of another. (Bereiter, 1980: 76) Moreover, Witte and Davis’s (1980) study suggests that syntactic complexity is not a stable trait of a person, but varies to fit different contexts (cf. II.3.37, 41 f; IV.2.60; VI.1.11).

    3.24 Third, complexity correlates with social standing (Beaugrande, 1981d). The higher syntactic complexity of middle-class speech over working-class speech was observed by Basil Bernstein (1962: 225ff) on precisely such indicators as number of dependent clauses. Oevermann (1970) found the same correlation for German school children. Apparently, the long contact of middle-class speech with the modality of writing has led to an elaboration of syntactic options for differences that working-class speech sustains with intonation and non-verbal signals instead. Perhaps sentence-combining researchers harbor the ambition to foster upward social mobility as well as maturation. If so, they should gather more extensive sociological data about their enterprise (II.3.29).

   3.25 Fourth, there is good evidence that a rise in syntactic complexity need not improve writing quality (cf. Perron, 1974; Callaghan, 1977; Sullivan, 1977; Stewart, 1978). Readers react to complexity in various ways (cf. Nold & Freedman, 1977). Sentence-combining might bewilder students with so many structures that “they lose control of their sentences” (Maimon & Nodine, 1979: 240). An extreme gap between complexity and quality was found among Gosnave’s (1977: 43f) aphasics who insisted on displaying their syntactic accuracy, fluency, and complexity, despite instructions “to {77} utter simple (that is, kernel) sentences” — yet their discourse was largely incoherent.

   3.26 Fifth, unskilled writers may uncritically equate “good writing” with complexity and struggle so hard that they attain the opposite effect (cf. II.3.46; VI.I.28): “An inexperienced writer may draw on the same passive constructions, the same circumlocutions as the bureaucrat, who uses these syntactic strategies deliberately, as a way of blurring and suppressing information. How ironic it would be if so-called “remedial” English were to produce no more than a mastery of bureaucratic syntax!” (Shaughnessy, 1977: 87). Bureaucratic syntax illustrates how complexity can damage readability (cf. Flesch, 1972; Williams, 1979b; Faigley, 1980; II.3.33; III.3.16; V.2.35; VI.I.26). Shaughnessy (1977: 77) compared sentence-combining to “finger exercises on piano” and cautioned that writers may “self-consciously decorate their thoughts rather than developing them.” In Williams’ (1979a: 28f) test population, all the teachers from high schools and most from colleges gave high evaluations to the heavily nominal style often criticized in classes and handbooks. Williams noted teachers “apparently take a nominal style as evidence of thoughtful mature writing” (cf. also Freeman, 1981; Nelson & Piché, 1981). If so, their quality judgements contradict the style they recommend to students. Williams added that nominalizations increase the ratio of words per clause-the key measure of maturity in sentence-combining research (cf. II.3.19).

   3.27 Sixth, research has to control all the other factors and variables besides sentence-combining that affect the quality or maturity of student writing. The massive study at Miami University was a model of circumspection. A control group (who read and analyzed essays, discussed rhetorical principles, and studied modes of developing essays) ran alongside the experimental group (who did only sentence-combining all semester). Some 48% of the subjects were tested 28 months later for lasting effects. The researchers also controlled for a stunning range of variables: holistic vs. analytic ratings; students’ scores on the ACT composite and SAT verbal tests;” instructors’ experience, teaching abilities, age, sex, academic rank, and degree; the testing room’s location, architectural style, temperature, light, comfort, and noise level; the time of day for the testing; and so on (cf. Kerek et al., 1980: 1086-1090, 1107-1120, and 1113f). Papers were rated for five entire days by 28 independent judges with advanced degrees and demonstrated expertise. Papers were so coded that the judges couldn’t tell which group or test occasion was involved. Every paper was graded four times analytically and four times holistically, each time by a different judge. {78} The judges were trained to use a uniform scale and grade each paper on its own merit, disregarding the factors of photocopy quality, handwriting, neatness, spelling, titles, and outlinings (Kerek et al., 1980: 1112).

   3.28 This exemplary thoroughness proved rewarding. The experimental group outscored the control group not only on the conventional measures of syntactic complexity listed in II.3.19, but also on the both the holistic and the analytic ratings, including criteria whose relation to syntax is far from obvious (Kerek et al., 1980: 1111): (a) “ideas: the extent to which the ideas of the essay are mature, insightful, and clear”; (b) “supporting details: the extent to which the ideas of the essay are supported by examples and details which are specific, appropriate, and fresh”; (c) “organization and coherence: the extent to which the thesis is developed in an orderly and logical way and to which sentence is linked to sentence and paragraph to paragraph through effective transitional devices”; (d) “voice: the extent to which the essay speaks with individuality and distinctness of tone” (cf. I.2.8.6); (c) “sentence structure: the extent to which variety, maturity, and effectiveness of sentence structure is [sic] achieved”; and (f) “diction and usage: the extent to which wording and phrasing are accurate, expressive, and concise, and to which the conventions of standard English are observed.” Only the “variety of sentence structure” mentioned in (e) directly relates to the syntactic manipulations stressed in sentence-combining. Still, the experimental group fared better on every criterion except (c) (Kerek et al., 1980: 1116f), even though the control group studied content organization, diction, usage, and rhetoric with the aid of McCrimmon’s (1976) Writing with a Purpose.

  3.29 The Miami study accounted for almost every factor except social variables (II.3.24). They might have probed social class status of the writers, not just verbal test scores (cf. I.3. I ff). Or, they could have selected judges from non-academic environments like industry and commerce, whose opinions would differ from those of English teachers (cf. Diederich, French, & Carleton, 196 1; Faigley et al., 1981; Hairston, 1981). In any case, what we need now is not further studies that replicate the results of the Miami group, but studies that explain such results. Until then, the recommendation that “college courses” should “use sentence-combining” as the central “organizing principle of the syllabus” (Kerek et al., 1980: 1148; cf. Daiker, Kerek, & Morenberg, 1981) seems premature.

    3.30 Several explanations are worth pursuing. The most facile would be the “Hawthorne effect”: spurred on by the novelty, discovery, and conspicuousness of their performance, an experimental group usually out-performs {79} a control group.¹ [The “Hawthorne effect” got its name from a study of working conditions vs. fatigue and monotony in the Hawthorne Works of the Western Electric Company in Chicago (Roethlisberger & Dikson, 1939).] The Miami study was a publicized, well-funded project in a small Ohio college where innovation would attract attention. But the Hawthorne effect could function only if the experimental group also knew how to manipulate the six criteria listed in II.3.28 —i.e., if they already possessed such writing expertise that they shouldn’t have been able to improve so markedly (indeed, shouldn’t need composition instruction). An increase in motivation couldn’t be decisive without the means to raise one’s writing quality.

   3.31 Another facile explanation is that the experimental group did more writing while the controls were reading and analysing essays (cf. Kinneavy, 1979a: 11). This account would agree with the behaviorist view that “conditioning” leads to learning in proportion to the “frequency of trials” (cf. II.2.5), and with the information-processing view that extended trials render processing “automatic” and hence more rapid and efficient (cf. III.1.21). However, some research shows that writing practice alone need not improve quality (e.g. Heys, 1962; Arnold, 1963), just as repetition is no guarantee for learning (cf. II.2.27; III. 1.28, IV2.38). Practice might simply rehearse disadvantageous strategies and impede progress by locking the student into bad habits (I.2.23.3). In sum, we can reject this explanation also.

    3.32 A more interesting explanation advanced by Kinneavy (1979a: I 1) is that the results tell us more about judgments of English teachers than about quality of student writing. Superior holistic ratings correlated heavily with superior sentence structure from analytic ratings.¹ [Kinneavy’s (1979a: 11) published figure of 64% variance-accounted-for came from an early draft of Daiker et al. (1978). According to Kerek (personal communication), the figure was withdrawn when regressions showed it to be unreliable. All the same, the study itself suggests that holistic scoring contains a strong reaction to sentence structure. Research is certainly necessary on how a holistic score really emerges (cf. Freedman & Calfee, 1982).]  Kinneavy (ibid.) opines that ordinary English teachers, whatever they say they grade for, “really seem to grade a theme by the quality of its sentence structure”; “maybe English teachers are not the great judges of a theme that they are usually assumed to be. Conceivably, by quite specific norms, [e.g.,] factual and organizational criteria such as are used to judge the usual news story in journalism, or the rigid evidence that a scientist might require, or the careful definitions that a logician might demand in a classification paper, these papers might rate as vacuous, illogical, and undefined.” ² [The copies of papers from the Miami project, kindly sent to me by Kerek, show that the students were writing personal essays on topics like “escaping from reality.” Kinneavy’s scientific, journalistic, and logical “norms” don’t fit such papers, so that his suspicion quoted here can’t be resolved. I can detect little difference in the basic content organization between pretest and post-test, but then I am not accustomed to assigning such topics and therefore am not well qualified to rate the papers. {80}

 3.33 Kinneavy’s reservation is important. There can be no such thing as writing quality in a vacuum, but only for a given audience with particular expectations, values, and requirements (I.3.13). Hake and Williams found that English teachers rewarded the nominal style they “perceived as better organized, better supported, and better argued than the corresponding verbal paper” (Williams, 1979a: 28; cf. II.3.26). Yet nominalizing a paper scarcely affects how its content is organized or argued. Moreover, nominal style was found by Coleman (1964) to be harder to read than verbal; and is a notorious encumbrance in bureaucratic writing (V1. 1.28). Accordingly, intensified research on the evaluative practices of English teachers is pressingly required (cf. I.2.19ff; 1.3.13-20).

    3.34 Still another explanation was tendered by the Miami group itself. Unlike most sentence-combining studies in the past, the Miami effort strove to integrate into sentence-combining some rhetorical considerations: “style, purpose and thesis, organization, logic and coherence, tone, punctuation and grammar, even diction” (Kerek et al., 1980: 1100). Sentence-combining may be neutral or negative unless students have clear motives for deciding when and whether sentences should be combined (Beaugrande, 1979b; Kinneavy, 1979b; Williams, 1979b; cf. II.3.26, 46). Measures of “rhetorical effectiveness” can hardly be attained by counting words in “T-units.”

   3.35 The most interesting explanation, in my view, is that a general process model of cognition and communication might allow us to define the relationships between sentence-combining and the criteria enumerated in II.3.28. Researching this point seems far more urgent than replicating the usual projects (II.3.29). I shall try to outline first the theoretical and then the practical conditions under which sentence-conibining might have a positive influence upon the overall complex activities of writing and writing instruction.

    3.36 The first theoretical possibility is that sentence-combining brings to conscious awareness the syntactic formats that the student has already acquired for speech and now needs for writing (Shaughnessy, 1977: 78), especially insofar as the latter modality is typically more complex and hierarchical than the former (V.3.13; VI. 1. 10). Proper sentence-combining practice could help students transfer their syntactic expertise from a familiar activity to an unfamiliar one. Their speech is often articulate, varied, and elaborate, but their writing suffers from inhibition and overload that lead to deceptively simple sentences. Explicit, orderly tactics to break apart or build up sentences are easier to follow than unqualified demands to revise or reword a passage, or to imitate models taken from skilled writers whose achievement looks unattainable.

    3.37 Second, sentence-combining should, by raising awareness (II.3.36), increase the familiarity of syntactic patterns. Their experiential complexity would then be lower, though their structural complexity would remain constant (cf. II.3.12.2; III.3.2.7; III.3.4.3). Practice in constructing {81} elaborate sentences should render part-whole relationships easier to recognize and manipulate, and again combat overload (cf. Maimon & Nodine, 1978, 1979).

    3.38 Third, the training could improve the fluency and frequency of writing sentences (cf. II.2.5; II.3.21), provided control increased also (Epes, 1981: 142). Writing either too smoothly and rapidly or too slowly and jerkily is equally risky. We need a method to define a strategic rhythm for the individual writer. The “speech rhythms” Jaffee (1977) observed to persevere for each person despite shifts in motivation, fatigue, attention, and mood, may correspond to personal “writing rhythms” (cf. Matsuhashi, 1979).

   3.39 Fourth, greater familiarity and fluency, and lower anxiety, should improve students’ confidence about managing sentences, and make it easier to vary, experiment, and innovate. Many students have been intimidated and feel incompetent to use grammar (cf. I.2; V.2). Explicit steps that demonstrate the formation of grammatical patterns can correct these negative outlooks. Discerning judgements can be made about non-strategic options within the grammar of the language, e.g. (Chomsky, 1965: 11):

   (9) I called the man who wrote the book that you told me about up.

(10) The man who the boy who the student recognized pointed out is a friend of mine.

 Students learn not to strive for sentences so complex as to be impractical (cf. II.3.46; VI.I.28).

  3.40 Fifth, the training could promote variety (cf. II.3.21), such as the “elegant variation” ambivalently treated by Fowler (1965: 148ff). Fowler warns that variety for its own sake may be gratuitous, even damaging. Writers do better to consolidate a redundancy, e.g. in (11), than to hide it behind variation, e.g. in (1a):  

(11) Vincristine makes the patient tired most of the time and keeps the person nauseated, too.

(11a) Vincristine makes the patient tired and nauseated most of the time.  

Fowler inveighs against writers who are “intent on expressing themselves prettily rather than conveying their meaning clearly.” Clarity sometimes demands stability, not variety, e.g. in parallelism (cf. IV.2.40). Variety should be used moderately as needed to prevent syntactic monotony.

   3.41 Sixth, sentence-combining could be applied to controlling the informativity of a text: the rate at which new materials are added to what has been said so far. Short, simple sentences are suitable if readers might find the content difficult or unfamiliar; or if the writer wants to convey the impression of important content, e.g. Gibson’s “tough talkers” (II.3.22) or newspaper writers (remember Hemingway’s background as a journalist). Longer, more complex sentences go better with easy or familiar content (IV.2.60; IV3.12, 17). Dependent clauses are good for incidental or uncontested {82} materials, as in Posner’s (1980: 389) illustration, where the fact of the ‘crossing’ is taken for granted:  

(12) That the soldiers crossed the border surprised even the intelligence service.

 Only if informativity is kept constant does a structurally complex sentence exceed the processing load of a simpler one.¹ [Taylor (1969) showed that latencies and pausings during sentence production correlated with topic difficulty rather than with length (number of words), structural complexity (determined by “immediate constituent analysis,” e.g. N. Johnson, 1965, 1966; cf. II.1.13), or sentence type (simple active affirmative declaratives vs. passives, negatives, passive negatives, etc.). Compare II.3.12.2.]

  3.42 Seventh, it follows from II.3.41 that sentence-combining could also aid in controlling the chunking of discourse (cf. I.4.5, II.2; III.1.10; III.3.2.7). Sentence boundaries often, though not always, function as conceptual boundaries as well (cf. Goldman-Eisler, 1972: 111; Butterworth, 1975: 82; Just & Carpenter, 1980: 345f). The components of an event naturally form a chunk, and may readily fit into a sentence. Bransford and Franks found that readers who had seen a series of short sentences about one event, e.g. (13), believed they had seen a single long sentence, e.g. (13a) (Bransford & Franks, 1971: 332):

 (13) The rock rolled down the mountain. The rock crushed the hut. The hut is at the river. The hut is tiny.

(13a) The rock that rolled down the mountain crushed the tiny hut at the river.

 as if the memory itself does some sentence-combining on occasion²,[In fact, the short sentences were interspersed among passages not even belonging to the same story. Yet the confidence in recognition was strongest for the combined versions like (13a),(cf. also Franks & Bransford, 1972; Griggs, 1974). Simple sentences are easier to read, but soon lose any advantages in speed or accuracy for answering content questions (Kintsch & Monk, 1972; King & Greeno, 1974). These various findings suggest that the length, complexity, and type of sentences affect the shallow, earlier stages of reading; more profound, lasting effects result from the density and integratedness of content.

   3.43 Eighth, syntax is operationally integrated with the other levels of discourse processing (III.2.3ff). An evolution in syntactic capabilities could create favorable, though not obligatory, conditions for an evolution on other levels. For example, the ability to form complex sentences could support the ability to create compact hierarchies of conceptual content, as skilled adult writers do more often than children (Caccarnise, 1981).

   3.44 If we admit, pending new research, that the eight possibilities enumerated in II.3.36-43 could offer a theoretical basis for the positive effects of sentence-combining, the findings of the Miami project begin to make sense. Judges saw improvements in “ideas,” “supporting details,” and {82} “organization and coherence” because syntax was better used to control informationality (II.3.41) and chunking (II.3.42). “Voice” and “diction” could improve though syntactic variety (II.3.40) and personal confidence (II.3.39). “Usage” could profit from a rise in the familiarity of available formats (II.3.37) and in the fluency to handle them (II.3.38). Of course, these correlations are speculative until they are empirically tested. Theoretical research on sentence-combining is particularly urgent in so far as its original foundation, transformational grammar, no longer seems (to me anyway) a viable theory of language.

    3.45 On the practical side, the success of sentence-combining depends on a carefully designed approach. For one thing, the basic sentences to be combined should be spontaneous, naturally occurring samples students might actually write, not the unrealistic, forced constructions found in some experiments. Incautiously reducing complex sentences to basic ones for the students to recombine can produce versions that distort the original meaning (O’Hare, 1973: 90ff):¹[ O’Hare (1973: 37) used “sentences virtually identical to those written out by Mellon’s group (cf. samples in Mellon, 1969). But Mellon was after all working directly with and from the linguistic theory; O’Hare was not, and should have designed more realistic samples.]

   (14) The assignments were English [vs. the English assignments].

(15) The reports were written [vs. the written reports].

(16) The formations were rising hundreds of feet into the air [vs. the formations rising hundreds of feet into the air].

 Despite linguists’ claims that transformations are “meaning-preserving” (Olson, 1977b: 19), such transmutations can alter or distort the sense by converting modification into assertion — a pragmatic distinction overlooked in a narrow syntactic theory (cf. II.3.12.2). In (14), assignments seem to hail from England; in (15), an action is announced rather than a kind of object; and in (16), the scenic marvels (according to O’Hare, already “one of Asia’s greatest wonders”) evoke redoubled stupefaction by quitting their moorings and floating aloft. Oddness results also from the placeholders needed to fill out grammarians’ rules (O’Hare, 1973: 81-92): 

 (17) Most teachers have learned something.

 (18) Something was only preliminary to something sometime.

 (19) A grammar is supposed to account for something and something.

 Real writers would produce such sentences only with satiric intent: an overview of teacher education (17), a “stonewalled” passage from a government statement (18), or the lesson we learned from mentalist linguistics (19). Yet the combined sentences students make out of these samples (one assumes) carry no sarcastic intent.

 3.46 Another practical requirement for sentence-combining is to {84} control the danger of gratuitous, muddled complexity, as in O’Hare’s (1973: 91):

(20) The princely heron, perched high on a rocky lodge whose height enabled the bird to survey the swirling blue-white waters that were below on three sides, tensed its wings for the spectacular plunge soon to be triggered by a school of approaching fish.

 is not just too awkward and too redundant (‘high/height’, ‘soon/fast approaching’) to be a model for student writing, but also misses the fact that the heron (the state bird here in Florida) feeds not by plunging from heights but by wading in ponds, viz . (20a). O’Hare must have been thinking of some bird like the fishing eagle.

(20a) These birds forage mostly by standing still or stalking very slowly at the edge of the water, waiting patiently for something to come within range, at which time they dart that dagger beak forward. (Department of Fish and Game)

Shortening sentences, removing dependent clauses and careless dumb mistakes makes it not less “mature,” but more readable:

(20b) The princely fishing eagle perched high on a rocky lodge and surveyed the swirling white blue waters below on three sides. The wings tensed for the spectacular plunge toward a fast approaching school of fish.

 Complex samples can be confusing too, e.g. (O’Hare, 1973: 89):

 (21) Connie’s constant chattering, which kept the hunters from hearing where the dogs were running irritated the men, who swore (that) they would never take her hunting again.

 where the involved syntax leaves it unclear if all the ‘men’, or only some of them, were ‘hunters’. Students who conclude that syntactic complexity is worth any amount effort should feel proud to write:

 (21a.) The hunters were kept from hearing, as they swore that never again would they never gain take the constantly chattering Connie hunting where the dogs were running, who irritated the men.

(21b) Connie who constantly chattered and who kept the hunters who swore never to take her hunting again from hearing where the dogs were running irritated the men.

 With 4 dependent clauses apiece, (21a) and (21b) are both more “mature” than (21) with 3 such clauses, according to Hunt’s “subordination ratio” (II.3.19). The point should be clear. Without a clear rhetorical account of when complex sentences are or are not appropriate, sentence-combining offers no guarantee that writing quality will improve (II.3.39). At times I have to discourage my students from straining for ungainly, unnecessary complexity, as when they wrote: 

(22) You go to registration at a time which is always inconvenient that is assigned by the office.

(23) There are also parts of the body, of which the appendix is a good example, which are supposed to be remnants of organs that were useful to our ancestors, and which although no longer any use to us have not yet disappeared.

  3.47 Clearly, sentence-cornbining should not consist of pasting odd, silly basic sentences, e.g. (14-19), together into tangles of complexity, e.g. (20-21b). Instead, students should deploy strategies to combine their own {85} spontaneous sentences as a solution to problems in phrasing, chunking, and redundancy. Sentence fragments can be integrated into nearby sentences (V2.36):

  (24) I guess that’s what makes a classic a classic. The ability to look completely different depending on how it’s used. (D’Angelo, 1979a: 571)

(24a) I guess the ability to look completely different depending on how it’s used is what makes a classic a classic.

 Comma splices (25) usually link clauses that should be chunked into one sentence (25a), rather than broken up by replacing the comma with a period (25b) (V2.43):  

(25) The school didn’t financially support the paper, all costs were raised by the journalism class.

(25a) The school didn’t financially support the paper, since all costs were raised by the journalism class.

(25b) The school didn’t financially support the paper. All costs were raised by the journalism class.

 The vague ‘this’ at the start of a follow-up sentence (26) can be clarified more concisely by combining (26a) than by inserting a head noun (26b):

   (26) They must know what they’re doing at all times. This can be quite a problem. (15 words)

(26b) Having to know what they’re doing at all times can be quite a problem. (14 words)

(26a) They must know what they are doing at all times. This requirement can be quite a problem. (16 words)

 Needless words and redundancies can be eliminated by combining two or more sentences into one (V1.2.28, 31):

 (27) Polygamy is a marriage in which one person is married to more than one person. A husband would have several wives, or a wife would have several husbands. Either way, the marriage is illegal.

(27a) Polygamy is an illegal marriage in which one spouse has two or more wives or husbands.

 Such applications integrate sentence-combining into the framework of writing overall as a useful remedy for commonplace problems rather than as an exercise unrelated to natural writing (cf. Moffett, 1968).

  3.48 In sum, mentalism, like behaviorism, has left us some tenets that deserve to be respected by future research, such as:

  3.48.1 “Linguistic theory is concerned with discovering a mental reality underlying actual behavior” (Chomsky, 1965. 4).

 3.48.2 “Linguistic theory” “presents certain empirical hypotheses” that “aim to categorize the schematism that the mind imposes in examining the data of sense and acquiring knowledge of language” (Chomsky, 1975: 13). {86}

  3.48.3 “A grammar of a language” should “provide an explicit analysis” of “how a sentence is understood” (Chomsky, 1965: 4f).

  3.48.4 The most striking aspect of “linguistic competence” is “the creativity of language” (Chomsky, 1966: 11).

  3.48.5 “It is questionable whether the theory of meaning can be divorced from the study of other cognitive structures” (Chomsky, 1975: 23; cf. 1.4.6).

 3.48.6 What speakers may say at a single moment need not reflect the extent of their knowledge (“competence”) about the language (cf. Chomsky, 1965: 4f).

  3.49 Though mentalism, again like behaviorism, offered only a small part of the total picture of language, some of its central notions remain useful. Its limitations should not blind us to the contributions it can make to future investigations.  

*    *   *    *   *    *   *    *   *    *   *    *   *    *   *    *   *    *

Click here to go to Chapter III

  Click here to go to Main Page