Click here to go to Chapter VI

 

Click here to go to the Table of Contents

 

VI

Frames, Schemas, and Plans

1. GLOBAL PERSPECTIVES ON KNOWLEDGE

1.1 Efficient processing of texts cannot afford to operate on a LOCAL scale. Procedures which could handle only single items or small groups of items would lack the requisite directionality and control that keep tabs on predictions and probabilities in such a diverse intersystem as communication. I noted in II.2.9, for example, that syntactic processing should treat single items as MICRO-STATES inside MACRO-STATES, so that the building of structures would have orderly priorities. In III.4.27, I argued that the SPACES in text-world models can be viewed as conceptual macro-states. However, these structures originate largely from within the text, i.e., they are BOTTOM-UP macro-states; there must also be TOP- DOWN macro-states coming from outside the text and supplying GLOBAL hypotheses about what is going on in the textual world (on top-down versus bottom-up, cf. I.6.5). In this chapter, I shall explore recent work concerning these large-scale stored organizers for knowledge.

1.2 Whether in memory storage or in actual utilization, configurations of knowledge can have at least four perspectives (III.4.II.7). First, knowledge can be viewed as an ARRAY in which elements are arranged such that access of potentially relevant elements is provided. This perspective is called a FRAME (cf. Minsky 1975; Charniak 1975c; Winograd 1975;¹ [1. Some researchers seem to treat “frame” and “schema” as synonymous, but the distinction is both crucial in theory and borne out by the major sources I cite. Of course, still other ways could be found to contrast them (cf. Tannen’s blinkered survey, 1979); Scragg 1976; Petöfi 1976; Metzing [ed.] 1979). For example, a ‘house’-frame would be a network of entries such as parts, substances, uses, etc. that houses have (cf. III.3.25). The format is one of links fanning out from a conceptual control center (III.3.8), with no single commitment to a sequence of actualization. Second, knowledge can be viewed as a PROGRESSION in which elements occur during actualization. This perspective is called a SCHEMA (cf. Bartlett 1932; Rumelhart 1975, 1977a, 1977b; Rumelhart & Ortony 1977; Spiro 1977; Kintsch 1977a; Mandler & Johnson 1977; Thorndyke 1977; Kintsch & van Dijk 1978a, 1978b; Adams & Collins 1979; Freedle & Hale 1979). For example, a ‘house’-schema could describe the order in which houses are assembled, or the sequences in which people can walk through them. The schema is thus much more committed to an ordered sequence of actualization than is the frame.

1.3 Third, knowledge can be viewed as relevant to a person’s PLAN in which elements advance the planner toward a GOAL (cf. Sussman 1973; Abelson 1975; Sacerdoti 1977; Schank & Abelson 1977; Allen & Perrault 1978; Carbonell Jr. 1978a; Coben 1978; McCalia 1978a; Wilensky 1978; Beaugrande 1979a, 1979b). For example, someone who wants a house or is told that somebody else wants one would summon up plans for building or buying a house. The ‘house-getting’-plan will look very different depending on the method selected. A plan to buy a house will also differ from a plan to burgle one—a factor affecting priorities of processing (R. Anderson & Pichert 1978). Fourth, knowledge can be viewed as a SCRIPT in which elements are INSTRUCTIONS to PARTICIPANTS about what they should say or do in their respective ROLES (Schank & Abelson 1977; Cullingford 1978; McCalla 1978a, 1978b). For example, a ‘restaurant’-script has instructions for the customer, the waiter, and the cashier (or the policeman if you don’t pay), to be enacted in an established pattern.

1.4 These four perspectives yield a gradation from general access toward operational directionality and order. Frames and schemas are more oriented toward the internal arrangement of knowledge, while plans and scripts reflect human needs to get things done in everyday interaction. One might argue that schemas are frames put in serial order, that plans are goal-directed schemas, and scripts socially stabilized plans (on the latter cf. Schank & Abelson 1977: 72f.). In this progression, the pattern becomes more selective, and expectations more definite at any given time of application; consequently, the episodic aspect dominates over the purely conceptual-relational one more and more. However, considerations of economy (III.3.18) lead me to suppose that much knowledge is shared by these perspectives. For example, the ‘house’-frame could be selectively activated to produce a ‘house-building’-schema by following “part-of” and “substance-of” links and then putting the results in an order dictated by knowledge about materials and construction. The frame would be useful for a descriptive text about existing houses, and the schema helps in telling or understanding a story about a particular house getting built. If people were then called upon to actually build a house themselves, they could convert the schema into a plan via further knowledge about how to buy or obtain materials, how to select a site, and how to procure the co-operation of other people. A professional contractor doubtless has a complete, detailed, and routinely applied ‘house-building’-script that other people do not possess.

1.5 These large-scale knowledge configurations supply top-down input for a wide range of communicative and interactive tasks. Their utilization is a form of PROCEDURAL ATTACHMENT, where operations are adapted and specified to fit a current requirement (Bobrow & Winograd 1977; cf. II.2.19; III.4.1). This attachment requires more processing as the task at hand becomes more detailed and precise. For instance, a schema or plan for ‘house-building’ undergoes more development for a large, luxurious edifice than for a small, modest one. Still, the availability of global patterns of knowledge cuts down on non-determinacy enough to offset idiosyncratic bottom-up input that might otherwise be confusing (cf. IV.2.9).

1.6 Depending on context and co-text, the selection of the appropriate global pattern may be difficult (cf. Wilks 1975a: 47, 1977b: 389; Collins, Brown, & Larkin 1977; Schank & Abelson 1977: 58; Charniak 1978; Rumelhart 1978; Woods 1978b: 9ff.). The consensus is that an understander must watch for cues and their “intersections” (Charniak) or “coincidences” (Woods). Obviously, the understander cannot afford to wait and gather large numbers of cues, or the pattern won’t be selected in time to be very useful. Hypotheses should be formed early and applied until a major snag is encountered (cf. Kuipers 1975). This procedure is not without drawbacks: the early hypothesis might be wrong and bias understanding such that contradictory cues would be overlooked for too long a time (cf. Bruner & Potter 1964). The understander may be able to adjust by treating the first hypothesis, now rejected, as a “near miss” (Winston 1975) that may remain useful for reasoning by ANALOGY (cf. III.3.21).

1.7 The occurrence of DETERMINATE cues is plainly more reliable than that of TYPICAL ones, and typical in turn more reliable than ACCIDENTAL. If a text starts right out with (Charniak 1978: 187):

       (172) The woman waved as the man on stage sawed her in half.

a ‘magician’-frame or ‘magic-trick’-schema can be confidently applied, even without having these domains named in the surface text. However, if the text beginning were less determinate:

       (173) John walked thoughtfully down the aisle.

too many candidate frames (supermarket, church, airplane) or schemas (shopping, getting married, getting on a plane) could fit. The understander would have to wait for a continuation, such as:

       (174a) He swiped a can of caviar from the display shelf.

       (174c) He swiped a bottle of tequila from the stewardess’s cart.

1.8 Bransford and M. Johnson (1973) deliberately constructed non- determinate texts for people to read. They found that the following sample was treated as nearly incomprehensible and was poorly recalled (1973: 392f.):

(175) If the balloons popped the sound wouldn’t be able to carry since everything would be too far away from the correct floor. A closed window would also prevent the sound from carrying, since most buildings tend to be well insulated. Since the whole operation depends on a steady flow of electricity, a break in the middle of the wire would also cause problems. Of course, the fellow could shout, but the human voice is not strong enough to carry that far. An additional problem is that a string could break on the instrument. Then there could be no accompaniment to the message. It is clear that the best situation would involve less distance. Then there would be fewer potential problems. With face to face contact, the least number of things could go wrong.

They supplied some subjects with a picture showing a young man presenting a serenade with guitar accompaniment to his girlfriend; she was positioned at the window of her sixth-floor apartment, obliging the man to transmit the song via a microphone whose loudspeaker was supported outside her window by six lighter-than-air balloons.² [2. The picture provides the background knowledge which the definite articles presuppose; even the articles become exophoric (cf. V.5).] The subjects with the picture understood the text right away and recalled more than twice as much as the others. This text describes a rare and improbable situation, but Bransford and Johnson (1973: 400) also prepared an unresolvably non-determinate text about the everyday activity of washing clothes, and the results were the same.

1.10 Non-determinacy can be introduced from the side of the receiver as well. One can take many texts and apply to them knowledge patterns that their producers may not have even considered. For example, managers of a recent seminar on “Marketing Warfare” proclaimed that ‘what works best in warfare also works best in marketing.’ They distributed posters with quotations from Carl von Clausewitz’s 1832 book On War, each quote being translated into the frames, schemas, and plans of American business, e.g.:

(176.1) Napoleon’s objective was not to merely outmaneuver but to annihilate the opposing force.

(177.1) Moral effects are greater on the side of the conquered rather than the conqueror.

(177.2) Translation: God is on the side of General Motors.

In support of their metaphoric outlook, the managers circulated a statement of a female business executive:

(178) In presenting my ideas to an all-male board, I’ve found I’m understood better when I use the military or football terminology of offensive-defensive.

1.11 Frames, schemas, plans, and scripts should be capable of INHERITANCE (cf. III.3.19). Inheritance would apply for relations of classes to superclasses and metaclasses. A ‘sun-frame could inherit from a ‘star’-frame, a ‘folktale’-schema from a ‘story’-schema, a ‘bank-robbing’-plan from a ‘steal’-plan, and a ‘pizza-parlor-script from a ‘restaurant’-seript. The storage address of knowledge would depend upon what patterns were more likely and efficient in use (see especially Fahlman 1977). A statement of CANCELLATIONS (cf. III.3.19; VI.3.4) could apply, e.g. changing the ‘visible at night’ of the ‘star-frame to ‘visible during the day’ for the ‘sun’-frame. Plausibly, text processing often requires several frames, schemas, etc. to interact with each other, leading to further modifications in context (cf. D. Bobrow & Norman 1975; Adams & Collins 1979).

1.12 The standard of INFORMATIVITY requires that understanding cannot involve perfect matching of input to a frame or schema; instead, there must be at least minor variables or discrepancies to uphold interestingness. Thus, an understander can’t afford to discard a frame or schema at the slightest difficulty. Other recourses could be: (1) check to see if the mismatching element is connected by determinate, typical, or accidental linkage to its environment in the frame or schema; (2) if accidental, continue as before; (3) if typical or determinate, check to see if the text is fictional. We saw in IV.4 that a newspaper writer constructed an article so that a ‘psychoanalyst-patient’-frame worked for the opening stretch of text, but had to be discarded in favor of an ‘anthropologist-chimpanzee’-frame later on. The discarding does not vitiate the usefulness of the original frame for understanding the context where it did apply.

1.13 When the text fails to match receivers’ frames, an operation of FRAME DEFENSE may ensue: the text is rejected or simply not understood in order to preserve the validity of the frame (Beaugrande 1978b: 9f.). A striking demonstration was provided by an editorial board’s comments on a paper I submitted to a prominent journal in educational research. One reviewer was a professor of English Education and another of Linguistics. The paper was rather critical of conventional linguistics and proposed to deal with the issues from a quite disparate direction. While the English professor remarked that ‘the untempered rejection of sentence grammars (context-free) as avenues of understanding reading is important (and, I believe, correct) and in need of restatement’, the linguist decried the paper as ‘polemical about linguistics, unnecessarily.’ The intriguing point was that due to frames, the two reviewers reached totally opposite judgments of the readability and style of the paper. The English professor’s remarks on these headings are in (179) and the Linguist’s in (180):

(179) Appropriate to this purpose, objective. Lucid. The subject matter is necessarily complex, including multiple, systemic inter-relationships. The writing style clarifies and exemplifies relationships as simply and directly as necessary.

In this fashion, evidence in support of frames can be garnered even among scholars who are reluctant to admit the existence of such mental constructs.

2. FRAME ATTACHMENT

2.1 A text whose topic is not familiar should make people uncertain about finding a frame. An experiment I conducted with the help of Richard Hersh at the University of Florida pursued this prediction by presenting the following brief text:

(181.1) Sunspots are believed to be caused by magnetic fields inside the sun. (181.2) These fields slow down the energy flowing up from inside the sun, (181.3) so that the gases above them are cooler and seem darker in color.

Two groups of subjects heard the text read aloud and were asked to write down as much as they could remember of it. The second group, however, was required to wait five minutes before writing, during which time no activities were imposed. I felt that direct reproduction from short-term sensory storage would certainly become impossible in that interval.

2.2 If the text were fully and accurately understood as is, the result might be a configuration such as that in Figure 23.

The entire content is encased in a BELIEF SPACE, evoked by ‘are believed’ in the text (cf. Hendrix 1975, 1978).³ [3.I include the inference that the ‘belief is held by ‘scientists’, as verified by our test results (V1.2.4). We also have an instance of an underlying relation-’cause-of-being mapped onto a surface expression. I use the convention of simply passing the link label ‘ca’ (for cause) along to the next node.] In addition to ‘sunspots’, ‘the main nodes are ‘magnetic ‘fields’, and ‘gases’, as indicated by their multiple linkage. The explanation for sunspots is a causal chain: (1) magnetic fields slow down energy; (2) gases above become cooler; (3) cooler places are darker; (4) darkness yields spots.

2.4 The most striking instance was the student who declared the text to be dealing with an ‘eclipse of the sun’, cued apparently by ‘sun’ plus ‘darker’. In contrast, the addition of ‘scientists’ as holders of the ‘belief’ was a fully reasonable inference for the type of material. We tabulated our protocols to find the concepts that fared the best, with these results: ‘magnetic fields’ won out with 25 out of 35; ‘dark’ then followed with 21, ‘gases ’and ‘cool’ with 18 each, and ‘slow’ had only 6 recalls. The tendency to utilize ‘magnetic ficlds’ as a subsuming frame over others is very manifest in the amount of associated entries readers added to that node. Consider these protocol fragments:

(182) Sunspots are believed to be caused by lines of magnetic force which radiate outward from the center of the sun.

Clearly, this content is derived from knowledge about magnetism rather than from the presentation. The same source may have led to their statements about ‘electricity’ (1 subject), ‘force’ (5), ‘radiation’ (3), and ‘disturbance’ (1); and three subjects relocated the fields ‘around the sun’, following a ‘lines of force’ notion like that in (182). On the other hand, the students had a hard time envisioning magnetism slowing down gases. One converted the ‘fields’ to ‘shields’ to make this effect more acceptable.

       (184) Sunspots in the sun are always dark in color.

and another whose protocol was almost the same (both on the no-delay condition!) may have understood or recalled almost none of the text, but knew of course what ‘spots’ should be. One subject drew the concepts of ‘cool’ and ‘dark’ even closer together by eliminating the need for causality:

       (185) The temperature of the gases on the sun varies in color resulting in dark spots.

2.6 The candidate frame of ‘sun’ however, was not heavily used, perhaps because it doesn’t provide much help for this particular text world. This frame may have been used by the subjects who mentioned ‘molten gases’, ‘hotter gases’, and ‘extra heat at the surface of the sun’, where an ‘attribute’ was inferred; and the recall of the ‘spots’ as ‘circles’ evidently by conflation with the form of the sun. Experts on astronomy would doubtless have made better use of the ‘sun’-frame—a sketch is provided in Beaugrande (1979j)— and had a ‘sunspot-formation’-schema securely stored.

2.7 These results only begin to suggest the intricacies of frame attachment. However, they already suggest that even a straightforward, short (37-word) text is processed via general knowledge patterns akin to the “advance organizers” envisioned by David Ausubel (1960). In the following section on schema attachment, I suggest that global knowledge patterns have their own priorities regarding what materials are important and they should get organised.

3. SCHEMA ATTACHMENT

3.1 To explore schema attachment, I return to the ‘rocket’ sample (35) given in full in III.4.20. That text is better treated by a schema than a frame, since it deals with a sequence of events rather than with a description of a rocket as such. Of course, a ‘rocket’-frame could apply to some portions, and we should probably designate processing as dominated by the schema, not as exclusively and exhaustively attached by it.

3.2 A schema can also be represented as a NETWORK, although with a fixed ordering. its nodes appear as a progression of EVENTS and STATES in a time continuum. Our surface text is conventional in that it follows the underlying time order consistently, although, as we shall see, the surface signaling of the various events and states is uneven.

3.3 The ‘flight-schema is regular and balanced, as is graphically clear from Figure 24.

The internal patterning of ‘location’ state exited via a ‘motion’ event that leads to entry into a new ‘location’ state runs recursively throughout. The initiation operator ‘ί’ applies to the initial ‘take-off’ event, and the termination operator ‘τ’ to the’ ‘land’ event. The minimal components of ‘flight’ would be an object in the class of’ ‘flying objects’ (or a subclass of it) that ‘takes off,’ ‘ascends’ to a ‘peak’, then ‘descends’ ‘near the ground’, and ‘lands’ finally ‘on the ground’. The actual text does not make all these events and. states explicit. It follows that if people nonetheless recall them, the evidence for the schema as a mental pattern is fairly strongly proven (more evidence cited in VIII.2.2).

3.5 If this schema were applied as top-down input to the reading of the ‘rocket’ text, its nodes would be inserted as CONTROL CENTERS into the text-world model, and text-world entries confirming each node would be hooked on, yielding a configuration such as we see in Figure 25.

The various densities reflect the different degrees of node confirmation. The cues that the rocket begins down ‘on the ground’ are reliable: ‘stood’ indicates a stationary location that is at

once asserted to be a known geographical region in the southwestern United States. At the moment of being ‘empty’, it could certainly not be in motion yet. Note that despite the lack of tense distinction (‘had weighed’ would be clearer), readers would infer the ‘empty’ state to be earlier than the ‘carry’ of the next assertion. The presence of the entire fuel supply also indicates the rocket still being on the ground, because none has been burned up.

3.6 As was suggested in III.4.29, the proposition ‘everything was ready’ can be taken as subsuming whatever was needed to ‘enable’ the ‘take-off’ event. That event is shown to be imminent by the ‘flares’ serving’ as a signal to fire the rocket’. The ‘take-off’ is expressed only by means of ‘rise’, though the “causal proximity” to ‘roar’ and ‘burst of flame’ points to the initial motion rather than to simply ‘ascending’. The following state of being ‘near the ground’ is only inferentially represented by the first ‘slow’ rate of rising, since at low altitudes, propulsion must work hardest against gravity and inertia. The entry ‘rise’ also represents the ‘ascend’ event, along with ‘sped upward’ and the increase in velocity (‘faster and faster).

3.7 The peak height of this particular flight is never mentioned at all, and must be inferred to have been reached somewhere between ‘sped upward’ and ‘return’. The notion of extreme altitude is at most suggested by the flame’s resembling ‘a yellow star’ and the rocket’s being ‘too high to be seen’ — reasoning by analogy in the first instance, and by disenablement in the second. The ‘descend’ event easily attaches ‘return’, and, by a further inference, the ‘watching plane’ whose altitude would be lower than the rocket’s.