The Role of Cognitive Schemata

8/12/2019 The Role of Cognitive Schemata

1/24

COGNITIVE PSYCHOLOGY 13, 2077230 (1981)

Role of Schemata in emory for Places

WILLIAM F. BREWER AND J AMES C TREYENSUniversiry o f ll inois at Ur bana Champaign

A study of memory for places was carried out to examine five hypotheses aboutthe use of schemata in memory performance: (a) that schemata determine whatobjects are encoded into memory: (b) that schemata act as frameworks forepisodic information; (c) that schema-based information is integrated withepisodic information; (d) that schemata facilitate retrieval; and (e) that schematainfluence what is communicated at recall. Subjects were taken into what theythought was a graduate students office and later were tested for memory of theroom with either drawing recall, written recall, or verbal recognition. Memory

scores for objects were correlated with schema expectancy and saliency ratings.Schema expectancy was positively correlated with recall and recognition. Ex-pected objects were inferred in recall, supporting the integration hypothesis.Comparison of recall and recognition data supported the retrieval hypothesis.Analysis of the written descriptions supported the communication hypothesis.Saliency was positively correlated with recall and recognition for present objects,but was unrelated to retrieval. Saliency was negatively correlated with recognitionfor nonpresent objects, suggesting a metacognitive strategy in recognition of high-salient objects.

Much current research on human memory deals with the intentional

learning of linguistic materials such as words, sentences, or stories. Theseexperiments tend to be analogs of school tasks such as remembering thenames of the states or the content of a paragraph in a textbook. Thepresent paper uses an experimental task that is intended to be more analo-gous to the types of incidental memory that occur in everyday life. Inparticular, we have chosen to focus on memory for places. We are in-terested in investigating the memory processes that are involved whentelling someone where a particular book is on a shelf or describing some-ones office after your first visit.

Recent attempts to deal with memory for stories and with perception ofreal-world scenes have led to the development of theories in which exist-ing knowledge plays a crucial role (Anderson Pichert, 1978; BobrowNorman, 1975; Bower, Black, Turner, 1979; Minsky, 1975; Neisser,1976; Piaget Inhelder, 1973; Rumelhart, 1980; Rumelhart Ortony,1977; Schank Abelson, 1977; Spiro, 1977). It seems to us that the studyof memory for real-world places requires theories of a similar nature.

The representational structures in these theories (schemata, frames,

We would like to thank Richard Anderson, Don Norman, Edward Lichtenstein, EllenBrewer, and David Dupree for comments on an earlier draft of this paper. This research was

supported in part by NIMH Grant MN 29562. Requests for reprints should be sent toWilliam F. Brewer, Department of Psychology, University of Illinois, Psychology Building,Champaign, IL 61820.

207OOlO-0285/81/020207-24 05.00/OCopyright 0 1981 by Academic Press Inc.

AU ri.@ts of reproduction in my form reserved.


2/24

208 BREWER AND TREYENS

scripts, plans, prototypes) will be referred to here as schemata. Schemataare knowledge structures or sets of expectations based on past experi-

ence. They exist at various levels of abstraction and vary in their struc-tural complexity (cf. Rumelhart, 1980).

Schema theories propose that perception, language comprehension,and memory are processes which involve the interaction of new (episodic)information with old, schema-based information. The basic assumption ofschema theories is that an individuals prior experience will influence howhe or she perceives, comprehends, and remembers new information.

Minsky (1975) suggests that perception is a schema-based process oc-curring over time which involves filling in details, collecting evidence,

testing, deducing, and interpreting, on the basis of knowledge, expecta-tions, and goals, Minsky hypothesizes that this complex process can takeplace rapidly because schemata already exist in memory which corre-spond to common environments, such as rooms. Information slots orvariables in the internal structure of the schema which have not been filledwith perceptual information are filled by default assignments based onstereotypic expectations derived from past experience. Kuipers (1975)hypothetical example of room perception illustrates default assignments:if a quick perceptual scan of a room indicates that there is a clock on the

wall, hands may be assigned to the internal representation by default,even though this particular clock does not have hands.

A number of schema theorists have focused directly on memory andhave shown that memory performance is frequently influenced byschema-based expectations. Bartlett (1932) found that subjects expecta-tions and experiences distorted their recall of an unusual North AmericanIndian folktale. Spiro (1977) found that recall of passages can be influ-enced by expectations based on information presented after reading thepassage. Piaget and Inhelder (1973) showed that memories can be mod-ified by schema development occurring between learning and recall. An-derson and Richert (1978) have shown that an activated schema can aidretrieval of information in a recall task. While these studies have clearlydemonstrated that schemata influence memory performance, the detailsof the process have not been worked out.

It appears to us that there are five fundamentally different ways inwhich schemata might influence memory performance: (a) they can de-termine what objects are looked at and encoded into memory; (b) they canact as a framework for new information; (c) they can provide schema-

based information which becomes integrated with episodic information;(d) they can guide the retrieval process; and (e) they can determine whatinformation is to be communicated at output.

Encoding. Minsky (1975) and Neisser (1976) have suggested that per-ception is guided by schemata. However, two different memory predic-


3/24


4/24


5/24

MEMORY FOR PLACES 211

FIG. 1. The experimental room used in the rating study and memory experiment (part ofleft side of room not visible).

objects such as staples and an eraser were low. The schema-expectancyrating was intended to provide an index of subjects graduate-studentoffice schema by asking how likely the object was in the context of theroom. Objects such as the desk and ceiling lights were high on this scale,while objects such as a model brain and a piece of bark were low.

Minskys (1975) paper on the structure of everyday knowledge pro-poses that knowledge about rooms is organized into a room frame. Theroom frame contains the information about rooms that one can be nearlycertain about before encountering a particular room. Since Minsky haspointed out the theoretical importance of this type of knowledge, we werecareful to include in the rating task items to test the room frame. Wedesignated seven items to be room-frame items: walls, floor, ceiling, door,


6/24


doorknob, light switch, and lights. These were chosen on a priori groundsas items that any room would almost certainly contain.

In addition to the objects actually in the room, a sample of 70 objectsnot in the room was chosen for rating on the same two scales. Theseobjects were selected to vary along the full range of both scales. Theseitems served as foils in a verbal recognition test.

The saliency rating task was carried out by 14 subjects, and theschema-expectancy rating task was carried out by 14 different subjects.Subjects were taken into the experimental of&e in groups of two or threeand were given booklets containing the 131 objects to be rated.

The subjects in the saliency rating task were asked to rate the items on a

6-point scale with the following instructions: Rate each object for hownoticeable the object is (or would be) in this room. . . . Remember thatsome of the objects are actually in this room and some are not. If theobject is in the room, base your rating on how noticeable that objectactually is. If the object is not in the room, base your rating on hownoticeable that object would be if it were in this room, given the generalcharacteristics of the room. A response of 1 was defined as, It wouldbe extremely unlikely that I would notice the object, and a response of 6was defined as, It would be extremely likely that I would notice theobject.

The subjects carrying out the schema-expectancy ratings were giventhe following instructions: On the following pages are lists of objects(some actually in the room and some not actually in the room). Followingeach object is a scale from 1 to 6. We would like you to use this scale torate each object for how likely the objects would be to appear in a roomlike this. . . . As an example, if you were in a kitchen carrying out thistask, you would probably give an object such as a refrigerator a highnumber and an object such as a bed a low number. When making yourrating ignore whether or not the object is actually present in the room.Simply base your rating on the likelihood that the object would occur inthis room, given the general characteristics of the room. A response of 1was defined as extremely unlikely that the object would be found in aroom like this, and a response of 6 was defined as extremely likely thatthe object would be found in a room like this.

The mean saliency rating and the mean schema-expectancy rating werecalculated for each of the 131 objects. The two scales are not independent.For the 61 objects present in the room, the correlation between saliency

and schema expectancy is -.41; for the 70 objects not in the room, thecorrelation is -.69. It appears that saliency is based on two factors: asaliency intrinsic to the object and a saliency derived from its schemacontext. For example, a .45-caliber pistol is intrinsically more salient thana paper clip. However, an item also increases in saliency if it does not tit


7/24


into the schema-based context. A spare tire is probably of intermediateintrinsic saliency, but in the context of a graduate-student office it be-

comes highly salient. This context effect appears to produce a portion ofthe negative correlation of saliency with schema expectancy. In addition,some of the negative correlation may be due to item sampling (especiallyfor the nonpresent items). When nonpresent items of high salience areselected, few will be likely to occur in a graduate-student oftice, and sothis intrinsic item bias will also lead to a negative correlation of saliencywith schema expectancy.

Examination of the schema-expectancy scores provided considerablesupport for the existence of a common room schema. The subjects were in

strong agreement that a room such as our experimental room is highly likelyto contain a floor, a desk, a calendar, an eraser, and pencils, but is highlyunlikely to contain a sewing machine or a .45-caliber pistol. The schema-extectancy data also support the a priori selection of room-frame items(items almost certain to be in a room). The six objects with the highestschema-expectancy scores were all room-frame items (door, floor, lightswitch, ceiling lights, walls, ceiling). The other a priori room-frame item,doorknob, was rated slightly below two strong office-frame items, deskand eraser.

The data also suggest that place schemata are not rigid preformedstructures, but instead are much more flexible, utilizing the full extent ofthe subjects knowledge. In the present task the subjects were taken to aroom in a psychology building and were asked to rate the objects in thisgraduate students office. In this context they gave high schema-expectancy ratings for experimental apparatus, worktable, and textbooks,even though it seems quite unlikely that these items would receive highexpectancy scores in a rating of a generic office or of a different type ofoffice, such as the office of a bank president.

The saliency ratings also showed considerable lawfulness. Subjectsgave high saliency ratings to large visible objects (walls, desk), toattention-attracting objects (Playboy centerfold, skull), and to objects notfitting the graduate-student office schema (spare tire, rolling pin). Sub-jects gave low saliency ratings to smalI ordinary objects (eraser, paperclips). Overall, it appears that the rating tasks gave lawful data reflectingpsychologically meaningful processes. The following experiment was de-signed to study the effects of schema expectancy and saliency on memoryfor places.

MEMORY EXPERIMENT

This experiment used the experimental room described above. Thegeneral procedure was designed to allow us to examine the episodic placeinformation retained after a brief exposure to the experimental room in an


8/24


unintentional setting. In order to examine different aspects of memory forthe room, three groups of subjects performed different memory tasks.

Subjects

The subjects were undergraduates fulfilling a course requirement.There was a total of 86 subjects: 30 subjects carried out written recall andthen verbal recognition; 29 subjects carried out drawing recall; and 27subjects carried out verbal recognition only. Subjects were exposed to theexperimental room individually and carried out the memory tests ingroups of one or two.

Procedure for Room ExposureThe experimenter took the subject into the experimental wing of the

Psychology Building and said, I have to check to make sure that theprevious hours subject has completed the experiment. The experi-menter then told the subject that he or she could wait in my office. Atthis point the experimenter opened the door to the experimental office,turned on the lights, started a hidden stopwatch, and asked the subject tohave a seat (all chairs but one had objects on top of them). The experi-menter assured the subject that he would return shortly and then left the

room, closing the door behind him. After 35 set the experimenter reenteredthe office and asked the subject to follow him. The subject was immedi-ately taken to a nearby seminar room where the subject was told that thereal purpose of the experiment was to test his or her memory for theexperimental office and then the memory task was given. Approximately1 min elapsed between the time the subject left the experimental room andthe time the subject began the memory task. The seminar room in whichthe memory tasks were given was quite different from the experimentaloffice. It was a long rectangular room with two doors, a single long table,

and 24 chairs.A questionnaire given after the memory task showed that most subjects

spent their time in the experimental room seated in the chair, lookingaround the room in order to guess what kind of person the graduatestudent was or to see if there were any indications as to what the experi-ment was going to be about. The deception appears to have been quitesuccessful. The crucial item on the questionnaire was, Did you think youwould be asked to remember the objects in the office? On this item, 93%of the 86 subjects responded no.

RecallProcedure. After being taken to the conference room, the subjects given the written

recall task were asked to write down what they could remember about the experimentalroom. The instructions stated, We would like you to describe for us everything you canremember about the room you were just in. . . . For each object please try to give its location


9/24


and as complete a description as you can provide (shape, size, color, etc.). Write yourdescription as if you were describing the room for someone who had never seen it. Imme-diately after the written recall task the subjects were given a verbal recognition test. Discus-sion of the data from this task will be deferred until later in the paper.

Subjects given the drawing recall task were provided with an outline of the experimentaloffice and were asked to draw in the objects they could remember from the room. Theinstructions stated, We would like you to draw everything you can remember about theroom on the provided floor plan. . . . Represent each object in the location you rememberit, and try to draw each objects size to scale. Label each object which you draw with itsname. Subjects in both recall tasks were asked to work for a minimum of 15 min and couldtake up to 30 min.

Results and discussion. Criteria were developed to classify objectsmentioned in the recall protocols as present objects, frame objects, orinferred objects. For the written recall task, present objects were objectsdescribed in enough detail so that someone could take the written pro-tocol into the room and locate the indicated object. Frame objects werethe same seven frame objects included in the rating tasks (door, floor,light switch, ceiling lights, walls, ceiling, doorknob). Inferred objectswere objects given in enough detail so that they could have been identifiedif they had been present, but were not actually in the room. The scoringcriteria were essentially the same for drawing recall. A number of items inthe recall protocols did not fit into these three categories and were elimi-nated from the analysis: (a) objects that were not in the room, but mighthave resulted from misidentification (birds on mobile for fish on mobile);(b) vague responses (the other chair had something lying on it). Inaddition objects that were parts of a present object (keys on typewriter)were eliminated to simplify the scoring.

In written recall, a total of 88 different objects were recalled by one ormore subjects. Of these, 7 were room-frame objects, 62 were presentobjects, and 19 were inferred objects. The recall data for each object aregiven in Table 1. The mean number of frame objects recalled per subjectwas 3.37, the median was 3, and the range was 1 to 6. The mean number ofpresent objects was 13.50, the median 14.5, and the range was 5 to 20. Themean number of inferred objects was 1.13, the median was 1, and therange was 0 to 4. A correlational analysis was carried out to investigatethe effects of object saliency and schema expectancy on written recall.Saliency and schema-expectancy scores were available for 43 of the ob-jects given in recall (room-frame objects and inferred objects were notincluded). The correlation of saliency with recall frequency and the cor-relation of schema expectancy with recall frequency were calculated;since saliency and schema expectancy were negatively correlated, partialcorrelations were also calculated. All of these correlations are given inTable 2. Examination of these correlations shows that there are clearlyindependent effects of saliency and schema expectancy.

The data from drawing recall essentially replicated the findings from


10/24


TABLE INumber of Subjects Recalling Objects in Written Recall

ObjectNo.

subjects ObjectNo.

subjects

Chair (next to desk)Desk

WallChair (in front of desk)Poster (of chimp)Door

Table (worktable)Shelves

CeilingTable (with coffee)Skinner boxChilds chair

FloorLight switch

Toy topBrainParts, gadgets (on

worktable)Swivel chair

Poster on ceiling 10*Books 9

Ceiling lights 9Poster (of food) 9Typewriter 9Bulletin board 8

29292924232222211615141212121211II

11

Clown light switch 8Coffee pot 8Skull 8Mobile 7Road sign 7Calendar 6Wine bottle 6Football-player doll 5Jar of creamer 5Pipe (cord) 5Postcards 5Tennis racket 5Blower fan 4

Colored patterns onceiling lights

Piece of bark*Filing cabinetFrisbeeJar of coffee*Poster (in addition to

those in room)ScrewdriverSnoopy pictureRotary switchesCactus

4

written recall. The correlation of the number of subjects recalling anobject in drawing recall with the number recalling it in written recall is .94.Apparently, at this global level of analysis, drawing recall and writtenrecall are roughly equivalent methods of measuring what information thesubjects have picked up in their brief exposure to the room. In drawingrecall a total of 77 different objects were recalled by one or more subjects.Of these, 61 were present in the office and 16 were inferred. Table 2 givesthe correlations between schema expectancy, saliency, and drawing re-call. The pattern of these correlations is very similar to that obtained withwritten recall.

Overall, the results in the recall tasks give strong support for the actionof schemata in memory for places. The evidence for the integration ofepisodic information with schema-based knowledge was very clear. Therewere a number of inferred objects in recall, and the inferred objects forwhich ratings were available received high schema-expectancy scores.


11/24


12/24


TABLE 2Correlations Between Schema Expectancy, Saliency, and Recall for Written Recall and

Drawing Recall

Recall Schemafrequency expectancy Saliency

Written recallRecall frequencySchema expectancySaliency

Drawing recallRecall frequency

Schema expectancySaliency

1.00.27 (.55*)* 1.00.47* (.64*) -.37* 1.00

1.00

.42* (.68*)b 1.00.36* (.66*) - .43* 1.00

(1n = 43 objects.b Saliency partialled out.c Schema expectancy partialled out.n = 41 objects..*p < .05.

the recall data taken by themselves do not distinguish between schemata

operating in encoding information, as a framework for information, byintegration, or in the retrieval of information. There is, however, otherevidence to suggest that an encoding hypothesis cannot account for this cor-relation. Loftus and Mackworth (1978) and Friedman (1979) have shownthat subjects examining pictures tend to spend more time looking atnonschema objects than at schema-related objects. This suggests thatschemata do influence looking time. However, if their findings can begeneralized to our more naturalistic situation, then the encoding hypothe-sis cannot account for our positive correlation, since as applied to the

looking-time data it would predict better recall for nonschema objectsthan schema-related objects, i.e., a negative correlation of recall fre-quency and schema expectancy.

The positive correlation between saliency and recall probably reflectsthe amount of attention devoted to the salient objects, but given thepresent findings, one could also hypothesize that saliency leads to astronger memory representation or to more efficient retrieval.

Overall, the inferences in the recall data provide clear-cut evidence forthe integration of episodic information and schema-based information.

Furthermore, if it is assumed that on some occasions present objects withhigh schema-expectancy ratings are recalled on the basis of schematicinformation rather than on the basis of episodic information, integrationwould also contribute to the positive correlation between schema-expectancy ratings and recall. However, this correlation might also be


13/24


due to the offtce schema acting as a framework or as a retrieval mecha-nism. The data from the verbal recognition condition were designed to

distinguish between some of the hypotheses about the use of schema-based information that were not distinguished by the recall data.

Verbal RecognitionProcedure. Subjects in this condition received only a verbal recognition test. After

being taken out of the experimental room to the seminar room, they were given a bookletcontaining a list of object names. They were asked to rate each item for how certain theywere that they had seen the named object in the experimental room. The instructions stated,We would like you to indicate how certain you are that you remember seeing each objectby circling one of the numbers from 1 to 6. Use 1 to indicate that you are absolutely certain

that you did not see the object. Use 6 to indicate that you are absolutely certain that you didsee the object. Use the numbers between 1 and 6 to indicate intermediate degrees of mem-ory. The verbal recognition test consisted of 131 object names; 61 of the named objects hadbeen in the experimental office and 70 had not.

Results and discussion. There was a strong effect of schema-basedinferences in the verbal recognition data, as there had been in the recalldata. Of the 51 nonframe items on the verbal recognition test with thehighest recognition scores, 13 were not actually present in the room. It isclear that these 13 inferences reflect the influence of the o&e schema-all but one are in the top one-third of absent objects when ranked in termsof schema-expectancy scores,

In order to examine the effects of schema expectancy and saliency onverbal recognition, a series of correlational analyses was carried out.Table 3 gives the correlations of saliency and schema expectancy with

TABLE 3Correlations between Mean Recognition Scores, Schema Expectancy Ratings, and

Saliency Ratings

RecognitionSchema expectancySaliency

SchemaRecognition expectancy Saliency

Present objects (without frame objects)1.00.21 ( .58*) 1.00.48* ( .69*) - .49* 1.00

RecognitionSchema expectancySaliency

(1 = 54 objects.

Absent objects1.00

.75* ( .52*p 1.00-.69* (-.36*) - .69* 1.00

b Saliency partialled out. Schema expectancy partialled out. n = 70 objects.*p < .05.


14/24


verbal recognition scores for present objects (not including the room-frame objects) and for absent objects. First we will discuss the positive

correlations of schema expectancy and verbal recognition for presentobjects. On the basis of the previous recall data, we concluded that thepositive correlations of schema expectancy with drawing recall and writ-ten recall could be the result of schema-based information operating as aframework in memory, operating as a retrieval mechanism, or becomingintegrated with the episodic information. In the present recognition data,the verbal recognition items themselves should serve as retrieval cues andso should eliminate any effect due to retrieval processes. Thus, it appearsthat we can eliminate the schema-based retrieval account of the recogni-

tion data. Therefore the positive correlation between schema expectancyand verbal recognition should be due either to schema-based informationoperating as a framework in memory or to schema-based informationbecoming integrated with episodic information from the room.

The positive correlation of schema expectancy and verbal recognitionfor absent objects can apparently be accounted for by only one schema-based explanation. The positive correlation indicates that the subjectswere more likely to state that they had seen a nonpresent object if theobject was a strong member of the office schema. Since these decisions

were being made for objects for which no episodic information was avail-able, the responses must have been due totally to old schema-basedknowledge, and therefore reflect the operation of the schema in integra-tion, not as a framework or as a retrieval mechanism.

Next we will examine the correlations between saliency and verbalrecognition. The positive correlation between saliency and recognition forpresent objects is probably due to increased looking time allocated to themore salient objects resulting in better memory. However, it is possiblethat the memory representation for salient objects is somehow simply

stronger than that for nonsalient objects.The correlation of saliency and verbal recognition scores for absent

objects shows a qualitatively different pattern of results from that ob-tained for objects in the room. The correlation for objects not present inthe room is negative. This suggests that when dealing with more ecologi-cally valid memory tasks, a simple recognition task is not so simple.The negative correlation apparently results from the following processduring the recognition task. When the subject is asked if some very salientobject (e.g., a rat in a cage, or a Playboy centerfold) was in the room, the

subject reasons: (a) if the object had been in the experimental room whenI was there, I would have noticed it; (b) if I had noticed such a salientobject, I would have remembered it; (c) I do not remember such an object;(d) thus, the object must not have been in the room. If a subject follows


15/24


this reasoning process, then the higher the saliency of an object, the Iowerthe likelihood of a false recognition response.

This I-would-have-seen-it-if-it-had-been-there phenomenon appears tobe a variant of a type of inference discussed by Collins (1978) in a paper onreasoning in answering everyday questions. Collins calls this type of in-ference a lack-of-knowledge inference and states that it is very common inhis data on answering everyday factual questions. Overall, the verbalrecognition results support the operation of schemata in integration andthe operation of a metacognitive strategy which allows subjects to avoidmaking false recognition responses for absent items of high saliency.

Verbal Recognition after Written RecallProcedure. The subjects who provided the written recall data were given a verbal

recognition test immediately afterward. Of the present objects given in the recall protocols,80% had been included on the verbal recognition test. Thus, for many items we had data onrecall and recognition for the same objects from the same subject.

Results and discussion. When subjects recalled an item in written re-call, they almost always (96%) gave it a rating of 6 (absolutely certain Iremember seeing the object) in verbal recognition. However, they alsogave ratings of 6 to many of the present objects that they had not been

able to recall. In order to examine the issue of retrieval from memory, aretrieval ratio was developed for each object. The retrieval ratio wasdefined as the number of times an object was given in the recall taskdivided by the number of times it was given a 6 in the verbal recognitiontask. Thus, if 12 subjects gave an item a 6 in verbal recognition, but only 6subjects had recalled the item on the written recall test, the retrievalratio for that item would be .50. A high ratio (near 1.0) indicates thatmost subjects who gave the object a 6 in recognition also recalled theobject. A low ratio (near 0) indicates that most subjects who gave the

object a 6 in recognition did not recall the object. Retrieval ratios werecalculated for all recalled present objects which were given ratings of 6 onthe verbal recognition test by at least three subjects. A correlationalanalysis was carried out on the 36 items that met this criterion. Thecorrelation of saliency and the retrieval ratio was - .04, while the correla-tion of schema expectancy and retrieval ratio was .56 @ < .Ol). Since thecorrelation of schema expectancy and saliency for this set of items wasonly -. 19, the partial correlations with retrieval ratio were not very dif-ferent from the simple correlations (.56 for expectancy and retrieval ratio

with saliency partialled out, and .08 for saliency and retrieval ratio withexpectancy partialled out).

The significant correlation of retrieval ratio and schema expectancysuggests that the office schema is used in the retrieval process. The near-


16/24

222 BREWER AND TRJZYENS

zero correlation between retrieval ratio and saliency suggests that theretrieval process is unaffected by an objects saliency. Given that an

object is in memory (as indicated by the recognition task), it is more likelyto be written down in the recall task if it is related to the offtce schemathan if it is not. A concrete example of this effect is as follows: Thetypewriter was ranked high in the graduate-student offtce schema (rank of12.5 among present, nonframe objects) and was given in written recall by9 of the 10 subjects who gave it a rating of 6 in verbal recognition. Incontrast, the skull was ranked low in the graduate-student office schema(rank 48 for present, nonframe objects on schema expectancy) and wasrecalled by only 8 of the 16 subjects who gave it a rating of 6 in recogni-

tion, despite its high saliency (rank 2 for present objects). Thus, the re-trieval ratio analysis provides strong evidence that the room schema isbeing used as a retrieval mechanism in the recall task. In all of theanalyses involving verbal recognition after recall the seven room-frameitems have been omitted. Comparison of recall and recognition scores forthese items suggests that the recall scores were influenced by the opera-tion of the room schema in communication, as discussed below.

Written Recall: Qualitative Analysis

In addition to the quantitative findings already reported, the recall pro-tocols from the subjects in the written recall condition provided qualita-tive information that allowed further examination of the use of schematain integration and in communication.

Schematadanonical location. Analysis of the recall protocolssuggests that the office schema contains some information about the ca-nonical location of objects. When books are inferred, they are almostalways recalled as being on the shelves, while inferred objects such aspencils and pens are almost always recalled as being on the desk. Objects

that were present in the room, but not in canonical locations, tended toshift to more canonical locations in recall. For example, there was ayellow pad on the seat of the desk chair. In the written recall condition,the only subject who recalled the yellow pad recalled it as being on thedesk.

Schemata-olor and shape. The use of schema-based informationcould also be seen in the recall of colors. The desk in the experimentalroom was brown. In written recall two subjects recalled the color cor-rectly, but four subjects recalled it as gray. There were similar effects for

shape. For example, the trapezoidal worktable was recalled as beingsquare by one of the subjects in written recall. Thus, the qualitativeanalysis of the recall data provides additional support for the effects of theintegration of episodic information and schema-based information.

Communication. The qualitative analysis of the recall data provided


17/24


important data on the use of schemata in determining how informationabout the room was communicated to their audience. In the recall in-

structions the subjects were asked to give each objects location and ascomplete a description as you can provide (shape, size, color, etc.). Yet,in practice, there were enormous differences across objects as to howmuch auxiliary information was given. Much of the variation here appearsto be attributable to the subjects use of Grices (1975) maxim of Quantity.Grice has argued that in carrying on conversations speakers obey a maximof Quantity, by which a speaker attempts to make the contribution asinformative as required, but not more informative than required. Thesubjects decision to report auxiliary information in the recall protocols

appears to be based on the subjects assumptions about their audiencesschema-based knowledge of the room and its contents and on the applica-tion of Grices maxim of Quantity.

Communication-size. There were two objects in the room that werenot of canonical size (the childs chair and the worktable). Size informa-tion was given for these two objects much more frequently than for anyother objects in the room. For the childs chair 67% of the 12 subjects whorecalled it used a size qualifier, whereas for the normal-size chair that thesubjects sat in, not one of the 29 individuals who recalled it used a size

qualifier. For the worktable 82% of the responses contained a size qual-ifier, whereas for the desk only 17% used a size qualifier. It appears thatthe subjects were obeying Grices maxim. They omitted size informationfor objects of usual size, since they could assume that this informationwas already available in their audiences office-object schemata; how-ever, for the objects of atypical size they added the size information.

Communication-material. There was a similar effect for informationabout the materials out of which objects were constructed. The subjectsnever gave auxiliary information about the materials for objects such as

posters, postcards, the frisbee, or the wine bottle, since they could as-sume schema-based knowledge concerning what materials these objectswere made from. The one object in the room that was slightly deviant withrespect to material was the chair the subjects sat in-it was constructed ofplastic. Auxiliary information about the material was given by 55% of thesubjects who recalled this chair. Auxiliary information about materialswas also given fairly frequently for the other pieces of furniture. Here, itwas not that the material was unusual, but that it was indeterminant (i.e.,most office furniture could be constructed of either metal or wood). The

rates of material qualifiers in these cases were: 24% for the desk, 25% forthe desk chair, 33% for the coffee table, and 18% for the worktable.

Communication-shape, orientation, and location. The results weresimilar for the other categories of auxiliary information. Shape informa-tion was rarely given, and when it was given it was for slightly noncanoni-


18/24


cal or indeterminant objects (e.g., the chair the subjects sat in had acurved back and 14% of the subjects included that information in their

recalls). Orientation information (as distinct from location) was rarelygiven except for the one object in the room in a highly unusualorientation-a large metal detour sign on the worktable pointing towardthe ceiling. In recall, 86% of the subjects gave orientation informationwhen recalling the sign, whereas the next highest rate of orientation in-formation was only 22%. Location information (for present nonframeobjects) was almost always given in the recall protocols (93%), presum-ably because the exact location of most objects is not given by the officeschema.

In general, the differential rates of recall of auxiliary information appearto reflect the use of schema information in conjunction with the maxim ofQuantity. Even though the instructions said to report auxiliary informa-tion, the subjects systematically omitted information that was derivablefrom the room and object schemata, and reported auxiliary informationwhen it deviated from the schemata or when it was indeterminant withrespect to the schemata. The subjects were asked to recall the room as ifthey were describing it to someone who had never seen it. The datasuggest that they were able to do this quite successfully. If they had beendescribing the room to the experimenter, there would have been no needto tell the experimenter that the worktable was small or that the chair theyhad been sitting in was made of plastic, since clearly the experimenteralready knew this. Instead, the schema information and maxim of Quan-tity were used as if the audience was an idealized average individual whoknows about graduate-student offices in general, but not about this par-ticular graduate-student of&e.

Communication-negative statements. There were 11 sentences aboutobjects in the room which contained negatives: 3 subjects stated that theroom was not large, 2 stated that there were no windows, and 6 state-ments occurred once-that there were no rugs, that the chair was notagainst the wall, that the desk was not covered with knickknacks, that thecoffee table was not made of wood, that the Skinner box had no top, andthat the worktable was not square. Without a theoretical construct such asthe room schema, the occurrence of these negative sentences is bizarre.The subjects were instructed to write down what they had seen in theroom, and thus had no apparent reason to produce any negations. How-ever, the explanation in terms of schemata seems quite straightforward.

When communicating the information about the room, the subjects wereconsidering their audiences office schema and pointing out the nonexis-tence of things that the subjects thought their audience might mistakenlyinfer on the basis of that schema. Thus, the list of negated sentences is, ineffect, an explicit list from the subjects of those aspects of the experi-mental room that the subjects found to deviate from their schema.


19/24


Communication -article usage. Another example of room and officeschema information determining aspects of the recall protocols is in arti-cle usage. In traditional grammatical accounts of English (e.g., Stockwell,Schacter, & Pat-tee, 1973), the article a is said to be used to introduce newinformation into a discourse, while the is used anaphorically to refer tosomething that has already been mentioned in the discourse. For exam-ple, I saw a platypus. The platypus was running along the river bank.More recent discussions of article usage (Linde, 1975; Norman &Rumelhart, 1975) have pointed out that a is used to introduce somethinginto the current shared knowledge of the speaker and hearer, while the isused for things that are already in the shared knowledge of the speakerand hearer. Thus, John moved into a new house last year. The kitchen isbeautiful. In this case, the speaker introduces the concept of Johnshouse into the shared knowledge of the speaker and hearer, and then canrefer to the kitchen, since kitchens have been introduced by the houseschema.

Since English article usage is sensitive to shared knowledge, we exam-ined the article usage in the written recall data to see if it reflectedschema-based knowledge. For each present object in the written pro-tocols, the percent usage of the on first mention was calculated. Theresults showed a clear trimodal distribution: The high-usage set consistedof 7 objects (door, 100%; doorknob, 100%; floor, 100%; ceiling, 94%; lightswitch, 83%; walls, 81%; lights, 44%). The low-usage set consisted of 6objects (bulletin board, 13%; typewriter, 11%; desk, 10%; shelves, 9%;chair, 7%; coffee table, 7%). The remaining 56 objects were never intro-duced with the on first mention. These results show remarkable sensitiv-ity to shared schema knowledge on the part of the subjects. If one weredescribing the experimental room to someone who has been in it, then itwould be quite natural to use an introductory the for unexpected andhighly salient objects (e.g., the skull, the road sign), since one can assumethat the other individual saw them and (since there was only one object ofthat type in the room) there could be no confusion about what was beingreferred to. The recall protocols show not a single example of introduc-tory the for this type of object; clearly the subjects were (as instructed)directing their recall protocols to an audience whose schemata containedno specific information about the particular experimental room the sub-jects had been exposed to.

Examination of the objects that fell into the high article usage set shows

that once it has been established that a room is being recalled, the subjectsassume that the room-frame objects are in their audiences room schema.It should be noted that the seven objects selected by the empirical crite-rion of high introductory the usage are identical with the seven a prioriroom-frame objects selected by the experimenters when the room wasbeing organized and scaled. Examination of the objects that fell into the


20/24


low-usage group suggests another subtle use of schema knowledge. All ofthese items are instances of an office schema (as distinct from a room

schema). It appears that a subset of subjects makes the assumption that itis possible to assume shared knowledge about objects in the office schemawhen communicating about an office. Overall, the article-usage data fromthe written recall protocols suggest that the subjects almost always as-sumed in their audience the presence of a room schema, sometimes as-sumed the presence of an office schema, and never assumed the presenceof a schema for the particular experimental room.

Communication-missions. Another aspect of the use of schema in-formation in communication is in determining what information can be

omitted. Very strong schema-related information such as room-frame in-formation appears to be frequently omitted in the written protocols. Thisreduced production of room-frame objects in the written recall task wasshown by subtracting an objects rank in recall (for present and room-frame objects) from its rank in verbal recognition (for present and room-frame objects; data from the verbal recognition only condition). Thesescores were calculated for the 15 objects with the highest recognitionscores. There were 9 objects with negative scores (i.e., rank in recallbelow that expected from rank in recognition), and 7 of these 9 were the

room-frame items. This analysis indicates that the room-frame objects arewritten down in recall less frequently than would be expected on the basisof their recognition scores, presumably because the subjects assume theroom-frame information is already known to their audience.

Overall results. The qualitative analysis of the written recall protocolsshowed the integration of episodic information and schema-based infor-mation in the production of location, color, and shape inferences. Inaddition, the qualitative analysis found strong effects of schema-basedknowledge determining what the subjects chose to communicate to the

experimenter: They reported auxiliary information that was unusual orindeterminant with respect to room and object schemata; they omittedauxiliary information given by the schemata; they explicitly denied theexistence of information they thought might be mistakenly inferred on thebasis of the schemata; and finally, they adjusted their article usage inaccordance with the schema information they assumed was available totheir audience, using the definite article to introduce objects that theyassumed were shared knowledge.

GENERAL DISCUSSIONThis experiment shows that it is possible to bring the study of real-

world place memory into the laboratory. The choice of an ecologicallyvalid situation, such as the unintentional memory for rooms, has beenimportant in theory development. Many recent discussions of schematheory have been relatively global and nonspecific. However, the com-


21/24


plex and meaningful nature of the information available in the room led toan intricate set of empirical relations in our experiment and allowed us to

be somewhat more specific in our theorizing and to show the operation ofschemata on a number of different aspects of memory for places.

Our initial scaling of the office for schema expectancy and saliencyuncovered a number of interesting characteristics of our subjects per-ceptions of real-world places. The subjects use of the schema-expectancyscale made it quite evident that they were relating our experimental roomto their long-term knowledge of offices, e.g., they thought that pencilsbelonged in our office, but that a .45-caliber pistol did not. The fact thatthe subjects gave high schema-expectancy ratings to objects such as the

experimental apparatus when they thought they were dealing with a psy-chology graduate-student office suggests that place schemata cannot beconsidered to be rigid frames, but are capable of rather subtle readjust-ments.

The negative correlation we found between schema expectancy andsaliency suggests that both of these dimensions must be considered in ananalysis of real-world places. Comparison of the ratings suggested thatsome objects have an intrinsic saliency, but that objects without a highintrinsic saliency are made salient if they deviate from the overall place

schema.The results of the memory experiment provided support for the opera-

tion of schemata in several quite specific aspects of place memory:Encoding. Our experimental procedures provided no data on looking

time per object, and so we have no evidence on the use of schemata inencoding. The work of Loftus and Mackworth (1978) and Friedman (1979)on looking at meaningful pictures suggests that schemata are involved indetermining looking time. Both of these studies have found that lookingtime is longer for nonschema objects. The encoding hypothesis states that

memory is a direct function of looking time. Therefore the looking-timedata as applied to the encoding hypothesis predict that memory for low-schema objects should be better, but our basic findings are that high-schema objects show better memory performance. If the looking-timedata can be generalized to our experimental paradigm of unintentionalexposure to an actual room, then the use of schemata to determine look-ing time cannot be used to account for our memory results.

Framework. The results of this experiment allowed no unique test ofthe hypothesis that schemata can serve as a framework which preserves

schema-relevant episodic information. It is possible that the positive cor-relations of schema expectancy with recall and with verbal recognitioncould result from the room schema serving as a framework to improverecall of schema-related objects from the experimental room. However,these positive correlations could also be explained by the integration ofepisodic information from the experimental room with old room-schema


22/24


knowledge, or, for the recall data, by the use of the room schema as aretrieval mechanism. Thus, there are several findings that could support

the framework hypothesis, but in each case, there is at least one otherplausible hypothesis that can give an equivalent account of the data.

Integration. The hypothesis that memory for places is often a result ofthe integration of new episodic information with old schema-based infor-mation was given strong support in this experiment. In both drawingrecall and written recall, the subjects recalled a number of objects thatwere not in the experimental room. The inferred objects were invariablyhigh-schema-relevant .objects. Thus, the recall of these items must havebeen due to schema-based knowledge about offtces becoming integrated

with the actual episodic information about the experimental room. Thehigh positive correlation of schema expectancy and verbal recognitionscores for objects that were not present in the room is also strong supportfor the integration hypothesis, since for objects not present, there couldbe no episodic information, and all of the effect must be due to old-schema knowledge. The finding that a number of items with high verbal-recognition scores were not actually in the experimental room and thatthese items were high-office-schema items shows the power of schemainformation in influencing place memory.

Retrieval. The hypothesis that schema information can serve as a re-trieval mechanism was supported by the analysis of recall and verbalrecognition data for the same objects from the same subjects. Given thatan item was strongly recognized in verbal recognition, it was more likelyto have been written down in the recall task if it was a high-office-schemaobject. Thus, the office schema apparently enabled subjects to retrieveobjects from memory that otherwise would have been inaccessible. Giventhis finding, it seems very likely that at least part of the general positivecorrelation of schema expectancy and recall was due to office-schema

information serving as a retrieval mechanism in recall.Communication. The analysis of the written recall data shows powerful

effects of schemata on the communication of responses. The subjectsreport auxiliary information that is unusual with respect to the roomschema, and tend not to write down information that they can assume isknown by their audience. They adjust their article usage in accordancewith the place-schema information they assume in their audience, usingthe definite article to introduce objects that they feel are given by theroom schema.

The finding that the communication effect in our experiment is to em-phasize recall of nonschema information and deemphasize the recall ofinformation given by the schema, avoids a problem that has frequentlyoccurred in other research. In studies of the effect of schemata on recall ofnarratives (Anderson & Pichert, 1978; Rumelhart, 1977; Thorndyke,1973, the general finding is better recall of schema-relevant information.


23/24


Most of these investigators have wanted to show an effect of schemata onmemory, and yet with this type of material it is always possible that the

schemata are operating in the communication of the recalls (i.e., the sub-jects choose to omit the nonschema information at recall, because theyassume it is not important). Thus, in studies of the recall of narratives, thepotential effects of schemata in memory are confounded with the effectsof schemata in communication. In the present experimental paradigm, thesubjects adopted the strategy of communicating the nonschema informa-tion and not communicating the information given by the schema. Theadoption of this communication strategy by the subjects in our writtenrecall task avoided the usual confounding of an effect of schemata in

communication with the other uses of schemata and allowed us to providemore precise tests of other hypotheses.

Saliency. The initial scaling of the experimental office demonstratedthat saliency must be distinguished from schema expectancy. It seemslikely that the positive correlation between saliency and memory perfor-mance found in all three experimental conditions reflects the amount ofattention devoted to the salient objects; however, other hypotheses arepossible. The negative correlation of saliency with verbal recognition ofabsent objects suggests the operation of a powerful metacognitive

strategy in recognition memory. The subjects use an I-would-have-seen-it-if-it-had-been-there strategy to avoid making false recognition re-sponses for items of high saliency.

Overall, the results of this experiment suggest that place schemata playan important and complex role in place memory. The interaction of theschemata in various aspects of the recall process produces an interestinginverted U-shaped function of the relationship of schema expectancy andrecall of objects, given that the objects are shown to be in memory by arecognition test. Objects of very high schema expectancy (e.g., room-frame objects) are not given in recall as much as might be expected, sincethe subjects assume that they are known to their audience. The objectsof high to medium-high schema expectancy are typically recalled. Theobjects of low schema expectancy are not recalled as much as might beexpected from their recognition scores because there are no schemata tofacilitate their retrieval.

REFERENCESAnderson, R. C., & Pichert, J. W. Recall of previously unrecallable information following a

shift in perspective. Journal of Verbal Learning and Verbal Behavior, 1978, 17, 1- 12.Bartlett, F. C. Remembering. Cambridge, England: Cambridge Univ. Press, 1932.Bobrow, D. G., & Norman, D. A. Some principles of memory schemata. In D. G. Bobrow

& A. Collins (Eds.), Representation and understanding: Studies in cognitive science.New York: Academic Press, 1975.

Bower, G. H., Black, J. B., & Turner. T. J. Scripts in memory for text. Cognitive Psychol-ogy, 1979, 11, 177-220.


24/24


Collins, A. Fragments of a theory of human plausible reasoning. In D. L. Waltz (Ed.),Theoretical issues in natural language processing-2. New York: Association forComputing Machinery, 1978.

Downs, R. M., & Stea, D. Maps in minds: Reflections on cognitive mapping. New York:Harper & Row, 1977.

Friedman, A. Framing pictures: The role of knowledge in automatized encoding and mem-ory for gist. Journal of Experimental Psychology: General, 1979, 108, 316-355.

Grice, H. P. Logic and conversation. In P. Cole 8z J. L. Morgan (Eds.), Syntax and seman-tics, Vol. 3, Speech acts. New York: Seminar Press, 1975.

Kuipers, B. .I. A frame for frames: Representing knowledge for recognition. In D. G. Bob-row & A. Collins (Eds.), Representation and understanding: Studies in cognitive sci-ence. New York: Academic Press, 1975.

Lichtenstein, E. H., &i Brewer, W. F. Memory for goal-directed events. Cognitive Psychol-

ogy, 1980, 12, 412-445.Linde, C. The linguistic encoding of spatial information (Doctoral dissertation, ColumbiaUniversity, 1974).Dissertation Abstracts International, 1975, 35, 4483A. (UniversityMicrofilms No. 74-28, 512)

Loftus, G. R., & Mackworth, N. H. Cognitive determinants of fixation location duringpicture viewing. Journal of Experimental Psychology: Human Perception and Perfor-mance, 1978, 4, 565-572.

Lynch, K. The image of the city. Cambridge, MA: MIT Press, 1960.Minsky, M. A framework for representing knowledge. In P. H. Winston (Ed.), The psychol-

ogy of computer vision. New York: McGraw-Hill, 1975.Neisser, U. Cognition and reality. San Francisco: Freeman, 1976.

Norman, D. A. Memory, knowledge, and the answering of questions. In R. L. Solso (Ed.),Contemporary issues in cognitive psychology: TZie Loyola symposium. New York:Wiley, 1973.

Norman, D. A., & Rumelhart, D. E. Reference and comprehension. In D. A. Norman,D. E. Rumelhart, L LNR Research Group, Explorations in cognition. San Francisco:Freeman, 1975.

Piaget, J., & Inhelder, B. Memory and intelligence. New York: Basic Books, 1973.Rumelhart, D. E. Understanding and summarizing brief stories. In D. LaBeIge & S. J.

Samuels (Eds.), Basic processes in reading: Perception and comprehension. Hillsdale,NJ: Lawrence Erlbaum, 1977.

Rumelhart, D. E. Schemata: The building blocks of cognition. In R. J. Spiro, B. C. Bruce, &W. F. Brewer (Eds.), Theoretical issues in reading comprehension: Perspectives fromcognitive psychology, linguistics, artificial intelligence, and education. Hillsdale, NJ:Lawrence Erlbaum, 1980.

Rumelhart, D. E., & Ortony, A. The representation of knowledge in memory. In R. C.Anderson, R. J. Spiro, & W. E. Montague (Eds.), Schooling and the acquisition ofknowledge. Hillsdale, NJ: Lawrence Erlbaum, 1977.

Schank, R. C., & Abelson, R. P. Scripts, plans, goals, and understanding. Hillsdale, NJ:Lawrence Erlbaum, 1977.

Spiro, R. J. Remembering information from text: The state of schema approach. In R. C.Anderson, R. J. Spiro, & W. E. Montague (Eds.), Schooling and the acquisition ofknowledge. Hillsdale, NJ: Lawrence Erlbaum, 1977.

Stockwell, R. P.,,Schacter, P., & Partee, B. H. The major syntactic structures of English.New York: Holt, Rinehart, & Winston, 1973.

Thorndyke, P. W. Cognitive structures in comprehension and memory of narrative dis-course. Cognitive Psychology, 1977, 9, 77- 110.

(Accepted September 11, 1980)

The Role of Cognitive Schemata

Documents

Transcript of The Role of Cognitive Schemata