The Role of Imagery in Sentence Memory: A...

10
The Role of Imagery in Sentence Memory: A Developmental Study Stephen M. Kosslyn and Gordon H. Bower Stanford University KOSSLYN, STEPHEN M., and BOWER, GORDON H. The Role of Imagery in Sentence Memory: A Developmental Study. CHILD DEVELOPMENT, 1974, 45, 30-38. This experiment tested whether children rely relatively more than adults on sensory imagery as the internal represen- tation of sentences in memory. If so, then children should show more confusions in recognition memory between sentences which evoke very similar imagery. Adults, on the other hand, should be able to encode, remember, and distinguish sentences differing conceptually even though these sentences evoke very similar imagery. The experiment used sentence pairs rated as evoking highly similar imagery. Ss studied 1 list of sentences and then judged a test list of sentences regarding whether each was a verbatim replica of 1 studied earlier. Compared with a studied counterpart, each test sentence was either verbatim the same, imaginally similar and conceptually similar, imaginally similar but conceptually different, or imaginally and con- ceptually different. Recognition confusion errors supported the hypothesis that children did not discriminate between same versus imaginally similar sentences whereas adults did. Al- though children comprehended conceptual distinctions nearly as well as adults, they forgot such distinctions more readily, so that imaginal similarity guided their recognitions. Several developmental psychologists (e.g., Bruner, Olver, & Greenfield 1966; Church 1961) have proposed that imagery is the pre- dominant mode of cognitive representation of information in children before they develop facility with language. While older children and adults might habitually encode informa- tion in tenns of linguistic abstractions (e.g., the "conceptual structures" described by Schank [1972]), it is alleged that children rely relatively more on concrete, sensory imagery. The present study investigates an implication of this thesis regarding children's and adults' memory for sentences. While previous devel- opmental investigations bave examined tbe role of imagery in memory for visual patterns (e.g., London & Robinson 1968, 1971), in memory for paired associates (e.g., Rohwer 1970; Wolff & Levin 1972), and in various types of transformations and problem-solving tasks (e.g., Piaget & Inbelder 1971), no previ- ous developmental studies have considered the role of imagery in memory for sentences. We shall be examining confusion errors in recognition memory for sentences. The logic of using recognition confusions is rougbly tbe same as tests for "stimulus equivalence" and is as follows: It is supposed tbat wben S studies and comprebends a particular sentence, be sets up an internal representation of tbe information he has extracted from it. Although S's imme- diate memory may include aspects of the sen- tence's exact phonetic (or literal) presentation, these "surface features" presumably decay from short-term memory, leaving as the more enduring residue something like the concep- tual or imaginal representation of the sentence studied. A list of studied sentences will leave behind a set of such memory structures. Upon presentation of a test sentence in a later recog- nition memory test, its surface structure will be encoded into a deep-structure form; this input structure is presumed then to be matched in some way against the set of memory struc- tures establisbed by the study list. If the deep structure of the test sentence is sufficiently "similar" to some one or more structures in memory (the exact metric is undefined), then The authors express their appreciation to C. J. Sengul and Sheri Robinson who assisted in collection of the data and to the staff of the Bing Nursery School. Research of the second author is supported by grant MH-13950 from the National Institute of Mental Health. Reprints may be obtained from Stephen Kosslyn, Department of Psychology, Stanford University, Stanford, California 94305. iCkild Development, 1974, 45, 30-38. © 1974 by the Society for Research in Child Development, Inc. All rights reserved.]

Transcript of The Role of Imagery in Sentence Memory: A...

Page 1: The Role of Imagery in Sentence Memory: A …web.stanford.edu/~gbower/1974/role_imagery_sentence...The Role of Imagery in Sentence Memory: A Developmental Study. CHILD DEVELOPMENT,

The Role of Imagery in Sentence Memory:A Developmental Study

Stephen M. Kosslyn and Gordon H. Bower

Stanford University

KOSSLYN, STEPHEN M., and BOWER, GORDON H. The Role of Imagery in Sentence Memory:A Developmental Study. CHILD DEVELOPMENT, 1974, 45, 30-38. This experiment testedwhether children rely relatively more than adults on sensory imagery as the internal represen-tation of sentences in memory. If so, then children should show more confusions in recognitionmemory between sentences which evoke very similar imagery. Adults, on the other hand,should be able to encode, remember, and distinguish sentences differing conceptually eventhough these sentences evoke very similar imagery. The experiment used sentence pairs ratedas evoking highly similar imagery. Ss studied 1 list of sentences and then judged a test listof sentences regarding whether each was a verbatim replica of 1 studied earlier. Comparedwith a studied counterpart, each test sentence was either verbatim the same, imaginally similarand conceptually similar, imaginally similar but conceptually different, or imaginally and con-ceptually different. Recognition confusion errors supported the hypothesis that children didnot discriminate between same versus imaginally similar sentences whereas adults did. Al-though children comprehended conceptual distinctions nearly as well as adults, they forgotsuch distinctions more readily, so that imaginal similarity guided their recognitions.

Several developmental psychologists (e.g.,Bruner, Olver, & Greenfield 1966; Church1961) have proposed that imagery is the pre-dominant mode of cognitive representation ofinformation in children before they developfacility with language. While older childrenand adults might habitually encode informa-tion in tenns of linguistic abstractions (e.g.,the "conceptual structures" described bySchank [1972]), it is alleged that children relyrelatively more on concrete, sensory imagery.The present study investigates an implicationof this thesis regarding children's and adults'memory for sentences. While previous devel-opmental investigations bave examined tberole of imagery in memory for visual patterns(e.g., London & Robinson 1968, 1971), inmemory for paired associates (e.g., Rohwer1970; Wolff & Levin 1972), and in varioustypes of transformations and problem-solvingtasks (e.g., Piaget & Inbelder 1971), no previ-ous developmental studies have considered therole of imagery in memory for sentences.

We shall be examining confusion errors in

recognition memory for sentences. The logic ofusing recognition confusions is rougbly tbesame as tests for "stimulus equivalence" and isas follows: It is supposed tbat wben S studiesand comprebends a particular sentence, be setsup an internal representation of tbe informationhe has extracted from it. Although S's imme-diate memory may include aspects of the sen-tence's exact phonetic (or literal) presentation,these "surface features" presumably decayfrom short-term memory, leaving as the moreenduring residue something like the concep-tual or imaginal representation of the sentencestudied. A list of studied sentences will leavebehind a set of such memory structures. Uponpresentation of a test sentence in a later recog-nition memory test, its surface structure willbe encoded into a deep-structure form; thisinput structure is presumed then to be matchedin some way against the set of memory struc-tures establisbed by the study list. If the deepstructure of the test sentence is sufficiently"similar" to some one or more structures inmemory (the exact metric is undefined), then

The authors express their appreciation to C. J. Sengul and Sheri Robinson who assistedin collection of the data and to the staff of the Bing Nursery School. Research of the secondauthor is supported by grant MH-13950 from the National Institute of Mental Health. Reprintsmay be obtained from Stephen Kosslyn, Department of Psychology, Stanford University,Stanford, California 94305.

iCkild Development, 1974, 45, 30-38. © 1974 by the Society for Research in Child Development, Inc. Allrights reserved.]

Page 2: The Role of Imagery in Sentence Memory: A …web.stanford.edu/~gbower/1974/role_imagery_sentence...The Role of Imagery in Sentence Memory: A Developmental Study. CHILD DEVELOPMENT,

it is assumed that S decides that the test sen-tence is one he remembers from the study list.

Focusing upon a particular study sentenceand its internal representation, one can thenimagine a range or domain of other sentenceswhich differ in their surface features but whichare all equally confused in long-term memorywith the originally studied sentence. Now,recognition confusion may occur for a varietyof reasons, including, for example, simple fail-ure to learn all of a complex proposition. Buta sure cause for confusing the test sentencewith the memory of the studied sentence is ifthe encoding mechanism maps them into thesame or very similar internal representations.Given these working assumptions, one canthen use systematic confusion errors in recog-nition memory as an index (albeit a "noisy"one) of "similarity" of the internal representa-tions in memory of two sentences. In turn,knowledge of which pairs of surface sentencesare encoded similarly and which pairs differ-ently may provide some leads in inferring thenature of the representations and the mecha-nism which encodes surface strings into theserepresentations. At least, that is the hope un-derlying use of confusions in recognition mem-ory for investigating cognitive representations(see Fillenbaum [1970] for discussion of someproblems with this methodology).

The present experiment inquires whetherthe internal representation of a sentence inmemory has a substantial imagery component;more specifically, we ask whether this imagerycomponent is more prominent for childrenthan for adults. The logic, again, is that if asentence is remembered in terms of an imagethat it evokes, then a new test sentence("lure") which evokes about the same imageryand which is encoded and judged primarily interms of this imagery will then be "recognized"as a repetition of the earlier sentence studied(e.g., a "false alarm").

Previous studies with adults by Begg(1971) and Anderson (1972) partially testedthis imagery hypothesis. While finding an ap-preciable percentage of false alarms to imagi-nally similar lures, Begg found that adultscould reliably discriminate between para-phrase lures and actual recurrences of sen-tences originally studied. (Paraphrases werepresumed to evoke the same imagery.) Ander-son also found that adults did not confusesentences primarily on the basis of their imag-ery. He argued rather that sentences were

Kosslyn and Bower 31

confused in memory according to similarity oftheir semantic or conceptual content and inde-pendently of the similarity of the imagery theyevoked.

A problem with these earlier studies isthat neither followed procedures to insure thattheir sentence pairs did in fact evoke highlysimilar imagery. To illustrate, one of the sen-tence pairs which Anderson claimed to be"imaginally the same but conceptually differ-ent" differed in providing a judgmental de-scription of an act: either "John signed hissignature on the check," or "John forged hissignature on the check." Adults can rememberlater which of these two sentences thev hadstudied. But this may occur because they usediffering elaborative features of the image to"carry" the otherwise latent judgment. Thus,in contrast to the "straight" check signer, theforger may have a handlebar mustache, a fur-tive manner, and do the signing with slowdeliberation. Of course, trained method actorsare skilled at making manifest, in auxiliary ac-tions visible to an audience, such subtle nu-ances of mood, intentionality, uncertainty, andfeelings. Accordingly, one might "use a goodactor" to portray in internal imagery the epi-sodes or states of affairs depicted by the studysentences. In this case, then, many of Ander-son's or Begg's sentence pairs might not heimaginally identical—which may account forthe degree of discrimination they observed.

To assure high imaginal similarity ofstudy and test sentences, we used pairs of sen-tences which had been prerated by normativeSs as evoking identical or highly similar imag-ery. To contrast the relative importance ofimaginal versus conceptual similarity in recog-nition-memory confusions, we compared false-alarm rates to imaginally similar sentenceswhich either did or did not have the sameconceptual meaning. Our main hypothesis wasthat adults would be able to discriminate be-tween reoccurrence of a studied sentence andpresentation of an imaginally similar lure,while children would not. Moreover, withinthe class of imaginally similar lures, adultsshould make more errors on conceptually samesentences than conceptually different sen-tences, whereas children should false alarmabout equally to these two types of imaginallysimilar lures.

MethodSubjects heard a study list (or lists, in the

case of the children) of sentences to learn. A

Page 3: The Role of Imagery in Sentence Memory: A …web.stanford.edu/~gbower/1974/role_imagery_sentence...The Role of Imagery in Sentence Memory: A Developmental Study. CHILD DEVELOPMENT,

32 Child Development

second test list was immediately presented,and S judged whether the exact verbatim rep-lica of each test sentence was or was not onthe study list.

MaterialsAll the sentences were concrete (see

Paivio 1971) and were composed of seven to11 words. One third of the sentences on thetest list were exactly identical with some coun-terpart in the study list; these sentences willbe referred to as the "sames." Another onethird of the test-list sentences were composedby rearranging words taken from the study-list sentences. These "different" sentences didnot express meanings similar to any of thestudy-list sentences. These should be easily re-jected if S. were storing any meaningful repre-sentation of the study sentences. The finalthird of the test-list sentences were chosen sothat they each evoked a visual image similarto that evoked by a counterpart sentence ofthe study list. Let us describe how these "im-agery" sentence pairs were constructed andselected.

One hundred and fifty pairs of sentenceswere generated and then rated on a seven-point scale for similarity of the evoked imagesby 25 graduate psychology students. The 25pairs of sentences which were rated as havingthe most similar images (see table 1) wereused in the experiment. An informal panel ofjudges initially determined that both sentencesin 10 of the imagery pairs expressed the sameconceptual meaning (e.g., "The flowers in thegreen vase were on the table," and, "The flow-ers on the table were in a green vase"). Italso was determined that the sentences in 10other imagery pairs expressed different mean-ings (e.g., "The lion killed the buffalo becausehe was hungry," and, "The lion killed thebuffalo, so he must have been hungry"). A laterrecheck of these "conceptual similarity" ratingswith 30 Stanford undergraduates producednearly perfect agreement with the informalpanel. Unfortunately, there was one pair forwhich the majority of the raters disagreed withour earlier panel. Using the raters' classifica-tions, then, among our imaginally similar sen-tence pairs we had nine that had the sameconceptual meaning and 11 that had differentmeanings. Pairs where both members ex-pressed the same meaning will henceforth bereferred to as "conceptually similar" (CS),and pairs where the meanings are differentwill be termed "conceptually different" (CD).These distinctions are marked in table 1. Con-

ceptually different pair^ were characterized bysentences that differed in the expression ofcausality, intentionality, time, judgments, andappearance versus reality.

For the adults, the 75 study-list sentenceswere tape recorded in a random order. Test-list sentences that were the same as study-listsentences were recorded in the same relativeposition on both the test list and the studylist. Imagery test-list items were recorded inthe same position as the corresponding study-list sentence. This procedure insures a constantlag between study and test of any target sen-tence. Sentences were recorded every 7 sec,and the test list followed 60 sec after the ter-mination of the study list.

To increase performance of the children,the set of sentences was presented as threeshorter blocks or sublists. Each block com-prised 20 study sentences followed by 20 cor-responding test sentences. These sentencesalso were 7 sec apart, and test sentences oc-curred 60 sec after the study list. Within eachblock approximately one third of the test sen-tences were same, different, and imagery tests.Also within each block approximately half ofthe imagery sentences were conceptually sim-ilar and half were conceptually different. Theorder of the sentences in the study lists wasrandom. As in the adults' list, the serial posi-tion of test-list items mirrored the study-listorder. The same conceptually same and con-ceptually different imagery sentences wereused with both adults and children. All 60 ofthe children's sentences were used in the adultlists; both age groups received the same pro-portion of each type of test sentence.

ProcedureThe original design of this experiment in-

cluded two sets of learning instructions. Withineach age, half of the Ss were to be instructedto learn each sentence by imaging its referentepisode or state of affairs, whereas tbe otherSs were to be instructed to learn by "attendingto the conceptual meaning" of each sentence.Two groups of adults were tested utilizing thisprocedure. However, no memory differenceswhatsoever emerged as a consequence of theseinstructions, and Ss' self-reported use of imag-ery did not correspond well with instructions.Thus, the adult data were pooled over theseinstructional differences, and such differing in-structions were not used with the children.

Adults.—Adults were tested in smallgroups of two to seven. They received printed

Page 4: The Role of Imagery in Sentence Memory: A …web.stanford.edu/~gbower/1974/role_imagery_sentence...The Role of Imagery in Sentence Memory: A Developmental Study. CHILD DEVELOPMENT,

Kosslyn and Bower 33

TABLE 1

CHILDREN'S AND ADULTS' PROPORTION OF "OLD" RESPONSES TO IMAGERY ITEMS

The top sentence of ekch pair was a member of the study list and the bottom was on the test list. To the right ofeach pair is the conceptual similarity designation (derived from adult ratings): CS, indicating conceptuallysimilar, or CD, indicating conceptually different. To the right of the conceptual similarity designation is theproportion of false alarm ("same," "old") judgments given to that pair by children (left number) and adults(right number).

1. a) Henry scraped deep into the ground with his hoe. /pc . ^^ nr.\b) Henry gouged the ground with his hoe. '

2. c) The ball rested on the box in front of the tree.b) The ball in front of the tree rested on the box.

3. a) The fly flew over the dog that was under the blanket.b) The fly flew over the blanket that was over the dog.

The dog chased after the frisbee into the bushes. ,pc. nnThe dog ran into the bushes after the frisbee. ^ '

The ambulance went down the street with its lights flashing. /pg. -^ .--.Using its flashing lights, the ambulance went down the street. '

The flowers in the green vase were on the table.The flowers on the table were in a green vase.

4.

S.

6.

a)b)

a)b)

a)b)

i n )7. a) Homer Smith was much taller than fat old Stephen.b) Fat old Steplien was much shorter than Homer Smith.

8. c) A falling raindrop was the reason Sally's nose was wet. /pc .b) Sally's nose was wet because a raindrop fell on it. '

9. a) The glass of Coca-Cola was half full. //-^c. 5A 45)b) The glass of Coca-Cola was half empty. ' '

10. a) The crayon rested in the boy's hand as he drew. (CD- 50 35)b) The crayon rested between the boy's fingers as he drew. ' '

11. a) I have washed my new car twice today. irx)- on 40)b) I washed my new car twice yesterday. ' '

12. a) Hank dug the hole furiously a day early. (Qj^- 40 10)b) Hank dug the hole furiously a day late. ' '

13. a) George picked up the pen to write on the paper. (CD- 20 0)b) George picked up the pen to doodle on the paper. ' '

14. c) Jimmy got his present because Santa knew he deserved it. (CD- 90 55)b) Jimmy got his present because Santa thought he deserved it. ^ ' '

15. a) The lion killed the buffalo because he was hungry.b) The lion killed the buffalo, so he must have been hungry.

16. a) A large brown cow was falling asleep on the grass. (CD- 60 55)b) A large brown cow was laying down on the grass. ' '

17. a) My father's friend Ed always looked very happy.b) My father's friend Ed always was very happy.

18. a) The big white sack looked exactly like a pillow. (CD- 70 15)b) The big white sack was really a pillow. ' '

19. a) Pat looked like she thought she knew the answer. (CD- 60 65)b) Pat looked like she really knew the answer. K . , j

20. a) The peanut butter and jelly sandwiches were made a few hours ago. (CD- 80 45)b) The peanut butter and jelly sandwiches were made an hour ago. ' '

instructions to attempt to remember the list of initiation of the test list Ss turned to a pagesentences that they were about to hear. The on which were listed numbers 1-75. Next tostudy list was then presented. During the min- each number were two brackets. One columnute between the end of the study list and of brackets was labeled "old" at the top, and

Page 5: The Role of Imagery in Sentence Memory: A …web.stanford.edu/~gbower/1974/role_imagery_sentence...The Role of Imagery in Sentence Memory: A Developmental Study. CHILD DEVELOPMENT,

34 Child Development

the other was labeled "new." The Ss were in-structed to listen carefully to each sentence onthe test list and to indicate whether theythought it had been on the study list by check-ing the bracket under "old" next to its num-ber. If they thought it had not been on thestudy list, Ss were to check the bracket under"new." It was emphasized that in order for asentence to be considered "old" it must matcha study-list sentence verbatim ("word forword").

Children.—Children were tested individ-ually. Before beginning the actual experimentit was necessary to assure that the child under-stood what was meant by "same" and "differ-ent" verbatim sentence.s. The child was testedfor his concept of same and different with apair of sentences differing by active-passivevoice ("The fox chased the rabbit around thebam" vs. "The rabbit was chased by the foxaround the bam"), with a pair of sentencesdiffering by a quantifier ("The boy ate all ofthe ice cream" vs. "The boy ate some of theice cream"), and with a pair of identical sen-tences ("They bought a lot of new furniturefor their house"). If the child expressed anyuncertainty, he was trained until he couldcorrectly classify sentences as verbatim sameor verbatim different. It was emphasized thatfor two sentences to be called the same theyhad to be exactly the same, word for word.After S's understanding of "same" and "differ-ent" was established the experiment properbegan.

Each S was instructed to listen carefullyto each sentence. (The study block was notpresented until we were sure he was attend-ing.) In the minute following presentation ofeach block of study items S was told that atest list of sentences would be presented, andhe was to say "same" if a sentence was exactlythe same (word for word) as one on the firstlist and "different" if it was different. Subjectsexperienced no difficulty in making these judg-ments. For most Ss, all three study-list blockswere presented during the same session. ThreeSs completed only two blocks during the firstsession, but finished the third several dayslater. Five Ss wearied and failed to completeall three blocks. Despite these losses of obser-vations, the sublist orders were so rotated thatthe total number of observations per sentencetype was equal, and all sentences were pre-sented an equal number of times.

SubjectsThe subjects were seven boys and five

girls ages 5-2 to 5-7 (median age = 5-4)from Stanford's Bing Nursery School popula-tion and 20 adults (12 males and eight fe-males) from an introductory psychology classat Stanford University.

Results

Since we wish to generalize our resultsbeyond the specific items used in this experi-ment, as well as the particular Ss, it is neces-sary to consider both items and Ss as randomeffects (see Clark 1973). Thus, for each com-parison t tests were performed twice, oncepooling over items exemplifying a given con-dition for each S (the standard method) anda second time pooling data from different Ssand analyzing in terms of the actual itemswithin each condition. Unless specifically noted,reported significance levels are representativeof both analyses.

The primary data are the proportions oftimes a test sentence of a given type is judgedto be a verbatim replica of a studied sentence(old). These proportions are shown in figure1 for the two populations and three categoriesof test sentences. Each proportion is based on500 observations for the adults and 200 forthe children. The adults clearly show veryreliable discrimination among all three typesof test sentences; separate t tests betweenadjacent conditions are all highly significantstatistically (same vs. imagery, p < .001;imagery vs. different, p < .001). The highfalse-alarm rate of adults to imaginally similarsentences is important in conjunction with theotherwise very high degree of learning (dis-crimination) shown for same versus differentsentences.

The children show a different pattern ofrecognition confusions. First, their overall dis-crimination is poorer than that of the adults;the children have a lower hit rate on sameitems and a higher false-alarm rate on differentitems. These main effects in themselves arenot of much interest. What is of interest isthat the children do not discriminate at all(in memory) between same and imagery sen-tences; observed proportions of "acceptances"were 66% and 61%, respectively, for the tyvosentence types. Although these proportions donot differ reliably from one another, they bothexceed the false-alarm rate (18%) to different

Page 6: The Role of Imagery in Sentence Memory: A …web.stanford.edu/~gbower/1974/role_imagery_sentence...The Role of Imagery in Sentence Memory: A Developmental Study. CHILD DEVELOPMENT,

Kosslyn and Bower 35

CObJCO

tnbJ

b.

o

O

OCso

Somt Imogtry Difftrwit

TYPE OF TEST SENTENCEFIG. 1.—Proportion of three kinds of test sentences accepted as "old" by children and adults

items {p < .001 in both comparisons). Sowhile the children clearly discriminate be-tween "meanings" which were or were notpresented, they do not discriminate in mem-ory between sentences which evoke highlysimilar imagery. The difference in this differ-entiation between children and adults wasassessed by two t tests, one across Ss and oneover times. For the first test, each S receiveda score reflecting his difference in old propor-tions to same versus imagery test sentences.These discrimination scores proved to be re-liably greater for adults than for children,f (30) = 2.90, p < .01. The items analysis was

performed by subtracting the mean proportionof old responses to the imagery items fromthe proportion of old responses to same items.Once again, these discrimination scores weregreater for adults than for children, t{43) =4.47, p < .001.

Our second interest concerned the dis-crimination in memory between imaginallysimilar sentences which are either CD or CS(see table 1). As might be anticipated fromthe results above, children and adults alsodiffered in this respect. The children fakealarmed more to CD items (65%) than to

Page 7: The Role of Imagery in Sentence Memory: A …web.stanford.edu/~gbower/1974/role_imagery_sentence...The Role of Imagery in Sentence Memory: A Developmental Study. CHILD DEVELOPMENT,

36 Child Development

CS items (56%), whereas adults showed thereverse trend, false alarming more to CS items(47%) than to CD items (36%). The adultsshow a drop in false alarming between CS andCD items, whereas children show a slight trendin the opposite direction. This interaction be-tween item type and age group was tested intwo ways. First, pooling across items, thedirected difference in false-alarm proportionsto CS and CD items was reliably greater foradults than for children, t{30) = 3.11, p <.01. Second, pooling across Ss, each item wasscored according to the difference in false-alarm rates to it for adults and children. Thesedifference scores were reliably greater for the11 CD items than for the nine CS items, t{l8)= 2.13, p < .05. The differences in falsealarming to CS and CD items can also betested within each age group separately. Suchtests reveal no significant differences for chil-dren; a significant difference for adults whenpooling over items, f(19) = 2.86, p < .01,but not for adults when pooling over subjectsand treating items as the random effect, i(18)= 1.37, p < .1. Inspection of false-alarmrates in table 1 reveals item-specific variabilityor inhomogeneity within the sets of CS andCD items; beyond simple "sampling error,"this variance presumably reflects the influenceon sentence memory of such potent factors asword salience, presuppositional focus, etc. Al-though adults more than children rememberand discriminate among sentences according totheir conceptual meaning, table 1 suggeststhat there are further subdivisions within theoverall CS and CD categories which may re-late to recognition false alarms in a more uni-form and lawful way.

Comprehension Control GroupThe observed age differences in memory

confusions may be thought of in two ways.Consider two sentences, A and B, which areimaginally similar and which adults judge tobe conceptually different. Children may con-fuse A and B in recognition memory eitherbecause (a) they simply do not understandor even detect the difference in meaning be-tween the two sentences even under optimalconditions of comparison, or {b) althoughable to comprehend the difference, childrentend to forget more readily than adults thattype of information that distinguishes A fromB. The issue can be settled readily by testingchildren for comprehension of the conceptualdistinctions in our CS and CD sentence pairs.To this end, an additional 10 children from

Bing Nursery School (six girls and four boys,ages 5-0 to 5-7, median 5-5) were tested indi-vidually on two types of comprehension judg-ments. In each task Ss listened to pairs oftape-recorded sentences, the first sentence ofeach pair read in a male voice, the second ina female voice, and then immediately comparedthe two sentences according to a specifiedcriterion. In the first task, they judged whetherthe sentence uttered by the female was "talk-ing about the exact same thing" as that utteredby the male. They were to say "different" incase the second sentence of the pair meantanything different from the first. The tape forthis first task contained our 20 sentence pairsrated earlier as being highly similar in theimagery they evoked (so-called imagery pairs);these were mixed in with 10 pairs sampledfrom the same sentence pairs used for thechildren in the earlier memory experiment. Asexpected, the children correctly categorized98% of the same sentence pairs as identical,whereas they categorized 5% of the imagerypairs as expressing exactly the same meaning.So on an immediate test, they can discriminatebetween same pairs versus imagery pairs.

Now, this discrimination among sentencesof an imagery pair may reflect little more thanchildren's ability to detect changes in wordingof the sentences. To help decide this issue,they performed a second comprehension test.The Ss heard the 20 imagery sentence pairsagain and this time were asked to decidewhether the two sentences were "talking aboutthe same thing or idea but with differentwords." The distinction was illustrated withseveral examples. As one instance, the targetsentence, "I ran across the room and touchedthe wall," was said to have the same meaningas "I touched the wall after I ran across theroom," whereas it has a different meaningfrom "I ran across the room and touched thefloor." A second illustration was the target,"The floor was dirty because paper and asheswere all over it," which was said to mean thesame as "Paper and ashes all over the floormade it dirty" but to have a different meaningfrom "Paper and tin cans all over the floormade it dirty." Following these instructions,the 20 pairs of imagery sentences were readtwice for each S. If S's second judgment forany pair differed from his first, the pair wasread again and S was queried regarding hiscertainty. Three responses were possible:"same," "different," or "not sure." In orderfor a response to be counted as "same" or

Page 8: The Role of Imagery in Sentence Memory: A …web.stanford.edu/~gbower/1974/role_imagery_sentence...The Role of Imagery in Sentence Memory: A Developmental Study. CHILD DEVELOPMENT,

"different," S either had to justify his decisionor resist attempts to get him to change it.The scores to be reported, then, are only forcertain and/or rationalized decisions; exclud-ing uncertain judgments causes change per-formance to be less than 50% on the two-alternative test.

To briefly summarize the results of this"gist judgment" task, the children's judgmentsagreed with the adults' 69% of the time forconceptually different pairs and agreed 79%for conceptually same pairs. That is, our chil-dren noted similarities and differences amongmeanings of sentence pairs much as our adultraters did. The difference between CD and CSpairs in amount of child-adult agreement wasnot significant, t{18) = 1.10, p < .10. Onemight expect that children's tendency to saythat sentences A and B "mean the same thing"would correlate with how often they falsealarm by saying "same" to B when A had beenpresented in the study list. But this Pearsoncorrelation (over all 20 items) was only + .13for the children, suggesting that memory con-fusions were being largely controlled by theimaginal similarity of the test sentence to thestudy sentence. This low correlation was indi-cated also by the earlier facts that childrenjudge "same meaning" much more often forCS pairs than for CD pairs, but yet their falsealarms in memory to these two classes of itemsare somewhat in the opposite order.

Discussion

The findings of Begg (1971) and Ander-son (1972) with adults have been replicated.Moreover, memory confusions for adults occurmore often to conceptually similar sentencesthan to conceptually different sentences, evenwhen both lures evoke the same image as astudy sentence. This implies that a sentence'sinternal representation which endures in anadult's memory probably contains conceptualdistinctions over and above those representedin the sensory imagery the sentence evokes.The fact that adults do discriminate betweensame (78%) and conceptually similar sen-tences (47%) may be explained in terms oftheir auxiliary encoding of and memory forsurface-related features of the study sentences(e.g., the focus or presuppositions of the sur-face assertions).

In full contrast to these results withadults, the children's equal acceptance of same

Kosslyn and Bower 37

sentences (66%) and conceptually different(65%) sentences suggests that the conceptualdimensions or features along which these sen-tence pairs differed are either not encodedfrom the original study sentence or are soonlost from the memory representation (or can-not be regenerated out of it). Our comprehen-sion data suggest that the children's difficultywas not in initially understanding the concep-tual distinctions in the imaginally similar sen-tences; rather, the difficulty lay in the children'srapid forgetting of these conceptual distinc-tions.

Our present interpretation of these datais somewhat as follows: A sentence is initiallyencoded along several dimensions including itssurface phonological form, its semantic (con-ceptual) relations, and the referential imageryit arouses. For simplicity, these may be thoughtof as three distinct memory codes that areclosely tied together. In our comprehensiontasks involving only very-short-term memory,the codes from the two sentences of a pairwere readily available, and the two sentencescould be discriminated at that time on thebasis of their surface phonology, their con-ceptual semantics, or their aroused imagery,depending on S's judgment task. But, ofcourse, forgetting—or loss of code specificity—occurs with the passage of time and/orinterpolated events. The data suggest that theforgetting rate for the semantic-conceptualcode is much faster for children than for adults.Since the children later have available inmemory only the "imagery codes" to comparewith the test sentence, they make many falsealarms to imaginally similar lures, and these arenot distinguished (as they are for adults) be-tween conceptually similar and conceptuallydifferent lures.

A final comment is that the conceptualdistinctions that the child forgets would ap-pear to be those which he learns late and findssomewhat hard to comprehend—dimensionssuch as causal attribution, temporal prece-dence, intentionality, appearance versus real-ity, knowledge versus belief, etc. Many otherdevelopmental psychologists have remarkedupon the slow pace of cognitive developmentregarding these conceptual, and relatively non-sensory, distinctions. Thus, the developmentalpattems of memory confusion errors foundhere seem quite consistent with this correla-tive information regarding cognitive develop-ment.

Page 9: The Role of Imagery in Sentence Memory: A …web.stanford.edu/~gbower/1974/role_imagery_sentence...The Role of Imagery in Sentence Memory: A Developmental Study. CHILD DEVELOPMENT,

38 Child Development

References

Anderson, J. R. Recognition confusions in sentencememory. Unpublished manuscript, StanfordUniversity, 1972.

Begg, I. Recognition memory for sentence mean-ing and wording. Journal of Verbal Learningand Verbal Behavior, 1971, 10, 176-181.

Bniner, J. S.; Olver, R. O.; & Greenfield, P. M.Studies in cognitive growth. New York:Wiley, 1966.

Church, J. Language and the discovery of reality.New York: Random House, 1961.

Clark, H. H. The language-as-fixed-effect fallacy:a critique of language statistics in psycho-linguistic research. Journal of Verbal Leamingand Verbal Behavior, 1973, 12, 335-359.

Fillenbaum, S. On the use of memorial techniquesto assess syntactic structures. PsychologicalBulletin, 1970, 73, 231-237.

London, P., & Robinson, J. P. Imagination inlearning and retention. Child Development,1968, 39, 803-815.

London, P., & Robinson, J. P. Labeling and imag-ining as aids to memory. Child Development,1971, 42, 641-644.

Paivio, A. Imagery and verbal processes. NewYork: Holt, Rinehart & Winston, 1971.

Piaget, J., & Inhelder, B. Mental imagery in thechild. London: Routledge & Kegan Paul,1971.

Rohwer, W. D. Images and pictures in children'sleaming: research results and educationalimplications. Psychological Bulletin, 1970, 73(6), 393-403.

Schank, R. C. Conceptual dependency: a theoryof natural language understanding. CognitivePsychology, 1972, 3(4), 552-632.

Wolff, P., & Levin, J. R. The role of overt activityin children's imagery production. Child De-velopment, 1972, 43, 537-547.

Page 10: The Role of Imagery in Sentence Memory: A …web.stanford.edu/~gbower/1974/role_imagery_sentence...The Role of Imagery in Sentence Memory: A Developmental Study. CHILD DEVELOPMENT,

This document is a scanned copy of a printed document. No warranty is given about the accuracy of the copy.

Users should refer to the original published version of the material.