Issues in Text Processes
1. TEXT TYPES
To progress from a study of abstract structures in possible sentences to the
study of texts as communicative occurrences, we must confront a new challenge in
the domain of linguistic TYPOLOGY. In descriptive linguistics, typology centered
on minimal units, i.e., on repertories for distinctive features, phonemes,
morphemes, etc. In transformational grammar, typology centered on a set of basic
sentence patterns and classes of rules for building other patterns. Alternative
typologies for sentences used categories like “declarative – interrogative
– imperative – exclamatory” (traditional grammar); or “process –
action – judgment – identification” (Brinkman); or “process – action
– feature – classification’ (“functional” grammar) (see Helbig 1974:
159, 186). These latter typologies suggest a fundamental confusion about the
nature of the sentence. It is people, not sentences, who
“declare,” “interrogate,” and “exclaim.” It is concepts
and relations that are the basis of “processes,”
“classifications,” and the like, not grammatical
formats. Hence, the usual typologies of sentences cannot offer a means of
classifying texts as occurrences in communicative interaction (cf. Morgan 1975).
If sentence typologies are simple but sterile, text typologies are dauntingly
vast and subjective. Early attempts to press conventional linguistic methods
into service for text typologies were discouraging. We may count up word classes
or measure sentence length and complexity (Mistrík 1973) with no certainty of
distilling out crucial distinctions. Being told that advertising texts have an
abundance of adjectives, and news reports have lots of verbs (Grosse 1976a) may
provide a statement of symptoms for deeper-lying tendencies, but certainly
doesn’t explain the types themselves.
The landmark colloquium at the Center for Interdisciplinary Research at the
University of Bielefeld in January 1972 (proceedings in Gülich & Raible
[eds.] 1972) brought new issues to light. The proliferation of binary
oppositions so well known in phonology was proposed, yielding such questionable
and diverse constructions as “± spontaneous” (Sandig) or “±
figurative’ (Stempel). The plus-or-minus sign, placed in front of any
convenient expression as if it could transform an intuitive notion into a
scientific one, was denounced as indicating the absence of all formalisms (Kummer)
and hindering the development of theory altogether (van Dijk) (Gülich &
Raible [eds.] 1972: 136, 181). In effect, such features don’t account for a
phenomenon, but simply mark it with one of a large, totally unsystematic set of
It might be more productive to study text types from the standpoint of evolution
and usage. The INTERTEXTUALITY I suggested as indispensable for utilizing texts
(I.4.11.6) evolves from social as well as linguistic factors:
A differentiation of social settings and participant roles leads to a
differentiation of situation types.
The differentiation of situation types engenders reliance upon those text types
held to have greater appropriateness (cf. I.4.14).
The accrual of episodic knowledge about situations and texts fosters
expectations about what is acceptable and effective in a given context.
People build strategies to fit those expectations and to control textual
The priorities of control result in the relative dominances of surface features,
e.g. word class proportions and syntactic complexity.
These surface dominances gain the status of heuristic patterns against which new
texts can be matched.
The patterns may exert influence back on the control strategies applied to
situation management (I.3.4.6).
In this view, text types cannot be defined independently of pragmatics (Dressler
1972: 95; Kummer 1972a; Schmidt 1972; notwithstanding Grosse 1976b: 119). People
use text types as fuzzy classifications to decide what sorts of occurrences are
probable among the totality of the possible (cf. IV.1.23.3). As such, the text
type can be defined only as strictly as considerations of efficient
applicability allow. Unduly stringent criteria, like the rigorous borderline
between sentences and non-sentences, can either (1) open up endless disputes
over the admissibility of unusual or creative texts to a type, or (2) lead to so
many detailed types that any gains in heuristic usefulness are lost. It has
often happened that preconceived notions about a text type have led people to
reject a particular text which later became an acclaimed and classic
representative. The history of literature is filled with examples.
Two approaches to the TYPOLOGY of texts are readily evident (Schmidt 1978:55).
First, one could begin with the traditionally accepted text types, e.g.
narrative, descriptive, literary, etc., and seek to define distinctive traits of
each one; second, one could undertake to define a theory of texts independently,
and then observe whether one obtains a workable typology. The issue may have to
be resolved by a compromise: in the development of a text theory, the
applicability to text typology should be envisioned such that traditional types
become definable. I shall adopt this approach here.
Perhaps the following definition of the notion might prove useful for further
text type is a distinctive configuration of relational dominances obtaining
between or among elements of: (1) the surface text; (2) the textual world; (3)
stored knowledge patterns; and (4) a situation of occurrence.
relevant dominances can apply to elements of any size, according to the
circumstances. Without stipulating exactly what a text must look like for a
given type, these dominances powerfully influence the preferences for selecting,
arranging, and mapping options during the production and processing of the text.
We can at most obtain FUZZY sets of texts among which there will be mutual
overlap. Some textual traits will be DOMAIN-SPECIFIC, i.e., peculiar to the
situation, topic, and knowledge being addressed.
1.8 Some conventional categories of texts in our own culture (on some very
different cultures, cf. Grimes 1975) could be explicated along these lines:
In DESCRIPTIVE texts, the CONTROL CENTERS in the textual world are in the main
object and situation concepts whose environments are to be enriched with a
multiple directionality of linkage. All link types of state, attribute,
instance, and specification will be frequent. The surface text will reflect a
corresponding density of modifier dependencies. The most commonly applied global
knowledge pattern will be the frame.
In NARRATIVE texts, the control centers in the textual world are in the main
event and action concepts which will be arranged in an ordered directionality of
linkage. The link types of cause, reason, enablement, purpose, and time
proximity will be frequent (cf. VIII.2.13). The surface text will reflect a
corresponding density of subordinating dependencies.1
[Accordingly, the version (164b) of the Hawkins protocols seems to be a more
fitting narrative than version (163) with ‘and’ used throughout (cf V.7.3).]
The most commonly applied global knowledge pattern will be the schema.la
[1a. Freedle and Hale (1979) show that a narrative schema, once learned, can
easily be transferred to the processing of a descriptive text on the same topic.
In ARGUMENTATIVE texts, the control centers in the textual world will be entire
propositions which will be assigned values of truthfulness and reasons for
belief as facts (cf. IV.1.23.1); often there will be an opposition between
propositions with conflicting value and truth assignment. The link between types
of value, significance, cognition, volition, and reason will be frequent. The
surface text will contain a density of evaluative expressions. The most commonly
applied global knowledge pattern will be the plan whose goal state is the
inducement of shared beliefs.
In LITERARY texts, the textual world stands in a principled alternativity
relationship to matchable patterns of knowledge about the accepted real world.
The intention is to motivate, via contrasts and rearrangements, some new
insights into the organization of the real world. From the standpoint of
processing, the linkages within real-world events and situations is
PROBLEMATIZED, that is, made subject to potential failure (cf. I.6.7), because
the text-world events and situations may (though they need not) be organized
with different linkages. The effects would be an increased motivation for
linkage on the side of the text producer, and increased focus for linkage on the
side of the receiver. This problematized focus sets even “realistic”
literature (reaching extremes in “documentary” art) apart from a simple
report of the situations or events involved: the producer intends to portray
events and situations as exemplary elements in a framework of possible
In POETIC texts, the alternativity principle of literary texts is extended to
the inter-level mapping of options, e.g. sounds, syntax, concepts, relations,
plans, and so on. In this fashion, both the organization of the real world and
the organization of discourse about that world are problematized in the sense of
VII. 1.8.4, and the resulting insights can be correspondingly richer. The
increase of producer motivation and receiver focus will also be more intense, so
that text elements will be assignable multiple functions (cf. Schmidt 1971a).
In SCIENTIFIC texts, the textual world is expected to provide an optimal match
with the accepted real world unless there are explicit signals to the contrary
(e.g., a disproven theory). Rather than alternative organization of the world, a
more exact and detailed insight into the established organization of the real
world is intended. In effect, the linkages of events and situations are
eventually de-problematized via statements of causal necessity and order.
In DIDACTIC texts, the textual world must be presented via a process of gradual
integration, because the text receiver is not assumed to already have the
matchable knowledge spaces that a scientific text would require. Therefore, the
linkages of established facts are problematized and eventually de-problematized.
In CONVERSATIONAL texts, there is an especially episodic and diverse range of
sources for admissible knowledge (cf. VIII.1.4ff.). The priorities for expanding
current knowledge of the participants are less pronounced than for the text
types depicted in VII.I.8.4-7. The surface organization assumes a characteristic
mode to reflect changes of speaking turn (cf. VIII.1.2ff.; VIII.1.18).
Even within this modest typology, we can see that types cannot all be explicated
along the same dimensions. Whereas there may well be dominances of concept and
relation types for descriptive, narrative, and argumentative texts, the concept
and relation types in the other text types are probably domain-specific in the
sense of VII.1.7. Moreover, description, narration, and argumentation will be
found in various combinations in the other text types. And finally, if text
types are dependent upon situational settings (cf. VII.1.4ff.), the basic
question is how people use CUES to assign texts of various formats to a given
People can seek cues outside the text itself. Some situation types are
institutionally defined regarding the text types to be used, e.g. a church
service (in Pike 1967). Explicit announcements may establish the situation type,
e.g. a political gathering. Appearances of particular speakers or of a
writer’s name in print can activate expectations about the forthcoming text
type. A printed format, as in poems or newspapers, or a characteristic title,
such as for drugstore novels, may be influential. Even a specific topic, such as
those in many technical reports, can act as a cue. In accordance with what I
hold to be a general principle of human processing (cf. III.4.15; IV.1.10;
IV.2.9), the less evidence there is in the immediately apperceived text, the
more the text receivers will gather and utilize all kinds of cues.
A single text can indeed be shifted from type to type by altering its situation
of presentation. For example, it has become fashionable to “find” poems by
removing texts from their original environments (Porter 1972), such as cooking
recipes (Nöth 1978: 29f.) or classified advertisements (Kloepfer 1975: 88).
Conversely, poems are converted into advertisements (Reiss 1976: 70). Although
the text remains stable, the audience’s processing procedures are placed under
different controls and priorities. A non-poem presented as a poem is subjected
to the intensified assignment of multiple communicative functions to language
options (Schmidt 1971a, 1971b; Beaugrande 1978b). Presented as an advertisement,
a poem undergoes an impoverishment of the functions of its elements.
For a linguistics of texts as communicative occurrences, the issue of text types
is one of global processing controls. People are probably able to utilize texts
without identifying the type, but efficiency suffers, and the mode of
interaction of speaker/writer and hearer/reader remains vague. It seems unlikely
that we can throw away the traditional text types; after all, they have
functions in language users’ heuristics. Here as in many other areas, we may
instead have to throw away the hopes for air-tight, exhaustive, and mechanical
sorting techniques that consult only formal features without regard for human
THE PRODUCTION OF TEXTS
In comparison to comprehension, the production of texts has been left unexplored
(Fodor, Bever, & Garrett 1974: 434; Goldman 1975: 289; Osgood & Bock
1977: 89; Rosenberg 1977: xi; Levin & Goldman 1978: 14; Simmons 1978: 26).
The plausible reason is that linguists’ analysis can be taken as a model for
language understanding much more readily than for language production (II.2.4).
If we take linguistics too literally, the production of utterances seems like a
miracle of computation (cf. II.1.2f.). R. Jacobs and Rosenbaum (1968: 286)
criticize traditional grammar for conveying the impression that “human
language is a fragile cultural invention, only with difficulty maintained in
good working order.” But the transformational grammar advocated by Jacobs and
Rosenbaum is infinitely more fragile with its endless lists of rules that can
barely be kept under control and, to this day, have never been assembled into a
complete grammar for any language (cf. Achtenhagen & Wienold 1975: 9f.).
It might seem desirable to have a language model that uses the same procedures
for both the reception and the production of texts (cf. Klein 1965; Harris 1972;
Simmons 1973; Simmons & Chester 1979). The mapping between the surface text
and the underlying text-world would then be SYMMETRICAL in either direction
(Simmons & Chester 1979). However, this procedurally advantageous approach
would not be plausible for humans. Mapping is, in some ways at least, clearly
asymmetrical in textual communication (cf. I.6.12; III.3.5)—even people with
good memories will report what they have heard and understood in words differing
slightly from the original presentation. However, there is probably considerable
symmetry among the operations of mapping from one level to another and back
again (cf. VII.2.11). In practice, pilot programs for generating have usually
been performed with what Goldman (1975: 290) calls “canned output”: a small
repertory of expressions that forces everything into the same format. An
alternative with more varied options selected by weighting probabilities has
been developed by Sheldon Klein and co-workers (1973) (cf. the “weighted
filters’ for paraphrasing advocated by Mel’čuk & Žolkovskij 1970).
But a more detailed, circumspect model of the motivations for selecting a
particular option (some of these will be outlined below) is still needed.
Procedurally, reversals of operations would cover some differences between
production and reception of texts, but by no means all.2
[2. The processing via a transition network would foresee parallel
control in prediction of occurrences, but reversed control in stacking and
building the network (cf. 11.2.7ff.).] A text producer has to map a plan onto
conceptually relational content, and the content onto a surface format; the
receiver maps the surface back to the content, and the content back to the plan.
But it is surely an idealization to suggest that the receiver arrives at the
same material which the producer started out with. In some cases, the producer
would prefer keeping the plan secret or creating the impression of a quite
different plan. The receiver may also adopt an unexpected personal outlook on
the presentation. Reversability is furthermore not applicable to the textual
operations in which production and reception are running in parallel: the
producer monitoring the reception, and the receiver predicting the production.
And the production involves much more active selection and decision processes
which consume more resources and attention than does reception.
These considerations suggest that text production can only be treated by a
linguistics of actualization. The older linguistic methods oriented toward
identification, generalization, and description (cf. I.1.10f.) were purely analytic,
whereas a linguistics for explanation, reconstruction, and management, such as
is needed to study text production, must also have a synthetic outlook.
Consider the issue of misfunctions. We can recognize fairly clearly cases where
our own texts have been misunderstood, and we can discover the causes in factors
like surface ambiguities or misleading expectations. But it is vastly more
difficult to recognize when a text has been misproduced, i.e., when operations
have been wrong rather than merely inefficient, ineffective, or inappropriate
(I.4.14). If we count ambiguities as mistakes, we end up with vastly fuzzy
borderlines, because language options are systematically ambiguous in their
potential, and their usages possess variable degrees of determinacy. If we count
ungrammatical surface formats, we include occurrences such as Milton’s famous
warning to follow Christ, or else:
Him who disobeys, me disobeys. (Paradise Lost, V, 611-12)
deal with misfunctions, we evidently need a language model which does not simply
discover and analyze structures, but also relates structures to processes
operating with greater or lesser satisfactoriness.
In face-to-face communication, decisions and selections often have a provisional
character. The speaker may reconsider and introduce revisions when difficulties
ensue—”self-initiated repair,” according to Schegloff, Jefferson, and
Sacks (1977). Due to numerous factors competing for limited time and processing
resources, the operational load for spontaneous speaking may become unduly
heavy. People expect, on the other hand, much more controlled organization in
written texts, where the producer has time to discover and develop an efficient
and effective arrangement. If processing is overloaded during the initial phase
of expression, the producer has opportunity of going back and reviewing results
with a specially distributed focus. Hence, merely writing down the same
utterances one might produce in conversation—a frequent practice of untrained
writers—should not be expected to result in satisfactory texts. Writing
demands that situational factors, such as intonation, gestures, facial
expression, and immediately available feedback, be given compensation by factors
specific to written organization. The aspects of participant roles, time, and
location seem unproblematic in face-to-face communication, where people are
immediately present. In writing, they too must be accounted for by the
organization of the text world and its expression.
Like reception, production must involve a satisfaction THRESHOLD where
operations are TERMINATED (cf. I.6.4). Just as a receiver might go on and on
with inferences and spreading activation, a producer might keep revising a text
over and over. At some point, a decision to cease must be made, based on the
intended effects of the text on its audience; taken by itself, production
appears to be an open-ended operation. I shall attempt to sketch out the various
phases of this operation before going on to actual samples. I shall be concerned
in particular with the production of written texts (for a more thorough
treatment, see Beaugrande, in preparation).
The production process can be seen to consist of PHASES.3
[3. I note a different, simpler phase model proposed by Milic (1971) in
VII.2.38. For a full elaboration of my own model, see Beaugrande 1984] The
phases are presumably not separate operations in a time sequence, but rather
stages of PROCESSING DOMINANCE during which some operations are accorded more
resources and attention than others. I would distinguish at least four phases:
PLANNING, IDEATION, DEVELOPMENT, and EXPRESSION. During the PLANNING phase, a
text producer focuses on the PURPOSE of the text as a step toward a personal,
social, or cognitive GOAL, and on the intended AUDIENCE of text receivers. A
TEXT TYPE is selected, and correlations set up between the various component
steps of the plan and the general criteria of the production process. I use the
term RELEVANCE to designate these correlations: knowledge or discourse is thus
not inherently relevant, but relevant only with respect to a task at hand (cf.
The IDEATION phase places processing dominance upon the discovery of CONTROL
CENTERS for cognitive content. An IDEA is the internally activated configuration
of concepts and relations, which lies at the foundation of meaning-creating
behavior, including text production. It is extremely difficult to judge how
ideas originate, because the operations involved are in part at least beyond the
reach of conscious control. As a comparison, we could envision the focus of
attention as a beam of light sweeping across an enormously elaborate network of
knowledge; whatever the beam hits becomes active and can be inspected with
regard to its RELEVANCE. To write a friendly letter, the ideation phase could
cast about for material bearing the traits of INTERESTING (i.e., not obvious as
a matter of course) and EPISODICALLY RECENT (i.e., experiences in one’s
personal environment that the text receiver would not already know). To write a
scientific text, ideation could focus on a pre-decided knowledge space with its
own dense internal connectivity. To write a news report, ideation would be
directed toward the episodic storage of a situation or event sequence. To write
a novel, the ideation of situations and events would be substantially less
controlled by episodic storage of the producer.
These early production phases of planning and ideation need not be dependent
upon language. The raw materials feeding into production are essentially points
and pathways of knowledge: concepts, relations, mental images, states of the
world (past, present, projected), emotions, desires, and so on. The
correlation of all of these entities among themselves and with natural language
expressions is, I suspect, accomplished via respective modes of PROBLEM-
SOLVING: search, testing, and traversing of access routes. To attain COHERENCE,
the access routes are established among knowledge points; to attain EXPRESSION,
access routes are established between knowledge points and language expressions;
to attain COHESION, access routes are established among expressions within a
surface format (cf. I.4.4); and to attain RELEVANCE, access routes must be
established between knowledge points or expressions (or whole configurations of
these) and the steps and conditions of the producer’s plan in the current
Although there is surely considerable ASYMMETRY among these various accessing
operations (cf. I.6.12), I would view the operations themselves as comparable;
while the materials to be managed differ, the SYSTEMATICITY of their management
is unified by a common commitment to search, access, and connectivity. The
operations all require CONTROL CENTERS that determine the DIRECTIONALITY of
search and access (cf. II.2.9; III.3.6; VI.3.5; VII.1.8.1ff.; etc.). They all
vary according to degree of detail from LOCAL to GLOBAL (cf. VI.1.1, etc.) and
from MICRO-elements to MACRO-elements (cf. II.2.9; III.4.27;. VI.4.7; etc.).
They all work toward a THRESHOLD OF TERMINATION where processing is deemed
satisfactory for the task at hand (cf. I.6.4; III.3.3; III.3.23f.; IV.1.6;
VII.2.7; etc.) These common factors only emerge in the synthetic outlook I
advocated for a linguistics of textual communication in VII.2.4, where
structures and rules are interpreted as processes and procedures (cf. I.3.5.8).
The DEVELOPMENT phase receives the results of planning and ideation, whether or
not language expressions are already in sight at this moment. This phase is
responsible for the detailed internal organization of concepts and relations. To
the extent that this organization is not stored as determinate or typical
linkage, ORIGINALITY results. Originality may even lead to the creation of
totally new concepts like “chaos theory”. However, I suspect that if we go
into sufficient detail, we may find even new concepts to be composed mainly of
established materials put together in new ways (cf. IV.3.14). As the development
phase goes forward, the control centers passed on from ideation continue to
spread and intersect. If the conceptual-relational configuration were mapped
into expression at an early stage, we would have a terse text that would appear
as an OUTLINE (not fully cohesive) or a SUMMARY (fully cohesive) of the text
that would be mapped out at a later stage;4 [4. The difference
between outline and summary would be that the outline possesses a fragmentary
surface structure, and the summary a regular one (e.g. more complete
sentences).] at a still later stage, the summary relationship would be
reiterated, and so on indefinitely. The oppositions of local/global or
micro/macro are thus relative to the scope of the perspective we adopt.
Some typical operations of development could be carried out via linkages of
specification and instance, e.g. ‘people – young people – my friends –
my best friend’. Time and location can be subdivided into steadily smaller
components, e.g. ‘what I did this year – what I did this summer –what I
did on a weekend at the seashore’. Or ‘life in the south-life in Florida –
life in Miami’. The priorities of development, including link types, are
strongly controlled by the text type, e.g. descriptive, narrative, or
argumentative (cf. VII. 1. 8ff.). A further factor influencing development is
the use of the global knowledge patterns we explored in chapter VI: frames,
schemas, plans, and scripts. These patterns act as channels for spreading
activation, alerting the writer about what components require specification in a
relevant context. For such a pattern to become active, TOPIC configurations need
to emerge from densities of linkage in the ongoing textual world (cf.
III.4.11.9). The type of linkage will affect the type of pattern, e.g. states,
attributes, parts, etc, for frames; event or action progressions for schemas;
pathways of goal-attainment for plans and scripts. The variables in the pattern
will be filled in with applicable individuals. Some modifications might also be
required to make the intended content fit. Nonetheless, the PROCEDURAL
ATTACHMENT of the pattern to the planned output makes decision and selection
much more efficient (VI.1.5).
Global patterns, though channelling the development of the text-world model, do
not necessarily determine the format of the surface text. The most supportive
pattern is the schema, which provides an ordered progression of underlying
events and actions. The writer is free to express those events and actions in
some other order than their temporal and/or causal sequence, provided signals
are given; but the pattern offers guidance even then. The frame, in the sense I
use the term here, is less obvious in its ordering. To describe a scene or a
room, a writer has some typical strategies, such as moving from higher to lower,
central to peripheral, mobile to stationary (cf. IV.2.3ff). Yet these strategies
could compete with each other, and they might fail to respect the nature of the
scene components per se, i.e. relative importance from a human perspective such
as a plan. As a result, the normal ordering strategies for text world content
are usually applied EPISODICALLY, in response to the demands of context and
interest. Consider, for example, Dickens (1836-37: 35f.) depicting a newly
It was a careworn-looking man, whose sallow face and deeply sunken eyes
were rendered still more striking than nature had made them, by the straight
black hair which hung in matted disorder half-way down his face. His eyes wore
almost unnaturally bright and piercing; and his jaws were so long and lank, that
an observer would have supposed that he was drawing the flesh of his face in,
for a moment, by some contraction of the muscles, if his half-opened mouth and
immovable expression had not announced that it was his ordinary appearance.
Round his neck he wore a green shawl, with the large ends straggling over his
chest, and making their appearance occasionally beneath the worn button-holes of
his old waistcoat. His upper garment was a long black surtout; and below it he
wore wide drab trousers and large boots, running rapidly to seed.
can observe here a number of strategies for describing a person. The general
direction is to begin with the face and move from there to the clothes, working
gradually from highest (shawl) to lowest (boots). Superposed on this
conventional design is a focus on unusual features: eyes that are ‘deeply
sunken’ and ‘piercing’, and jaws that are ‘long and lank’. An episodic
comparison to a man contracting his facial muscles follows for emphasis. As the
writer passes on to describe the clothes, focus is directed to cues indicating
poverty and neglect. The spatial ordering is generally preserved: ‘neck –
shawl – chest – beneath […] waistcoat – surtout – trousers –
boots’. The writer’s selection is motivated by his plan to introduce shortly
afterwards a remarkably dismal tale of ‘want and sickness’ narrated by this
This illustration suggests how the EPISODIC tendencies of organizing text-worlds
can be controlled by DIRECTIONALITY. No writer would want to describe every
aspect of someone’s appearance, and Dickens’ careful depictions are more
detailed than the average. The writer should fovus attention on to those
portions of the available material which are INTERESTING (i.e. not predictable),
and RELEVANT (i.e. fitting to the plan for guiding the presentation of the
textual world along a given course). The untrained writer is hard-pressed to
make a selection and shifts about in a maze of episodic pursuits, assembling a
mass of discontinuous or superfluous details. Notice the unity of the Dickens’
passage despite the divergent descriptive strategies. The features he mentions
are relevant not merely by belonging to the same character, but by suggesting a
consistent impression of the ‘striking’ (interest) and ‘careworn’
(writer’s plan) aspects.
Like the development phase, the EXPRESSION phase during which the actual surface
text emerges is subject to a range of control factors. I would propose at least
three CONTROL LEVELS that are important for the mapping operations in the
The ORGANIZATION of EVENTS, ACTIONS, SITUATIONS, and OBJECTS in the textual
world exerts certain influences upon the organization of the surface text. I
reviewed in IV.2.6ff. the experimental literature regarding this issue, such as
the strategies of moving from higher to lower, central to peripheral, changing
to unchanging, mobile to stationary, earlier to later, and so on. I cited in
III.4.18 some correspondences between text-world organization and the use of
tense, voice, and mood. Harald Weinrich (1977) shows how tenses in French tend
to convey either a descriptive or a narrative perspective on the textual world.
Halliday and Hasan (1976: 40) suggest that the characteristics of objects are
cited in a certain order when modifiers are linearized in a noun phrase such as:
(191) two high stone walls along the roadside
number is followed by size and substance (on modifier positioning, see also
Vendler 1968; Martin 1969; Danks & Glucksberg 1971).
The STANDARD SEQUENCING OPERATIONS for imposing a linear format on English texts
must be respected. I suggested in section II.2 that the basic phrases, clauses,
and sentences of English act as frameworks for judging what surface occurrences
are probable. These frameworks are not obligatory, but they must be kept in mind
when departing from them, because they still act as a means of orientation. The
correlations between sequencing operations and text-world organization can be
arbitrary on occasion. In English, it is customary to place expressions of
location before those of time, while in German, the reverse order is preferred;
yet the event or situation may be the same. Such formatting standards reduce the
decision-making load, not so much concerning what to say as when to say it.
Sometimes, sequencing operations require conceptually empty placeholders without
justification in the text world, e.g. the dummy ‘it’ used for expressions of
the weather (cf. V.5.4.2). And the LINEARIZATION PROGRAMS often fail to reflect
grammatical dependencies via direct adjacencies (II.2.7ff.).
2.16.3 The INFORMATIVITY of text-world entries can also affect the order in which they are expressed in surface formatting (cf. IV. 3). The general trend is to mention new or focussed material after known or marginal material. For expressing configurations of familiar content, sentences will generally be longer and more complex than for expressing unfamiliar (cf. IX.4.6); perhaps the more problematic coherence of the unfamiliar content is compensated via a less problematic cohesion. The distribution of focus depends not only on the internal linkage of knowledge (whether pathways are predictable vs. problematic), but also on the RELEVANCE of knowledge organization to the producer’s plan (cf. VII.2.8): the arrangement of materials is co-ordinated with the ordering of steps in a plan (see for instance Dr. Haggett’s utterances among (96-104) in the stage play of VI.4.17). This control level thus applies not only to what to say and when to say it, but also to why. Let us pursue the interaction of these control levels by envisioning how a writer might describe a simple event sequence. The writer observes a man leading a dog whose bright-colored collar attracts a child; in order not to be grabbed, the dog breaks its leash to escape. The man spanks the child. If we arrange this much content in a network of concepts and relations, we might obtain Figure 28.
three animate agents: man, dog, and child, appear as object nodes with their
respective actions and attributes. I include some descriptive traits, e.g.
‘old’, ‘ugly’, ‘small, for purposes of demonstration. The writer’s
task is to find a surface expression, that is, to find a sequential connectivity
that captures the conceptual connectivity of the text world. This is a special
instance of PROBLEM-SOLVING: mapping out points in problem space according to
the already solved and connected points in a problem space on a different level
(cf. VII.2.10).5 [5.
Burton’s (1976) “semantic grammar” functions by using these two levels in
close correlation; compare also the “cascading networks” (Woods &
Brachman 1978b) depicted in III.4.14]
The easiest solution would be to chop up the network into individual events and
parse each event onto a surface structure according to PREFERENCES. The agent of
an action (or, for inanimate objects, the instrument) is then mapped onto the
grammatical subject, the event/action onto the verb, and the affected entity
onto the direct object (cf. Bever 1970).6
[6. This strategy does not apply to ‘ergative” languages, in which an
“ergative” case for agency is differentiated from a ‘nominative’
(Dressier, personal communication).] To include the third control level, that of
informativity, we stipulate that the agent or instrument be known, and the
action or affected entity be new. This could result in the following text:7
[7. For still greater banality (cf. 1. 1. 16), we could start out with
‘kernel” sentences:’ ‘I saw a man. The man was old.’ [etc. ad nauseam]
I saw an old man. (192.2) He was leading an ugly dog. (192.3) The dog was
wearing a bright orange collar. (192.4) The collar attracted a small child.
(192.5) The child grabbed at the dog. (192.6) The dog broke its leash. (192.7)
The leash hurt the man’s hand. (192.8) The hand spanked the child.
surface text is perfectly clear and cohesive, and there are no obstacles to
coherence and comprehension. The “process-type” actions (in terms of
Halliday 1967a) are expressed with the continuous form (‘be’ + verb + ‘ing);
the “uniplex” actions (in terms of Talmy 1978) are expressed via the simple
forms (here, simple past). Notwithstanding, the text is objectionable. It is
monstrously uninteresting to read, precisely because the continued use of
preferences makes such a predictable and repetitious pattern. The mapping is not
efficient because each underlying node has to appear so often in surface
structure: ‘dog’ in four sentences, ‘man’ and ‘child’ in three each,
and ‘collar’, ‘leash’, and ‘hand’ in two. To suggest how our network
from Figure 28 has been divided up for the surface text, I partitioned the
diagram of Figure 29 with dotted-line spaces for each sentence (small numbers
are sentence numbers) (on network partitioning, see Hendrix 1975, 1978).8
[8. Each space encloses its nodes and link labels. The numbers of dots of
enclosing lines match the sentence numbers of (192).]
can observe that REDUNDANCY is graphically visible partition overlap. It might
be a general definition that redundancy can be formalized as the overlap of
systemic unity partitioning of an actualization network for the next
The writer has good reason to be dissatisfied with this particular mapping. Let
us consider how an alternative version can be generated which, being based on
the same network, counts as a PARAPHRASE of the first (cf. III.3.11.10). This
new version will adopt a more flexible standpoint regarding event boundaries and
interestingness, and will cut down on redundancy. This procedure is not
comparable to sentence transformations, though the latter are also of paraphrase
character (II.1.11): transformations are done by an autonomous syntax in which
interestingness or efficiency of communication play no distinct role.
Let us follow the production of the new version along and observe how decisions
are made. The opening sentence is a strategic place to introduce the topic (cf.
VII.3.9). Here, the topic is not so much the old man, as (192.1) would imply,
but the events involving the dog being on a leash. It is therefore expedient to
load this topic material onto the opening sentence, yielding:
1) I saw an old man leading an ugly dog.
predicate has been expanded with a participial modifier, so that the topic
material is effectively located toward the end of the sentence. A gain in
efficiency is also attained by reducing the number of sentences, and hence the
number of focusable predicate slots; moreover, the single occurrence of
‘man’ in (193.1) supplants the redundant occurrences in (192.1-2) with no
loss of clarity.
The next task is mapping out the events involving the collar. Because a collar
is typically in “containment’ of a dog, and a determinate
“instrument-of’ leading a dog, there is no motive to assert the collar’s
presence in a separate sentence such as (192.3): a predicate slot is wasted on
content that is easily predicted. Instead, the predictable relation can be
mapped onto a possessive modifier dependency and the, predicate slot filled with
the unpredictable event in which the collar figures as instrument:
The dog’s bright orange collar attracted a small child.
this paraphrase saves resources by conserving a predicate slot and cutting out a
second expression for ‘collar’ vis-à-vis (192.3-4).
The TURNING POINT of this little story (cf. VIII.2.7) is composed of the events
of grabbing and leash-breaking, because these deflect the course of things most
decisively. As I shall argue in Chapter VIII, the turning point of a story is
usually accompanied by MOTIVATIONAL STATEMENTS that justify the central actions
(cf. VIII.2.23ff.; VIII.2.32). We might want to state the relation between the
‘grabbing’ and the ‘breaking’ that is left implicit (192.5-6). A
subordinative junctive preferentially indicative of purpose will do:
In order not to be grabbed, the dog broke its leash and ran away.
‘run away’ action can be derived from the original text-world model via
inferencing based on knowledge of purposes. This presentation helps to integrate
the least predictable event into the reader’s ongoing text-world model.
The final task is mapping out the conclusion. The ‘hand’ node is linked to
one event as ‘affected entity’ and to another as ‘instrument’
‘If we want to map out the node only once onto surface expression, we
need a sentence format that has a passivizing and an activizing constituent. The
passivizing construction is preferentially the passive voice, or the past
participle. We could then have a choice between:
The man’s hand was hurt by the sharp tug and spanked the child.
Hurt by the sharp tug, the man’s hand spanked the child.
decide between them, a writer should consider the knownness or inferrability of
the underlying events. The stronger those factors are, the less likely one is to
devote a separate subject-predicate construction to the surface expression of
the event. Because breaking a leash is very likely to hurt the owner’s hand,
(193.4b) seems the better selection. It has the added advantage of a surface
symmetry with ‘hand’ located between the expressions of the two events in
which it was involved, thus enacting the balance where one hurt is the reason
for the inflicting of another.
Following the decision process along as shown, we arrive at this version:
I saw an old man leading an ugly dog. (193.2) The dog’s bright orange collar
attracted a small child. (193.3) In order not to be grabbed, the dog broke its
leash and ran away. (193.4) Hurt by the sharp tug, the man’s hand spanked the
still pretty trivial in its content, (193) is better reading than (192) in its
form. The redundancy of (192) has been dramatically cut down: ‘dog’ in three
sentences, not four; ‘man’ and ‘child’ in two each, not three; and
‘collar', ‘leash’, and ‘hand’ in one, not two (cf. VII.2.18). The
savings allows the addition of some further material in (193), e.g. ‘in order
to’ and ‘ran away’, yet the total word count is still less than (192): 43
versus 47. The new version is thus more READABLE then the old, conveying as much
with fewer expressions, yet maintaining interest by motivated variety of
structuring. The partitioning of the textual world into sentence-length spaces
for (193) is illustrated in Figure 30.
The lower redundancy appears as reduced overlap.
Strategic control upon decision and selection is crucial. The mere loading of
more material onto more intricate sentence frameworks offer no certainty of
producing a worthwhile text. Untrained writers, who want to break out of the
monotony resulting from using the same mapping over and over, may fail to retain
the necessary control. Consider the following expression of the same text-world
model loaded onto a single sentence:
An old man I saw whose dog’s leash, attached to a bright orange collar,
attracting a small child who grabbed at the dog that broke its leash, hurt his
hand, spanked the child.
uncontrolled overloading of the sentence structure yields two participial
dependencies and four relative clauses. This structuring is in principle
allowable by rules of syntax proper. It cannot be the task of a grammar to state
at what length or degree of complexity a sentence is no longer allowed for a
language (188.8.131.52). Moreover, redundancy has been reduced still lower than in
(193): ‘dog’ mapped only twice and ‘man’ only once. Still, the text is
far less readable than (192) or (193). Too many events are packed into modifiers
and relatives as if they were already known to the reader. Distinctions between
predictable and non-predictable events are flattened. The reader’s attention
is scattered all around with no cues as to what might be important. For example,
the phrase ‘the dog that broke its leash’ wrongly suggests that a previously
mentioned dog with this action should be in the reader’s active storage.
These samples illustrate how writers must correlate the strategies used on the
three control levels depicted in VII.2.16. After noticing and developing the
internal structure of the events, the writer has to utilize the sequencing
operations of English in accordance with reasonable rates of informativity. The
arrangement of expressions for knowledge depends decisively on what the reader
is expected to know and to find interesting. The writer cannot fill in every
detail, nor make every underlying relation explicit. The writing and revising
process is terminated with a reasonable balance between what is said and what is
known, between what is said and what can be supplied by spreading activation or
inferencing, and between what is informative and what is dispensable.
To produce a text of enduring quality, substantially more extensive processing
is demanded. Both the original search for knowledge and the subsequent mapping
must be carried on with great circumspection. I shall attempt to follow the
processes that might create the following Shakespearean sonnet (number 33)
(discussed also in Beaugrande 1979e, 1979i):9 [9 Here, as in
all other Shakespeare quotations in this volume, I follow the Kittredge edition
Full many a glorious morning have I seen
Flatter the mountain-tops with sovereign eye,
Kissing with golden face the meadows green,
Gilding pale streams with heavenly alchemy;
Anon permit the basest clouds to ride
With ugly rack on his celestial face,
And from the forlorn world his visage hide,
Stealing unseen to west with this disgrace:
Even so my sun one early morn did shine
With all-triumphant splendour on my brow;
But, out, alack! he was but one hour mine,
The region cloud hath masked him from me now.
Yet him for this my love no whit disdaineth,
Suns of the world may stain where heaven’s sun staineth.
The writer’s problem for this text is especially delicate: to present the
poetic expression of a complaint to a particular addressee in such a flattering
way that reconciliation is by no means precluded. The PLANNING phase decides to
characterize some of the addressee’s actions negatively, yet without direct
confrontation. The underlying macro-structure of events for this communicative
situation is built along these lines: (1) addressee treats speaker as friend;
(2) addressee changes to unfriendly treatment; (3) speaker enters a negative
emotional state; and (4) speaker complains, the text itself being the
The avoidance of confrontation can be navigated by strategies of role division
and content selection. The SPEAKER of the text (here ‘I’) is kept distinct
from the PRODUCER, and the ADDRESSEE (here ‘he’) from the audience of
RECEIVERS. The outcome is that the personal message fades into the
background—a common principle in literary and poetic communication. The
content is selected via ANALOGY. The actual event sequence is displaced by a
sequence from another topic domain and yet kept recoverable via strategic
placement of cues.
The planning phase sets up a pathway toward the goal: create a linkage among
entities of knowledge that will make the underlying event series discoverable
via PATTERN-MATCHING. The IDEATION phase accordingly searches knowledge stores
for a TOPIC IDEA that will he a CONTROL CENTER for a textual world entailing a
contrast of positive and negative events. The topic idea is readily accessed
from the UNIVERSE OF DISCOURSE (I.1.3) for Shakespeare’s cultural setting: the
workings of nature as the background of human activities. This general knowledge
frame offers some obvious contrasts; for example, day versus night comes to
mind, but is too irreconcilable and determinate, whereas an accidental contrast
would be more relevant to the writer’s plan. Accidental contrasts are
available in the unstable domain of ‘weather’ (especially in England). If
the TOPIC IDEA were ‘change in the weather’. the DEVELOPMENT phase can
easily attach the contents of a ‘weather-frame, e.g. ‘sun’, ‘sky’,
‘clouds’. and so on, along with their attributes, locations, motions, etc.
To suggest what a typical person’s ‘weather’-frame might look like, linked
onto a ‘landscape’-frame, I provide a network diagram in Figure 31.
seems safe to assume that at least this much commonsense knowledge is well
As the development phase continues, the positive state called for by the plan
can attach a favorable state of the weather; that state should also have an
early time indicator to match the early stage of the personal relationship
between speaker and addressee. It follows that the morning sunrise is a natural
selection, allowing the plan-relevant flattering attribute ‘glorious’. We
notice, that the underlying element ‘sun’ in the ‘weather!-frame is not
explicitly attached yet, but introduced via a further analogy: a
‘person’-frame. The “parts-of” a person that correspond to the sun
include those sharing the same “form,” such as ‘eye’ and ‘face’. The
‘person’-frame is exploited by the mention of humanlike actions:
‘flatter’, ‘kiss’, and ‘gild’. The first two of these suggest the
subclass ‘friend’, pointing back to the underlying macro-structure of events
In this fashion, several commonsense frames are attached concurrently to build a
text-world model. The opening stretch (lines 1-4) describes the morning light
and its effects on some typical elements of the landscape. But the selection of
expressions is so designed as to point the reader away from that domain toward
‘actions of a friend’; otherwise, entries lick ‘flatter’ and ‘kiss’,
being incompatible with the ‘weather’-frame, cannot be integrated into the
textual world. The phase of conceptual-relational development (VII.2.1 1) has
recovered some incidental knowledge from these frames, such as attributes,
locations, and parts. The cognitive outcome of this multiple attachment is to
impel the reader to recognize the analogy required by the writer’s plan:
events of the weather versus events in a personal relationship.
This design process becomes a pattern to be repeated in the next stanza of four
lines (5-8). The sunny morning is set in opposition to the ‘clouds’ which
block out the light. The opposition spreads outward into the attributes and
motions of clouds, rendering them uniformly negative: ‘basest’, ‘ugly’,
‘forlorn’, ‘steal, ‘disgrace’. A series of elements is presented, the
integration of which requires the interface of the ‘weather’-frame and the
‘person’-frame: ‘face’, ‘forlorn’, ‘visage’, ‘steal,
‘disgrace’. The analogies for asserting that ‘clouds’ can ‘ride on’
and ‘hide’ the ‘face’ of the sun that moves ‘to west’ are derivable
from the ‘weather’-frame-based knowledge about locations and motions. The
outcome is the complex structure of concepts and relations, many shared among
frames, represented in Figure 32.
concepts of ‘morning’ and ‘cloud’ appear as topic nodes with the
material from the first four and second four lines, respectively, being
connected on. We see the further connectivity between the two knowledge spaces
that result, including both equivalences and oppositions. The mastery of
Shakespeare as a text designer is attested in the multiple justification he had
for all of the selections and arrangements. He implements his global plan of
complaining via a textual world with an inherent apperceptual power of its own.
He presents high-informational occurrences as discrepancies and discontinuities
between elements of frame-based knowledge; in downgrading the occurrences, the
reader is irresistibly impelled to recover the planned underlying message.
The mapping of the textual world onto surface expression must also conform to
the formatting demands of the text type ‘sonnet’. This requirement creates a
special problem setting where cohesion must be managed such that a closely
patterned arrangement is obtained: (1) syntactic arrangement; (2) line
arrangement; (3) sound arrangement; and (4) lexical arrangement. Shakespeare’s
constitutive principle for all of these levels is above all EQUIVALENCE (cf.
Jakobson & Jones 1970). In regard to syntax, six lines contain the
configuration “preposition-modifier-head,” the preposition being ‘with’
in all cases (2, 3, 4, 6, 8, 10) (in line 8, there is a determiner rather than a
modifier). Three of those lines (3, 4, 8) also begin with a present participle
expressing an action belonging to the ‘person’-frame. The syntax also
interacts with the line divisions. The first two groups of four lines, and the
last three groups of two have a clear internal cohesion. The first eight lines
form a single sentence; lines 9 and 10 form another sentence; 11 and 12 are a
run-on sentence, easing perhaps the transition to the marked separation of
sentences in 13 and 14. These divisions accord well with the flow of content:
(1) positive early events (1-4); (2) negative events as a change (4-8); (3)
comparison of these events to the speaker’s own experience (9-12); and (4)
withdrawal of the complaint (13-14).
These divisions are characteristic of the ‘sonnet’ text type as employed by
Shakespeare (whence the enduring term ‘Shakespearean sonnet’). The couplet
at the end is often opposed to the rest in content and format. Here, in effect,
it deflects the whole impact of the statement so far; and it breaks the
alternating rime patterns with consecutive rime. The internal organization is
also reflected in the rhythm pattern. The first four lines have a syllable
distribution of 12-10-10-11; the second have 10-10-10-10; the third group has
11-10-11-10; and the couplet is 11-12. The four-line groups are thus all
distinctive, and the 12-syllable pattern of the first line returns in the
last—just as the speaker hopes that the harmonious early stage of the personal
relationship may return.
The careful interlocking of mapping options is, as we shall note, essential to
the writer’s plan. The lines (9-12) begin with the junctive ‘even so’ to
signal that knowledge from the first eight lines should be kept active and
re-applied. That signal is reinforced by lexical recurrences and equivalences:
‘morn’ (9) looking back to ‘morning’ (1); ‘splendour’ (10) to
‘glorious’ (1);’but one hour’ (11) to ‘anon’; ‘region’ (12) to
‘heavenly’ (4); ‘cloud’ (12) to ‘clouds’ (5); and ‘masked’ (11)
to ‘hide’ (7). Such extensive correlation supports the transfer of knowledge
from an already constructed model space to an ongoing one-an example of
TEXT-INTERNAL INHERITANCE via pattern- matching (cf. IV.4.5: V.7.1). The
intriguing aspect here is that the negative terms of lines (1 -8) have no
correlates in (9-12). The characterization of the addressee’s actions
regarding the speaker’s situation in (9-12) is accomplished entirely by
inheritance from (1 -8). Moreover, the writer is careful not to personify the
‘sun’ in (9-12). As a result, the complaint is delivered with the greatest
mediation and indirectness. Whatever negative comments are made about the
addressee are filtered through an ennobling analogy in which the latter figures
as nothing less than the ‘sun’. Of course, the ‘sun’ is not at fault if
‘clouds’ intrude; and the message is even so arranged as to keep some
surface distance between the ‘sun’ and those negative terms that do appear.
To conclude, the final couplet withdraws the complaint as inappropriate to a
being of such grandeur. The contrajunctive ‘yet’ in line (13) signals a
surface reversal, but the drift of content organization has been conciliatory
The total text-world model for the sonnet is diagrammed in Figure 33.
have drawn in the various recurrences, equivalences, or class inclusions that
render the text-world model so uniquely motivated in its design (cf. 1.4.14).
The exceptional density of linkage holding so many entities in place is
indicative of the extraordinarily skillful selection and decision processes of
the writer. The surface text is so designed as to elicit spreading activation
within several frames at once. The intersections, as such, are unforeseeable and
hence interesting, and yet convincing by virtue of their dense connectivity. The
processing that recovers such a configuration underneath the already intensely
structured surface expression is the foundation of AESTHETIC EXPERIENCE:
discovering a multiplicity of functions among elements of the message (cf. VII.
1.8.5), and overcoming problematic linkages by finding their motivations.
No one would deny the staggering difference between the Shakespeare sonnet and
the ‘ugly dog’ story. But I would surmise that the production processes for
both are analogous: a macro-structure of events is selected and developed
according to content-internal standards and criteria of informativity; the
result is mapped onto a surface structure under interactive controls. The
effectivity of the results varies because of the differences in expenditure of
processing resources. Louis T. Milic (1971) was led by such differences to
postulate two phases of text production: (1) the selection of “stylistic
options” needed to produce any surface structure at all; and (2) the making of
“rhetorical choices” by evaluating and improving upon what has been
generated. Milic concedes that these two phases cannot be separate in real
time—a point I have stressed for my own model with four phases. But I wonder
if Milic might be drawing a line along the inappropriate dimensions. A good
share of Shakespeare’s rhetorical power is antecedent to anything like the
selection of stylistic options: it originates in his ideation and development
phases, e.g. the interfacing of frames for ‘weather’ and ‘person’.
Milic’s scheme appears to imply the notion I rejected in IV. 1. 17 that
all metaphors have commonplace, literal equivalents.
My experiments regarding the production and reception of creative texts have
been inconclusive so far, due to empirical obstacles of obtaining creative
behavior under reliable conditions. In one set of tests run by Walter Kintsch
and co-workers, subjects who recalled a Shakespeare soloquy did undertake to
rephrase the content in everyday language. Those subjects who did not only
recovered a few bits of the original. Until conclusive evidence to the contrary
is obtained, I merely claim that creativity is an intensification of normal
production processes rather than something altogether different. The question of
whether all content can be accorded creative treatment remains in debate. Given
sufficient licence (and a bent sense of humour like mine), one could wring a
poem even out of the ‘ugly-dog’ story:
Not many a dotard gentleman I spy
disfeatured dog on lanky leash,
with collar orange a child nigh,
to ‘scape its rank rapacious reach;
stinging hand the man requites the prank,
the infant nether flank.
so did God in Eden new-made beasts display
our childish fancy in parade;
we who snatch and seize in wanton way
harrow hence the habitants He made.
we deem ourselves creation’s dears
blight the earth till heaven interferes.
rotten tosh, I know, but maybe it can help to suggest once again the endlessly
pliant relation of language to content.
RECALLING TEXTUAL CONTENT
Many years ago, Sir Frederick Bartlett (1932) obtained experimental evidence
that recall is not merely a REPRODUCTION of what people experience, but also a
Royer (1977) sees three positions, with ‘construction’ being more moderate
use of the text user’s own disposition. A possible means of reconciling these
positions is discussed in Beaugrande (1980c).] Since then, a series of
experiments showing pervasively accurate recall (e.g. Gomulicki 1956; R. Johnson
1970; B. Meyer & McConkie 1973) apparently challenge Bartiett’s viewpoint.
However, these new results are no genuine refutation of Bartlett. Conventional
psychological tests are routinely designed so that people have little motivation
to integrate the content of texts into their store of useable knowledge, because
the tests lack relevance to everyday life. Also, our educational system stresses
rote memory work so heavily that, placed in a formal test situation, people may
strain to the utmost in order to render every detail as exactly as possible. I
shall explore some new data and suggest ways in which reproduction and
We must bear in mind that a person’s recall protocol is a text in its own
right (Kintsch & van Dijk 1978a: 374). The production of the protocol, at
least under natural conditions, ought to entail the developmental and selective
processes I outlined in the foregoing section. The recently processed original
would of course be an important source. But if people are building their own
cognitive models of a textual world, their recall should naturally include
material they supplied themselves by spreading activation, inferencing, and
updating (cf. 1.6.4). They should be especially prone to add material if their
protocols would otherwise lack cohesion in expression or coherence in the
At first glance, VERBATIM recall ought to be straightforward beyond dispute.
Output looks exactly like input, so that we feel comfortable about considering
memory a mechanism of “trace
abstraction” (Gomulicki 1956) like a tape recorder or photographic plate. And
yet the possibility cannot be eliminated that seemingly verbatim recall could
result from reconstructive processes (cf. VI.3.12; VII.3.16). Suppose someone
understands a surface expression by recovering the appropriate concept. If that
concept had only the original expression as its plausible name, recall would
probably be verbatim. But we cannot conclude that the person abstracted a trace
of the surface structure and simply reproduced it. From this consideration, it
would follow that verbatim recall may be telling us more about the availability
of alternative names for the concept in a particular text-world than about the
general memory strategies of people at large (cf. VI.3.12).
3.4 To pursue the interactive roles of surface expression and text-world coherence in processing, I designed a reading experiment for the Computer Laboratory of Psychological Research of Walter Kintsch and co-workers at the University of Colorado. My test was conducted by varying versions of the same text (cf. Bower 1976; Jones 1977; Thorndyke 1977), although the parameters of variation were, as far as I know, rather unusual. I created five alternative versions of the ‘rocket’ text cited in III.4.20 to be presented to separate groups of readers, mostly first-year college students venturesome (or naive) enough to enroll in an elementary psychology course, which requires them to act as guinea pigs. No test subject saw more than one version. The five alternatives read as follows:10 [10. Further tests have revealed some weaknesses in the design of these samples. I have made improved versions and run them, including one in German for a group of German native speakers. The outcome is discussed in Beaugrande (1979d).]
[INVERTED] (1. 1) Empty, it weighed five tons. (1.2) For fuel it carried eight
tons of alcohol and liquid oxygen. (1.3) There it stood in a New Mexico desert:
a great black and yellow V-2 rocket 46 feet long.
Scientists and generals withdrew to some distance and crouched behind earth
mounds. (2.2) Two red flares rose as a signal to fire the rocket. (2.3)
Everything was ready.
Trailing behind it sixty feet of yellow flame that soon came to look like a
yellow star, the giant rocket rose slowly and then faster and faster amid a
great roar and burst of flame. (3.2) Radar tracked it at 3,000 mph when it soon
became too high to be seen.
As the rocket returned at 2,400 mph and plunged into earth a few minutes after
it was fired, the pilot of a watching plane saw it return to a point 40 miles
from the starting point.
[ORNAMENTAL] (1. 1) In a bleak New Mexico desert, a vast black and yellow rocket
towered 46 feet into the sky. (1.2) In order to lift this five-ton colossus into
space, eight tons of alcohol and liquid oxygen were stored in the fuel chambers.
(2. 1) Scientists and generals scrambled for cover behind mounds of earth as the
signal for launching blazed forth: two bright red flares.
Amid a deafening roar and a blinding burst of fire, the giant ascended with
mounting speed. (2.3) Its trail of yellow flame became a distant star poised on
the outer verge of human vision. (2.4) The eyes of radar alone could follow the
traveller’s flight at 3,000 mph.
High above the earth, a pilot watched from an observation plane as the rocket
retraced its path, slowing to 2,400 mph. (3.2) Only forty miles from the place
of departure, the huge aircraft came to rest. (3.3) The giant was home again.
[CONDENSED] (1. 1) With eight tons of alcohol and liquid oxygen as fuel to carry
its five- ton frame, a 46-foot black and yellow rocket stood ready in a New
Mexico desert. (1.2) Upon a signal of two red flares, scientists and generals
withdrew to crouch behind earth mounds. (1.3) With a trail of yellow flame that
soon resembled a star, the rocket ascended with increasing speed. (1.4) Radar
clocked it at 3,000 mph after it had passed out of sight. (1.5) Within minutes
an observation plane recorded the return at 2,400 mph and plunge to earth 40
miles from the launching site.
[DISORGANIZED] (1.1) It was in a desert in New Mexico where, forty-six feet of
black and yellow, a great rocket stood. (1.2) Of its thirteen tons of total
weight, five tons of empty weight were added to eight tons of fuel, this being
alcohol and liquid oxygen.
Behind mounds of earth scientists and generals, when everything was ready,
withdrew, crouching. (2.2) To fire the rocket, two red flares were given as a
With behind it sixty feet of yellow flame, the giant rocket rose with a great
roar and a burst of flame faster and faster after starting slowly. (3.2) Before
it became too high to be seen, the flame soon looked like a yellow star would
look. (3.3) But radar tracked it upward, speeding to 3,000 miles in an hour.
A few minutes after it was fired, the pilot of a watching plane saw its return
to be at a speed of 2,400 mph and plunge to earth 40 miles from the place where
it all started. (4.2) What goes up must come down.
[MISLEADING] (1. 1) In a New Mexico desert, a yellow and bleakly isolated rocket
stood already waiting for take-off. (1. 2) When empty, it had weighed five tons.
(1. 3) Now, when fuel, being alcohol and liquid oxygen, was added, it weighed
thirteen tons. (1.4) Ready to fly as a wild blue wonder, it stood there
motionless, waiting for the signal station to start the take-off.
When everything was red as the station, two warning flares sent scientist and
general alike to shelter areas provided at a distance pointed out by large
With a roar and a burst of flares, the giant rocked on its pad and then rose
colored fire traced its flight into the sky’s open space. (3.2) Behind it
trails its yellow path that soon comes to look just lightly distinct from a
star. (3.3) When it was too high to be a scene of human observation, it was
tracked by the reader of radar screens. (3.4) Its speed was clocked as 3,000
(4.1) A few minutes after, it returned, observation planes clocking it at 2,400 miles. (4.2) The rocket, descent aimed toward the starting point, plunged down to the earth 's surface 40 miles from the launching padded by landing gear.
Version (197) was produced by inverting stretches of text, so that the order of
presentation was turned around. Version (198), deployed ornamental expressions,
including apperceptually salient imagery and metaphors. Version (199) was
produced by compacting the original. Version (200) was jumbled by deliberate use
of poor planning, such as we find in hasty rough drafts. Version (201) was
designed to actively mislead readers in eliciting expectations that are
overturned by bizarre occurrences.
Half of the subjects were taped reading aloud; the other half read silently. The
texts were then removed, and subjects wrote down "everything they could
remember in their own words." The total numbers of propositions in each
version were calculated according to the usual Kintsch methods (cf. Kintsch
1974; A. Turner & Greene 1977), and the protocols were scored according to
the amount of propositions recovered.11 [11.
This count follows a list of propositions formatted like (31) in IlI.3.4: close
to the surface text (the text expressions are usually used as concept names).
But there are problems of formatting expressions such as ‘not particularly
different from’ (how many propositions?): My network format places only
concepts with their own substance onto nodes, and loads all relational signals
onto the links.]
Kintsch and I both expected that the variations in the text would make important
differences in ease of reading and recall. To our initial amazement, quantitative
recall varied across all six versions only to a statistically insignificant
extent! There was a rise up to 54% for version (198), and a drop to 41 % for
(200), but the original (35), and (197), (199), and (201) were all recalled with
a ratio between 43% and 47%. This finding suggests the powerful role of prior
storage and processing strategies in imposing cohesion and coherence, even when
deliberate obstacles are presented. Procedural attachment was evidently able to
offset the oddly arranged surface formats of (200) and (201), for example by
attaching the 'flight'-schema I discussed in V1.3. There were, however, some
intriguing qualitative differences in recall, which I shall review.
Because of the inversion of the opening paragraph, version (197) entails a
postponement of announcing the topic (compare the original in III.4.20). The
text opens with cataphoric pronouns (‘it’) for which the co-referent is
not supplied until after a delay (cf. V.4.9). The effect was a strikingly
different distribution of attention for this version than for the original. On
version (197), 8 out of 10 readers correctly recalled both of the fuels, whereas
only 3 out of 10 who saw the original did as much. The topic postponement
apparently forced readers to utilize the opening stretch of material heavily in
order- to identify a subsuming frame or schema; subsequently, the material was
better organized and more available for recall. Interestingly enough, a
replication of the test by Richard Hersh and Roger Drury at the University of
Florida found that a five-minute pause of non-activity
before the writing of protocols reduced this difference in availability down to
According to the mechanics of the “Von Restorff effect” (cf. IV.2.2),
unusual items in a presentation draw attention to themselves, but in exchange,
other items are learned less well (Posner & Rossman 1965; Waugh 1969).11a
[11a I later managed to obtain Von
Restorff ‘s dissertation in Berlin and found she had in fact viewed the issue
from the opposite side such items hinder the formation of a Gestalt—thus with
a reductive, not supportive effect.] We found here that total recall for
original and inverted versions was about the same. Whereas readers of the
inverted one did well with fuels, they did badly with the colors (‘black’
and ‘yellow’). Only 3 out of 10 recalled them, whereas 8 out of 10 readers
of the original remembered—exactly turning around the proportions for fuels.
We attribute this difference also to a noticeable favoring of initial and final
text elements across all six versions (compare Garrod & Trabasso 1973). This
finding calls to mind the superior learning of initial and final entries in
word-list testing (Murdock 1962). But we have also to consider that the final
entries of several versions (original, 197, and 199) are confirmations of the
final schema nodes (cf. VI.3.7), and the ending of (200) is a well-chunked cliché.
The topic postponement in (197) elicited another noteworthy effect. The tendency
to present topic material in the opening statement of texts (cf. van Dijk 1977a:
150) is violated. In creating their protocols, our subjects should regularly
move the mention of ‘rocket’ into that position. On the no- pause tests, 3
out of 10 subjects preserved the beginning as ‘Empty, it weighed five tons’,
and 5 just said ‘The rocket weighed 5 tons’. On the pause condition, not a
single subject began with the pronoun version; 7 out of 18 opened with ‘the
rocket weighed...’. The rest of the subjects in both groups had different
openings, but all subjects, other than the three using pronouns, opened their
protocols with a mention of the topical ‘rocket’.
We were fascinated to observe that our test subjects varied the style of their
protocols according to the version each group happened to read. Version (198)
demonstrated this phenomenon with special clarity. Half of our subjects toned
the ornamental style down to commonplace expressions, such that the dramatic
opening was recalled as ‘a rocket waits for lift-off’ or ‘a 46ft. rocket
was launched’ (probably drawing on the ‘flight’-schema). The other half
used stylistically marked expressions, even adding to those offered in the text:
‘the giant colossus spewed forth a huge yellow flame’; ‘the burst of
explosive noise is deafening and the explosive fire is blinding as the rocket
zooms away’. Whereas the first half prosaically remembered a ‘radar
transmitter’ following the rocket ‘in the atmosphere’, the second spoke of
‘the eyes of radar’ directed toward a rocket ‘on the verge of human
The condensed expression of version (199) also inspired imitation in protocols.
The subjects expressed their recall in lengthy, complex sentences like this one:
‘the rocket filled with tons of fuel and oxygen took off after two flares were
shown and the scientists had hidden behind mounds of dirt.’One subject made
his entire protocol a single unbroken sentence, splicing phrase and clause
boundaries together with ‘and’ or with commas:
With 8 tons of alcohol and liquid oxygen for the 5 ton rocket, the rocket is
signaled by 2 red lights and the scientists and generals crouch down behind an
earth mound, the rocket takes off with a trail of yellow light, and the radar
clocked the rocket at 3,000 mph as soon as it got out of sight, and a plane
clocked the rocket at 2,400 mph when it was returning back to earth, and it
landed 4 miles from the launching site.
like this remarkable protocol was elicited by any other version. People are
manifestly able to remember style even when they cannot reproduce the exact
expressions they read. When the impression is strong enough, people even
intensify stylistic tendencies. It is difficult to say what kind of “trace”
style may impart—probably not at all like the notion in Gomulicki’s (1956)
“trace abstraction.” There is evidently a storage for selectional controls
on style similar to those applied when people are producing a given text type
for a specific audience (cf. VII.1.4ff.).
Readers did surprisingly well understanding and remembering the poorly written
version (200). It would seem plausible that lower readability would make reading
slower and more difficult; but we did not even observe any slowness in reading
aloud (for a discussion of that issue, see also Coke 1976). We attributed this
lack of effect to the efficiency of schema attachment. For instance, the text
suggests a number of misleading relations: that rockets begin with ‘fuel’
and ‘add on empty weight’ (200.1.2); that the ‘scientists and generals’
were aiready ‘behind earth mounds’ when they ‘withdrew’ (200.2. 1); that
the ‘flares’ somehow ‘fired the rocket’ (200.2.2); that the rocket’s
‘faster’ speed was reached before the ‘slow’ one (200.3.1); and that the
‘radar’ was ‘speeding upward’ rather than the rocket (200.3.3). In their
protocols, subjects showed little confusion, rearranging things into a more
reasonable coherence. The oddly inverted ‘faster and faster after starting
slowly’ was put back in normal order by 4 out of 10 subjects. All the same,
the bad style did interfere with the organization of expression in protocols, as
we can see here:
What goes up must come down. A rocket standing tall yellow and black took
off. Which was part oxygen and part water, the fuel. 3,000 miles. Before it went
out of sight, it looked like a yellow star should look. A big yellow flame. At
the end it came falling back to earth. Scientists and soldiers huddled behind a
barrier. Then crept out.
see confusion about when to mention recalled items, especially in the
displacement of the ‘scientists and soldiers’ to a later time. The high
incidence of sentence fragments (4 out of 10 units), of which there are none in
the original, indicates a disoriented control of text production as a
side-effect of the original’s poor style.
Subjects used various methods for dealing with the misleading version (201).
Some noticed and recalled the discrepant expressions exactly, having apparently
expended increased energy on them, e.g. ‘radar reader’, ‘the giant rocked
on its pad’, and ‘nearby shelters pointed out’. Others recalled instead
more probable expressions than those in the text: ‘launching, padded’
(201.4.2), for example, elicited ‘launching pad’ in 8 of 10 protocols. 3
subjects reacted to the mysterious phrase ‘when everything was red as the
station, two warning flares [ ... ] (201.2. 1) by moving ‘red’ in front of
‘flares’ where it had been in the original. One replaced ‘red’ with
‘ready’. Other solutions included: ‘everything was red and ready at the
control tower’; and ‘As the instrument panel became as red as the rocket
officials’. These results testify to the importance of PROBLEM-SOLVING in
composing a textual world in the face of discrepancies and discontinuities.
Compromises were drawn between what had been presented and what made sense. We
see some ambivalence between these two loyalties, as in ‘red and ready’;
similarly, the odd expression ‘be a scene’ (201.3.3) where ‘be seen’
could fit, caused one subject to write: ‘the rocket could no longer be sceen’.
Badly or misleadingly organized surface structures should cause people to make
errors, because their predictions do not match the text. These discrepancies
between printed text and reading aloud are termed MISCUES by Kenneth S. Goodman
(Goodman & Burke 1973; Allen & Watson [eds.] 1976). Goodman shows that
miscues are normal in all kinds of reading and favor substitutions of visually
similar material, just as speech errors favor acoustically similar replacement (Fromkin
[ed.] 1973). Colorado tests with ‘rocket’ were taped to pursue the source of
miscues. Some errors arose from simple botching of hard words and replacing rare
expressions with more common ones (examples in Beaugrande 1979d). Others,
however, clearly arose from people’s intention to make their protocoIs
cohesive and their text- worlds coherent.
3.14.1 Miscues due to COHESION concern sequential connectivity in areas such as surface fluency, definiteness, and co-reference. Taped versions showed how readers were planning ahead: ‘the flame soon...’(200.3.2) was changed to ’the flame was soon’.,’[ ... ] flares. Amid a [ ... ]’ (1 98.2.1-2) to’ flares aimed at’; ‘ascended with’ (198.2.2) to ‘ascended into’. Definiteness was signaled by the addition of articles: ‘signal for launching’(198.2.1) became 'signal for the launching” with the schema being supportive of expectations; ‘to earth’(199.1.5) became ’to the earth’, where ’earth’ is of course a unique entity (cf. V.3.3.4). Parallelism was introduced, evincing a tendency to re-use already apperceived structures (cf. IV.4.4):’a great roar and burst’(35.3. 1) was altered to ‘a great roar and a burst’, while ’rose as a signal’(197.2.2) was altered to rose as a signal rose’ (with a shift from ‘purpose” to ‘proximity of time’ for the junctive ‘as).
3.14.2 Miscues due to COHERENCE occur when spreading activation of already processed concepts provides material interpolated at other points. After reading about a rocket ’fired’ in the presence of generals’. a reader said ‘war’ for ‘roar’ (35.3.1). In the co-text ‘a pilot watched from an observation plane as the rocket retraced its path, slowing’ (198.3. 1), the major relation is "apperception”; and one subject read ‘showing’ for ‘slowing’. Expectations were influential again when ‘aircraft’ (198.3.2) was read as ‘airforce’. The ‘flight’-schema was probably the culprit when readers said ’landing’ for ’launching’ at the conclusion (199.1.5). A test series with a different sample brought one finding worth mentioning here. One version of a Whitman text contained the co-text ‘from the hills the cannon were thinning them [the soldiers]’ ‘ No fewer than 7 out of 10 subjects read the passage as ’from the hills the canyon' , the canyon being the place one arrives when going ’from hills’, as our Colorado students well know. Another version with the co-text ’cannon from the hills’, with ‘cannon’ being activated before ‘hills’, elicited no miscues at all.
3.14.3. Our subjects obligingly made many of the miscues designed with malice aforethought into version (201). Thus, ’distance, pointed’ was read as ’distant point'; 'be a scene’ as ’be seen’; ‘reader of radar’ as just ‘radar'; ‘launching, padded’ as ’launching pad’, and ‘giant rocked’ as ’giant rocket’. One young lady struggled valiantly but vainly with ‘giant rocked’, saying instead ’giant rocket’ three times in succession in a state of rising duress. She must have experienced a conflict of control between concept activation and local impulses of articulation.
3.15 These data suggest that surface sequencing has an important influence on text processing, even though quantitative recall was not severely affected. I now review the data concerning the interaction between a text world and the processing strategies based on prior knowledge of text receivers. I deal mostly with the data for the original (35) and the inversion (197) containing roughly the same expressions. The processes to be explored are:
3.15.1 CONCEPT RECOVERY, during which expressions make conceptual content active in working memory, can be documented when readers recall other expressions than the ones they actually encountered (cf. iii.3.5).
3.15.2 SUPERCLASS INCLUSION is shown when readers recall more general class names than were used in the text (cf. 111.3.22f.; Ausubel 1963).
INFERENCING is observable when readers undertake to bridge apparent
DISCONTINUITIES (missing linkage) or GAPS (empty nodes) in the textual world
which they set up in their minds (cf. 1.6.9; III.4.29f).
3.15.4 SPREADING ACTIVATION is manifested when readers report additional material which they associated with the text-world materials in their minds (cf. 1.6.4; III.3.24).
RECOVERY is evident in all protocols. I cited in VI.3.9ff. some examples of how
the concepts in the ‘flight"-schema were expressed quite differently in
protocols than in the original. For example, our subjects did not usually say
that the rocket ‘rose’; instead, they used other expressions such as the
following: ‘went up’, ‘shot up’, ‘was launched’, ‘ascended’,
‘lifted off,' ’took off,’ 'climbed’, ’moved’, ’soared skyward’,
'took to the sky’, ‘fired’, ‘set off, ‘was released’, ‘blasted
off, and ’started out’. Some of these expressions include an
“initiation” component, while others do not -- perhaps because of the
structure of the ‘flight-schema (V1. 3.9). I could have paraphrased the text
to deploy the more popular expressions, e.g. ‘take off’ (used by 29 out of
72 subjects). Yet we would be on shaky grounds in claiming that we were getting
verbatim recall (cf. VII.3.3) We might simply be getting back a more probable
SUPERCLASS INCLUSION was at work where our subjects recalled ‘plane’ as
‘aircraft’ (cf. VI.3.4), and ‘radar’ as ‘machine’. ‘Scientists and
generals’ were subsumed under ‘men’ and ‘the people that control the
rocket’--in the latter case, the overly general ’people’ is narrowed down
again with the ascribed agency. The selection of superclasses influenced the
recoverability of the subclasses and instances. A subject who mentioned
‘something composed of two chemicals’ went on to guess ‘hydrogen’ and
‘oxygen’, because ‘alcohol’ is (for all too obvious reasons) less
typical in schoolroom chemistry. A subject who mention ‘propellants’, on the
other hand, remembered ‘alcohol’ but substituted ‘gasoline’ for
‘oxygen’. Four subjects apparently leaning toward chemicals recalled as a
fuel ‘nitrogen’, which does not even burn.
3.18 INFERENCING as found in protocols confirms the bridges I postulated in III.4.29. The inference that ’scientists and generals’ were present in order to observe the rocket was made so often that 24 subjects recalled it as part of the original. 17 subjects made the relation between ‘ready’ and the ‘take-off’a part of their protocols. Inferencing filled not only these gaps in the presented version, but also gaps due to decay in storage. Having forgotten the ,scientists’, a subject mentioned ‘generals and soldiers’, the latter being supplied as people for the generals to order around. Another subject who forgot the landing concluded that the ’rocket', being a subclass of ’spacecraft’, went ‘into orbit’.
Our subjects were evidently aware of GAPS, which they did not attempt to fill.
Instead, they created placeholders: ‘somewhere in New Mexico’; ‘alcohol
and something else’; ‘something composed of two chemicals’; ‘generals
and others’; and ‘scientists and something else were behind earth dunes’.
There are two plausible accounts for this phenomenon. Either there is some
psychological reality to the notion of “model space," such that people
walking through mental storage could notice indistinct or empty areas. Or else
people retain some traces of input, but not enough to recover entire elements.
Although the two processes doubtless interact, it is useful to distinguish
between SPREADING ACTIVATION and INFERENCING. Spreading activation is based on
ASSOCIATION and results naturally from concept -activation in ideation or
comprehension without specially directed impulses. Inferencing is based on
express PROBLEM-SOLVING and is directed to overcoming discontinuities and gaps.
Spreading activation runs via the organization of prior knowledge in episodic
and conceptual memory (cf. 111.3.16); inferencing runs via the particular
organization of the textual world at band. A plausible tactice would be to
represent spreading activation in the WORLD-KNOWLEDGE CORRELATE, and inferencing
in the TEXT-WORLD MODEL. These differences suggest that we might make a
consistent distinction in the processes of knowledge management and recovery.
UPDATING (1.6.4) and INHERITANCE (III.3.19) during comprehension, and
RECONSTRUCTION and REPRODUCTION during recall (VII.3.1), could each have an
associative function running on spreading activation and a problem-solving
function running on inferencing. So far, it has been difficult to get an
empirical handle on questions such as: (1) how much material is added right away
during comprehension, and how much is added later for recall; and (2) what .
language users consider as a discontinuity or gap. I shall review the changes for
test subjects made according to LINK TYPES, rather than according to the
LOCATION was by far the most popular addition, probably because it is so crucial
to the understanding of ‘flight’. The opening scene was described with new
details. The ‘desert’ became ‘desert plains’, where launching
‘took place under a bright sun’. Reasoning that rockets are shot off
far from population centers, subjects recalled events ‘in the middle of a desert’
where the rocket ‘Iay alone’, or even ‘far out on a Moroccan
desert’. The rocket’s attribute of ‘great’ led people to remember
how the ‘rocket towered over the many scientists and technicians below’.
The expected flight was no doubt responsible for mentioning the rocket ‘on
a launch pad pointing toward the sky’. When time came for take-off, one
subject wrote: ‘the rocket blasted off, up. and away from the launch
pad’. Later events were described like this: 'at
its peak it
reversed and plummeted downward on its journey back to earth’.
Locational proximity is, of course, useful for continuity. Subjects recalIed the
opening scene with the scientists and generals ‘gathered all around the
rocket’; and ‘Iand mounds were surrounding the lift-off. Proximity
served to compensate for changes. The readers who converted the 'mounds’ into
a 'mountain’ had to fit the larger item in somewhere into the background: ‘a
rocket is in from of a mountain in Arizona’; or ‘a rocket is in
front of a mountain where the people that control it are' (with a new
function supplied). The pilot was introduced in a “nearby plane’ or ‘aboard
an airplane’. One subject who ended his protocol with the rocket ‘hovering
over’ (inspired by Star Trek?) used proximity to compensate for
having forgotten the landing.
TIME was the next most frequent type of addition, in accordance with the
importance of time in a ‘flight’-schema. Because time is a steady flow
during ‘flight’, the addition of proximities is natural. For instance, the
content of the second paragraph was drawn together in that fashion. The events
took place ‘when the time came for the rocket to be launched’; and
‘scientists crouched behind mounds as the rocket was launched’.
Later, ‘while it was ascending, radar tracked’ the rocket. The
missile Was seen from a ‘plane flying at the same time the rocket
was’. On the other hand, the time expressions in the original (‘soon’,
‘in a few seconds’, ‘a few minutes after) were among the expressions
reproduced least frequently by our subjects. People seem to create their own
time cues as needed for organizing textual worlds.
CAUSE and ENABLEMENT were contributed occasionally. Some readers sensed a
discrepancy between the rocket’s great distance and its continued visibility,
and they inferred: ‘being very big in size, the rocket could be seen by the
naked eye for quite a distance ‘; ‘the bright yellow flame could be seen
from very far, a pilot in a plane was even able to see it’. Other readers
inferred the final fate of the rocket: ‘a huge explosion followed the
rocket’s impact’; ‘going approximately 2,400 mph, it must have made
quite a recess in the earth's crust’.
3.25 The frequency with which APPERCEPTION was contributed may be due to the emphasis on that relation in the original, or else to readers’ reliance on mental imagery (cf. Paivio 1971). The latter source indicates the need to explore compatible modes for language and vision (see III.3.18). One subject introduced a new instrument for observing the rocket and still felt that visibility must fail eventually: ‘we can see it by satellite, but it speeds up and we lose track of it’. Another subject was especially insistent about continuity of apperception:
(204) They watched the take off and paid attention to the flames that followed the rocket until they could not be seen anymore and then they looked into a radar detector to find the distance of the rocket. A pilot in a plane watched the rocket, he saw it go up and return down to earth.
AGENCY is not too important for an unmanned ‘flight’-schema, because
propulsion and gravity do most of the work. However, some readers thought that
the ‘scientists and generals’ at the site ought to be doing more than just
‘Iurking behind mounds’ (as one subject put it). Four readers recalled them
as agents of setting off the flares, and three others had them directly fire the
rocket. The preferential treatment was via apperception: 24 subjects had their
scientists observing the rocket (cf. III.4.29)-a typical activity for scientists
ATTRIBUTES were flled in when motivation arose. To explain visibility, the
rocket’s ‘trail’ was made ‘huge’ and the ‘star’it resembled
‘bright’. On the other hand, four readers supposed that, because of the
distance, the star should be ‘tiny’. Three made the pilot’s plane ‘small’,
perhaps to contrast better with the ‘great’ rocket.
We obtained a scattering of further link types. MOTION was transferred to the
‘star’ which the rocket ‘Iooked like’: ‘a star rísíng in the
sky’ and a ‘shootíng star’. PURPOSE was assigned to the whole
text-world: ‘an experíment had taken place’. SUBSTANCES were
envisioned for the rocket’s trail which figured as ‘a tail of burnt
gases’, ‘a stream of exhaust’, and a yellow stream of fire
which turned into smoke’ (note the similarity of these inferences to
some content in the world-knowledge correlate in 111.4.36). PARTS were added
when a subject remembered ‘a flame from the rear of the last stage of
the rocket’. COMMUNICATION was assigned to the ’pilot’ who ‘reported’
the rocket’s return (four protocols).
This evidence confirms the postulate that meaning can be viewed as process
(cf. III.3). However diverse individual concepts and expressions may
in themselves, there appears to be a limited set of strategies for putting them
together in texts and textual worlds. These strategies apply both to the
immediate needs of cohesion and coherence in reception (e.g. reading) and
production (e.g. protocol-writing) of texts, and to the organization of
knowledge in the mind. Decay can thus be offset, though the results will often
lead to modification of the original material; decay appears to proceed even in
such a short time space as the five-minute pause we used in our testing. I now
propose six theses about the interaction of text-presented knowledge with prior
stored knowledge. Although I arrived at these theses independently, they bear
striking resemblances to the notions of David Ausubel (1960, 1963; Ausubel &
Fitzgerald 1962) working in the tradition of Bartlett (1932).
Text-presented elements are prívíleged ín storage and recall if they match
stored world-knowledge patterns. Following
the widely publicized space program, it is not surprising that our readers made
sense out of the rocket text, even on the badly expressed versions. News
coverage favored the scientific over the military aspect for political motives;
and our readers remembered the scientists in 42 out of 72 cases, but the generals
in only 14. 10 of those 14 were on the no-pause condition, suggesting that the
‘military’frame withstood decay less well than the ‘science’-frame.
Because rockets are known to be powered by combustion, the recall of ‘flame’
by 48 subjects was not surprising. 32 remembered the presence of 'fuel’. The
‘yellow’ color of the
Text-presented elements are privileged if they are attachable 10 major nodes
and links of a global stored knowledge pattem, such as a frame or schema. I
reviewed the evidence for the ‘flight’-schema in V1.3, and cited the fact
that schema-related elements were recalled with the highest absolute frequency.
I looked at the influence of frames on the understanding and remembering of the
‘sunspot’ text in VI.2, where ‘magnetic field’ was most often taken as
the subsuming pattern. I explored the support that knowledge about plans
provides for deciding what to say and for understanding actions and utterances
of a stage play in VI.4. I shall further investigate the use of schemas for
stories in VIII.2.
Text-presented elements are altered to produce a better match with
first glance, the ‘rocket’ text contains little disturbing material. Yet the
“chunking” together of knowledge might lead to improving linkages still
further. One subject converted the colors ‘black and yellow’ to ‘silver’,
the color of metal. The location ‘desert’ impelled people to transform the
‘earth mounds’ into ‘sand dunes’ and ‘rock formations’. Evidently,
reflecting on the “purpose” of these mounds to shelter scientists, one
reader made them into ‘concrete bunkers’. The comparison of the
‘rocket’ to a ‘star’ was also smoothed over by making the star into a ‘glow’,
a ‘blur’, or a ‘dot’. Particularly striking was the
compensation of a reader who may have seen V-2 rockets in World War II stories;
he recalled the ‘launching of a captured German V-2 rocket’.
Alterations were also required when readers had rearranged other portions of the
textual world. A reader who had recalled the rocket’s trail as ‘red’ by
conflation with the ‘flares’ had to say later: ‘red streaks in back
of the rocket at the beginning turned yellow’. After stepping up the
length of the rocket from 46 to 1,000 feet, a subject depicted a ‘massive’
rocket that ‘erupted’ like a volcano with its trail becoming a ‘brilliant
display)’ of fire’. By inferring that the rocket passed out of sight
because of speed, not distance, another subject had to match things up again:
‘it is only going 3,000 mph, but it speeds up and we lose track
of it and it comes back at 20,000’. These text-internal rematchings are
indicative of the cybernetic regulation outlined in 184.108.40.206.
Text-presented elements become conflated or confused because they are closely
related in world-knowledge. Readers
occasionally expressed concepts that combined the content of those in the
original. ‘Withdraw’ plus ‘crouch’ added up to ‘hide’; ‘roar’
and ‘burst’ to ‘explosion’; and ‘return’ and ‘plunge’ to ‘descend
back’, ‘tum back down’, and ‘come crashing back down’. The
‘flares’ were conflated with the rocket’s ‘flame’: ‘the rocket went
off in a burst of flares’; ‘when the rocket was launched it looked like a
red flare’; ‘lhe rocket was followed by a long fiery flare’. Two
subjects moved the ‘red’ color from
the flares to the rocket’s (yellow) trail, and one compromised upon’ ‘orange’,
The star suffered a like fate to the flares when a subject declared: ‘as
the rocket leaves it trails bright yellow and red flames like a burst of stars’.
Text-presented elements decay and become unrecoverable if they are neutral or
accidental in world-knowledge. The
prime illustration here was the treatment our subjects gave to quantities.12
Possible implications of the treatment of numerical vs. measurement expressions
were noted in 220.127.116.11.]
They reversed the respective weights of frame versus fuel, or calculated 5 and 8
tons as 5,000 and 8,000 pounds. The rocket’s length telescoped between 26 and
1,000 feet, and its speeds ranged from 300 feet per minute (a mere 3.13 mph) to
300,000 mph. The distance between take-off and landing accordioned between 60
feet and.164 miles. Speeds were converted to altitudes, e.g. ‘2,400 feet in
the air’. A quantity with no use was supplied: ‘there was something about
‘2,400’. Many subjects hedged their recall with ‘about’ or ‘approximately’,
or claimed that radar merely ‘estimated’ the speed. Others
contented themselves with ‘very fast’, ‘a certain speed’,
and ‘lots of fuel’. The rocket’s colors went by the same route.
About one half of the subjects recalled no colors. One fourth recalled both
‘black and yellow’, and the rest vacillated between these two and ‘red’
‘green’, ‘white’, ‘silver’, and ‘blue’.
Text-presented entries become indistinguishable from inferences and spreading
activations on the text receiver’s part. I
have already surveyed considerable evidence here in VII.3, and further evidence
is given in VI.2, VI.3, and VIII.2 (cf. also Beaugrande 1979d; Beaugrande&
Miller 1979), The longer
the times are which elapse between text reception and the producing of a
protocol, the harder it should become to uphold such a distinction, unless the
receiver assigns special importance to it. Harry Kay (1955) noted a striking
effect in this regard, His subjects not only made important changes upon the
material they read; the renewed presentation of the text was not employed to
return toward the original. Instead, the subjects kept their own versions time
after time, consistently preferring the latter to the version in the text
These six postulates are not without some correlates in conventional psychology,
even though the work of Bartlett, Ausubel, and their followers was often
ignored. The privileged status of entries which match stored patterns agrees
with the finding that prior frameworks aid retention on associative learning
tasks (cf. Jenkins & Russell 1952; Bousfield 1953; G. Mandler &
Pearlstone 1966; Bugelski, Kidd, & Segman 1968). Alteration for the sake of
a better match has been observed in tests where people changed anomalous
sentences to make better sense (Herriot 1969; Fillenbaum 1971, 1974;
Strohner & Nelson 1974). Conflation of associated elements agrees with the
interference of similar “stimuli” (Gibson 1942; Anisfeld & Knapp 1968;
3.31 One promising direction I am pursuing is to observe how cognitive processing creates the individual text-world models of particular readers through the interactions of text knowledge and world knowledge. I design these models and then compare and contrast them with the model for the original text. Eventually, this proceeding may provide a dynamic notion of “text meaning” as the core operations of content processing performed by a representative group of text receivers. That case could highlight the vital importance of INTERTEXTUALITY (cf. 18.104.22.168) for defining textuality. After designing a number of protocol models, I am confident that the strategies used by text receivers might be systemized along the lines I have undertaken to sketch out in this section. A graphic representation for the coherence of whole texts can show regularities of STORING and FORGETTING, which might not be visible via other means. I shall briefly portray this approach as applied to our ’rocket’ sample.
3.32 The following protocol was written by a subject who read the original version on the no-pause condition:
(205) In a New Mexico desert, a V -2 rocket waited to be launched: it was 60 feet tall and weighed 5 tons empty. The generaIs and technicians stepped back behind dirt mounds and launched two red flares signaling the launch of the rocket. The rocket sped upward at increasing velocity, leaving a 60 ft. exhaust flame behind it. It reached a velocity of 3,000 mph on the way up and later an airplane pilot clocked it at 2,400 mph on the way down.
This subject divided his attention between the preparations for launching and the actual flight. His first paragraph subsumes what had remained of the first two original paragraphs; and his second the surviving content of the original third and fourth. The suppression of paragraph transitions has noticeable effects: ‘the generaIs’ suggests that their identity should be clear from the situation; and the junctive ‘and’ before ‘later an airplane pilot’ downplays any discontinuity between the ‘way up’ and the ‘way down ‘. I show this compacted text-world model for (205) in Figure 34. I adopt the following conventions: as far as possible, reproduced materials appear in positions corresponding as far as possible, reproduced materials appear in positions corresponding to those in the original text world model; identical expressions appear as they are; alternative expressions for the same concepts are marked with ‘*’, and placeholders or hedges with ‘§’; and concepts added by inferencing and/or spreading activation are in square brackets (cf. III.4.29).
3.33 As a result of compacting, the representation of schema elements is simplified over the original. The ‘take-off event is represented by ‘launch’ and ‘sped upward’, the former being in the old position of ‘fire’ and the latter moved up from a later mention in regard to speed (cf. 35.3.5). This displacement creates a need for a signal of the ‘ascend’ event which is provided with ‘on the way up’ to accompany the specification of ’velocity’. That samr expression was simply varied to take care of ‘descend’: ‘on the way down preserving the balance of the ‘flight’-schema (VI.3.3) in miniature. The subject’s focus on ‘velocity’ throughout the second model space drew attention away from the ‘Iand’ event that is hardly definable in terms of speed The subject doubtless knew that the rocket landed, but focus was directed; elsewhere.
The modifications and additions performed by this test subject agreewith my
remarks so far, e.g. that the ‘rocket waited to be launched’ (having
the opening state look forward to the first schema event, cf. VI.3.13). ‘Exhaust'
was supplied as a “substance-or’ of the flame (cf. VII.3.28). To offset
decay and change, PROBLEM-SOLVING techniques were evidently used to create new
linkage among surviving material. After ‘crouch’ was lost, ‘stepped
back' (synonym of ’withdraw’) was linked directly to ‘dirt
mounds’ (rather than ‘some distance’). The agents ‘technicians and
generals’ were linked to a new action of ‘launching’ the
‘flares’, so that their presence is motivated (cf. VII.3.26). The general
tendency was to lose whole spaces of the text-world: model rather than random
isolated nodes. For instance, the knowledge of fuels, the noises at the
take-off, the use of radar, the resemblance of the rocket to a star, the
landing-all of these spaces disappeared. It seems reasonable suppose that if a
CONTROL CENTER is removed, whatever is hanging on it will be hard to
recover or integrate. Of course, this tendency reflects not a rule, but at most
3.35 We can contrast (205) with a rendering that shows somewhat different results, also from the no-pause condition:
A big black and yellow rocket, 46 feet long and 200 tons, was in the Arizona
Everyone was waiting for this missile to be
fired into the sky. Everything was ready.
All of a sudden it fired into the sky. As the
general and others
watched, they saw the rocket and a long trail
of flame following it into the sky, until they could
not see it any more.
Radar picked it up and estimated its speed at
A plane down below saw the rocket coming back down to earth in a ball of flame at approximately 2,400 mph, and then crash right into the earth.
subject arranged the material into shorter paragraphs than those in (205).
Evidently noticing a loss of material, she filled out her textual world by
copying repeatedly off the same concepts and concept configurations. We have
here a special use of CONTROL CENTERS to produce REDUNDANCY rather than
I draw these redundancies as links of equivalence or recurrence in Figure 35.]
Everything was ready’ was expressed redundantly as ‘everyone was
waiting’ -- notice again the inference linking up to the ‘firing’. The
location ‘into the sky’ was mapped three times onto the surface text.
The observation by the ‘generals’, also mentioned three times, incorporated
the inference that they were the ones who lost sight of the rocket at a certain
height (cf. IIIA.33). The ‘tracked’ of the original radar was subdivided
into ‘picked up’ (with initiation) and ‘estimated’ (with
3.36 Decay made a number of inferences necessary to keep things connected. Due to the loss of knowledge about preparations for take-off (flares, shelter), the subject reasoned that the rocket must have taken off all of a a sudden’. Having forgotten the ‘scientists’ (though ‘others’ suggest awareness of the gap) and retained one ‘general’, the subject chose the superclass ‘missile’ as suitable for military purposes. The subject transferred her own uncertainty about speeds into the text-world via ‘approximately’ and ‘estimated’ (cf. VII.3.29.5). The ‘plane’ was inferred to be ‘down below’ the rocket, and the landing was envisioned as a ‘crash’. The ‘trail of flame’ encouraged the wrong inference that the rocket would ‘come back down in a ball of fire’ (more likely, the fuel would have been exhausted). The model for this textual world is presented in Figure 35.
The representation of SCHEMA elements also indicates the interaction of text
knowledge and world knowledge. The concept ‘fire’ was transferred from the
original context of the flare signal to the representation of the ‘take-ofr
event. The ‘ascend’ event appears to have been satisfactorily covered by the
three uses of 'into the sky’. The ’descend’ event was expressed as ‘come
back down’, which captures both the ‘return’ and the ‘plunge’ in
the original. Having stepped up the rocket’s weight from 5 to 200 tons, the
subject readily concluded with a ‘crash’ landing of added impact.
3.38 In a recall protocol with the fiveminute pause, the state of decay was more advanced than in the samples we have seen so far, but coherence was nonetheless maintained with similar strategies. The text reads as follows:
(207) Far out on a Moroccan desert, a V-2 rocket was prepared for its blast-off. It was 26 feet 10ng and was silver in color. Once the count-down was completed, the rocket gracefully soared upward, leaving an 80 ft. flame behind. Soon the rocket was out of sight, and the flame was only seen as a tiny twinkle of light. The space increased to 24,000 mph as it rose into the atmosphere. Then it was seen dropping to the ground 40 feet from its original place of blast-off.
The network diagram for this text-world is shown in Figure 36.
The schema events are all represented. Having lost the knowledge spaces
involving the personnel and the flares -- the rocket being ‘prepared for
its blast-off' may have made them superfluous---:the subject filled in his
prior knowledge that a ‘count-down’ would be ‘completed’. The
‘take-off’ and ‘ascend’ concepts are subsumed in ‘soared’, analogous
to the ‘rose’ of the original. Indeed, the expression ‘rose’ was itself
still available, and was deployed to introduce a further stage of flight ‘into
the atmosphere’, so that the displaced original verb ‘sped’ became the
noun ‘speed’. The inferred ‘increase’ of the speed may
have been due to the intensity indicated by ’soared' or to the further
inference that, on entering the ‘atmosphere’, the rocket would have
to confront the gravitational pulI of the earth. The ‘descend’ and ‘landi”
events are expressed via ‘dropping to the ground’. The final ’place of
blast-off' folIows the original, and harks back to the expression introduced
by inferencing into the opening sentence.
The treatment of non-schema-based entries relied on inferencing. A metal rocket
could be ‘silver’, and a ‘desert’ is known to exist in ‘Morocco’
doubtless deemed a remote, unpopulated area (cf. VII.3.21). The attribute ‘gracefully’
is suggested by ‘soared’,
and the ‘twinkle’ of light (‘twinkle’ probably being
spreading activation from ‘star’) should be ‘tiny’ at a great
distance. The specification of ’atmosphere’ as the highest location
of the flight could be attributable to prior knowledge about space exploration.
By viewing these three protocols as textual systems, rather than as only
sequences of sentences or propositions, we can notice some aspects of quasi-cybernetic
controls on knowledge utilization. The decay of system components leads to
regulative compacting and inferencing. If the compacted or inferred elements are
not fulIy compatible to their environments. compensations are made along the
lines of problem solving: Finding pathways to connect points in spaces.
Schema-related materials survive with the greatest endurance. Decay works
against accidental elements, and tends to suppress whole spaces rather than
leaving space fragments floating unattached.
Hierarchical trees for text worlds (e.g. Bower 1976; Mandler & Johnson 1977;
Meyer 1977) suggest that only unimportant details ought to be lost. But we have
seen here that important elements are also subject to various kinds of
modification. We can illustrate the difference between human processing and
ideal hierarchical processing by contrasting (205) through (207) with the
summaries generated by a computer from the same text. Robert F. Simmons designed
a LISP representation of the content in Horn clauses of successor arithmetic
using a generalized production system interpreter (cf. Simmons & Correira
1978). The program and its summary protocols are given in the Appendix at the
end of the book.
What I have been able to offer here is no more than a modest beginning toward a
psychological theory of text processing, far from offering precise, let alone
statistical, predictions capable of verification or falsification. Instead, I am
searching for plausible accounts after the fact. Perhaps this approach has the
advantage of not being committed a priori to ignoring certain issues that a
ready-made theory might find unaccountable. In closing, I must stress the
central issues must be pursued through interdisciplinary co-operation among all
sciences searching for a limited, powerful, and unified set of processing
strategies to account for the constantly observable fact that people do use and
retain textual meaning ands purpose in alI modes of communicative activities.
Click here to go to Ch VIII
Click here to go to Table of Contents