ED- 113.702

AUTHOR

INSTIT.UTIN

SPONS AGENCY

PUB DATE --NOTE

EDRS PRICEDESCRIPTORS

,

..."" .

'DOCUMENT RESUME

CS 002. 208;

Gentner, Donald. R.The Steuctur0 andi,Recall of garrtive Piose.California Univ., La Jolla. Center. for HumanInformation Processing. . .Office of. Naval Research, Washington, D.C., Personneland -Training ResearCh Pro ams Office.Oct 75 4

\a. s

21p, . (.

r'MF-$0.76 Hc,-$1'.58 Plus , 4 ,

Aural 'Learning;. *Cognitive Processes; *Grammar;.

, \.* Higher Education; Lis,teningeCoarprehension,; "*Memtry;. *Narration; Oral Communication; .*Recall)(Bsycholoqical) ; Retention Audi ; SentenceSfrubture; Tape Recordings ... . .,.,_ . ,0

A'B.STR1CT1' * 1 4 .

. , . . This study examines recall. of narrative proSe_forevidence of unde-rlying structures such es .the grammar !sed 1n the '

...'story's sentence` structiire. S'ubjectS listeed to repeAed ..

'.. presentations of a tKpe recofding ,of two pages from a history, bbok,with, verbals collected 'after' each presentation. The subjects used in'hi. s e'xperiment were 13 undergraduate,students at. ,the ,University of

. , californf San Diego, who volaunteefed foi the experiment in returnfbr class credit ,,or paymentSV.The results ,show that serial structuresand stofy gramliik 'structure are iniportant. factors in the 'memory andrecall of-,'narrive passa.ges. (Author/RB) 4

W.

..: / t.

. .

****1!**********************,**41;:******************* ******e********Ic****r, * Documents acgocired by :BRIG (include many informal unpublished A's

* materials not available fr/m other sources. ERIC makes every effort *.*.t.c:S obtain) the best bopy, available. Nevertheles, items of marginal ** treprodtleibility axe often encountere,d And this ;affects the quality *

a

4 4C of the racrofiche and 'hardcopy teproduc4on -ERIC makes available ** via the ERIC,Document Reproduction service (EbRS)'.1 EDRS is not '* ,

.. , ,* responsible,-for the quality' of the original document: Reproductions ** supplied by EDRS are the best that tan be made from the original,. ************************°*******4******.************i*********;***********44,../ .

. .4 , i ..

: ,

I"

G

I

kt.

V

US DEPARTMENT OF NEALTH,EDUCATION &WELFARENATIONAL INSTITUTE OF

(EDUCATIONTHIS D CL/MENT HAS BEEN .REPROOUCE0 XACTLY AS RECEIRFO FROMTHE PER, NOWOROANQATIONOR.GINATA NG IT 'Om% OF VIEW OR OPiNIONSSTATED D NOT NECESSARILY REPitESENT OFFI 'JAL NATIONAL ;NSTITurt OFEOUCAT, POSITION OR, POLICY

The Structure and Recall of Narrative Prose

Donald. R. Gentner

Y

A

Contract Number: N00014-7.5-C-0757

ARPA Order Number: 2284 -4,

Task Adthority Identification Number: NI 1

Program Code Number: 5W10

EffectiVe Date of Contrictr.: 1 FebriSaryl 1975

Contract 'Expiration Date: 31 Januiry, 1976.

Amount of Contract: $106,509 '

360

Organization: University of California, San Qiego- ,

Principal Inveftigator: Donald Ar Norman )[Phone (714) 452-2947].

'

Scientific Oificer:Dr. Marshall Farr -, Director

,)

Personnel and Training Research Programs,'Office of Naval Research

4

-;c :..

The Jews andconclusions contained iri this document are those bf thea'ufh r and should not be interpretets.necessarily representing theofficial policies, either'expressed..pr implieaf of the, Advanced ResearchPr:ojectsAgency, the Office of Naval Research, or the U.S Government.

This resear ch was supported by the:Advanced-Resparch Wojects Agency andthe Office of-Naval Research of the Departmfft of Defense and was monit050by ONR under Contract No..N00014'-75-C-0757.%

ipproved for public release; distribution unlimftpd.

2 "-.,

t

UNCLASSIIEW 1

Ee FoTy CLAiSIFICATICfN OF THIS PAGE (khan Data Entered,

-REPORT DOCUMENTATION PAGE' N

READ INSTRUCTIONS_Lo BEFORE COMPLETING FORM

I REPORT NUMBER 2. GOVT ACCESSION NO...

13 hEciPIENTS cA,TALoa NUMBER.; --.

. ..

4 TiTLE :and Subtitle) -,The Structure and Rdcall of Narrative Prose .

...1

' . .

5 TYPE OF REPORT 8 PERIOD COVE.... e

,...

Technical ReportA ,;.. A.

6 PERFORMING ORG. REPORT NUMBER

8 COIreillZci oil 615AN T NUMBER(.).,

NO0014-75-C-0757 .. -. ;

.

I A') T,4C/R. ,, i

Donald R. Gentner .

.

-.'

9 PERFORMING ORGANIZATION NAME AND ADDRESS

Dr. Donald A. Norman . /..' . .

Center for Human Informatiop Processing°University of California, San Diego-

-.1 -

410 PROGRAM ELEMENT.PROJECT TASKAREA 8 WORK UNIT NUMBERS *

.

.

NB 46-360-1

.

t! k,ONT ROLLING OFFICE NAME Ati0 ADDRESS,

Personnel and Training Research Programs,

Office of Naval Research (Code 45S) ..

., -...

krlingfq0VA 22217 e 0

12. REPORT DATE

Octoker, .

, 1975

13 NUMBER OF PAGES ''

13-

14 MON! TONING'AGENCY NAME & AODRESS(ti dillerent from ControOng °Dice). ......L, ts :'1..Zt

..

%J..

15 SECURITY CLASS. (ol thie.report)

.W .....

Unclassified15., DECLASSIFICATION/DOWNGRADING

SCHEOUIE -.

16 I.---SISTR3y----a'ArIONSTATEMENT (ol this Amport) ,J (

Approved r public release; distribution''. . './

:liatedinsi?0,.20,ifdiiforetufromw0

--i

Ai. .

.._

unlimined. .

I .

.6

port)

..

0

M Of$TRiBUTION STATEMENT (ol the abstract

E-

18 ' SUPPLEMENTAR NOTES 1

' ..,.

aI

. , I

i .ti.

19. KEY WORDS (continua our revers. ide it ncmary and identify by block number) ' .

. .

Memory; cognitive psychology, recall, prose recall, learning, text°recall,kunderstanding prose, prose structure. /. : * .

. .. .$ .

I ftL

20 ArisTRACT (Continua on roveis side it ncsary and identity by block numbs)This report .describes,a'study of tht recall (A narrative prose: 4' Subjectsj

listened to repeated presentations of a tape recording of two pages'frum ahistory book, with verbal recalls collected, after each presentation. The .

elements of the passage were organized according to a serial,structure based' onordor'in the passage and a story_grammar structure based on causal relations.WhilV the. serial structure at first influenced which elements of the passagewere remembered, as 'the subjects remembered more of the passage, the story '

gramthar structure became the dominahi influence over the elemehts remaibered,oh..._dillachluent recallZ,-

_

!JAN 1473 EDITION OF I NOV (5 IS OBSOLETESiN 0102-1-F;04 6601

tiSECUNITY-CLASSIFICATION'OF THIS PAGE (When Deta.lntiiked).

lv

V

k

THE STR CTUR.E,AND RECALL OF NARRATIVE PROSE

Donald R. G6ntner. .

University of California: San Diego

La :Jolla, California 92093

. . `..t

{Learning a co mpleX story often equires several readings. Each re-..

:4'

ireading of the passage adds to Ale overall understanding, with the new..

klowledgePicked up at a reading being added co the general semantic- .

structure of the previously acquired knowledge.. Recent developments in

he study of semantic memory, iNormgn, Rumelhart, and the LNR reearchA

_gAuP, 1975;' Rum -Marc, 1975) suggLt ways ofinlestigatini.how a per-,

. .

-

sonls understanding of a prose passage develop with Successive expo-i .

(

sures to the passage. In this

\\)

study, subjects listened several times

to a tape recording of "prose passage, and their developing knowledge

was followed by collecting recall4 after each presentation of the prose

ipasIage.

4 tIn the pasd\few ears, a'number of studies have been published

on the organization awl recall of prose. Generally t y have beena 1 J 4N.,7 .0

concerned, withidescriptive prose, and the structures proposed for,

1 t f (, t the organization cif prose have been Simple hierarchies. ;Meyer ant h..

%,44;1,, .

* .

. 1

McConkie (1973, see also.Merer, 1975) 'analyzed

t

destriplive-Passages\,..

? . )...,' ,

.s...aboutnuclear reactors and parakeets into apierarchY of nidea'units".

P

bi t and found that iSeg units higher in the hierarchy were more frequently

..

.

I . ,

reca14ed than idea units lower in the hierarchir..Atederiksen (1972

7

4

1975) l'ooked at tile ` recall of set rentiOns id prose and reportedI

,t

that most errors were relate to the acquisition process, rather thandipw.....J." 1A .

Athp Memory retrieval proceg's.11,Crothers (1972) haedeveloped.ra method. .

..-, .. ..,..1..

I .. .-

.for characterizing stimuli and scoring responses fot recall of descriptive

prose involving set relations.

. /

A I

/

a

.2

4

Monk and Kintsch-(1974) measured reaction:times_ to answer, true-.

false quespions,or recognize -sentences plter reading-par raphs of.

varying length. They found that tht time to correctly respond "true"

or recognize an /old sentence increased with the length of the papagraph,

while tiee time to, correctly respond "false" or reject a new sentence

%was independent of the length of the pardtrlph.-

Rdmelhart (1975) has proposed a grammar Ott understanding narfativer

. 41..prose and representing the in.formation.in a neitwok structure. In the

Rumelhaft story grammar, storieare decomposed into smaller 'units,1'

,^\slich a settings, episodes, actions, reactions, events, and goals.

These smaller units, which.msy be decomposed further, are interrelated

. ,

with caval predicates,Such as IRITIATIVE..NOTIVATE and ALLOW. For

S

example, an epis9de costs of an'event which INITIATES a reaction.. ,

....f

C , .

(A portion of the stru ture which is prbduced when the story,grammar. .

text is Aowntn Figure 1.) The Obry igimmar. R

4A . .

l' ,I...s applibd to a histar-t. ,

....

,. . , , , ....,'

,

. frovides a useful tool for investigating the developing knowledge.

.-.., ...

-.

p ,%., ,-. c''';'

structures of the learner. Thig,paper examines reaf f'na,r-rati.ve

.,.." 'Y .. -

, prose for evidence of underlying sbructuzes suchdas th le proposedr . -: 1

.I . _

by the story grammar., 1 e s4

Tha text led in this stud was a passage about two pages in

4 4ngth (around 925 words)sfrop Morison's The OxfA Hisly oi the,/

tr.

s ,

f

American-People '(1965, pp. 638 - 640).. A text from a hisppry.,

.

4 . .

matbook wasZchosen becduse of its /similarity to materials used in (real

:.t e,- . .

educational situations andt becau e1,it could be'analyzed with the

story itammar which had been deviPloped to describe naratives ,of human

1.4.:

..,4

. 4.

It

t

actions. A portion of the text is quoted iwthe result section' of

3

this paper. The text is a self-contained description OF General Grant's

early forays in the WeSt: The.capture of Forts Henry and.Donelson and

the battle of Shiloh.

4 t, /4'The experiment consistedlofpjaying a tape recording of the text

four timesand co Meeting verbal recalls after each presentatior of

r the text The subjecn were first told that they Would be lisening1. -6

to a tape recorcting of approximately'two pages from a history lbok,

that the.recording would be Played tour times, and that after the ,

recording was finished each time they would be asked to tell all they,//

4couli renaer. The tape recording, bichlasted about six minutes,a

r

was then played for the subject/and' A it finished, the subject was .

asked to tell, everything he could

jec,t

.. c) 1, a 1

0i t

1 I

.5

0 I1

.wasetlayed Again, sand the'subject-was then asked.to tell .a.11 that he, ,

'. .

... . ,

% e

i:ould remember, including things that he had meniOned earlier;N This

,

i A

sequence wad repeated for I,tobal of four teials, where a tri,arconsists,

* ,... , 3

of a'presentation of the'tape recorded passage and the subsequent recall.

remembet from the passage. (fhe sub -

ted that he could'not remember anything else, t-h passage

. .

The subjects used in this experiment were JA undefgradUate students1

Ali .t

.at the University.of California; San Diego who had volunteered for the

...

. . .('t:

. '

. -.

..-experiment in return fbr'clasg credit or payment. No Subject had any ..

< .

4t '. 4'

unusual knowledge or interestAin.the American CivilbMar... e,. ' , e

4

<II a

,

RFSVETS

ry

4,4

Method of Analysis.

In order to analyze the recalls, the Mori, ton text, cohsiting oC

47-sentences, was divided-into 143 "facts.."- These "facts".correspond

to the units described by the story grammar such as setings, actions

-consequences, goals, and plan S. Although the method of. divtding the

text into 'facts wasinot completely systematic, in4teneral a fact

corresponds Eo a phrase or simple sentprice in th original text.

1-

Many o f the,facts could be(ecomposed into simpler propositions,

but.that was not necessary for this analysis. Examples of several

factsstan be seen-6s the numberedstatemehts in Figure 1..

..1 '

The Story Grammar Structure i . .1

V

s

' . . . text .

'The facts*derived from the Morison text were structured accord-. ;. , i

_

ing to d)e,st,ory grammar. A small` poiti on If the story .grammar1 ... tor .. , .

....,,

1. ,structure is. shown Iningurq 1. 'This structure -represents some o

o1.

: -,the facts deriVed frbm the fotlowing section of text':

.

., .

- . .7%. e

',./ ..,In the.suMmer of,1861 the, Confederates began. to throw 4P

O

l% * 1

I ..kea T t h war k s at various' points along .C4. Missiissip0. iaherp.,

- .:.i .

... /.st, the old Spanish...forts used 0 choke `down %rade.' ,In . ..

r. . . . , , : .

(11 . :t 1

4oterto force'a paspagp past them J. 13. Eads, an,:engineer,therm,

. ,.. 4 e / 0' . . .

1 .,t r .

of St. Louis, constructed a.fleet of.rivey gunbodts, each ..

. ) ... , .'

o . +

'with a Ortiaiy.arMored asemate shaped like a Mansard. . . .

A- .4 ' * tt- . I,

roof, and a flat-bottome 'hull... . ( . 0 4,

). ,

Less, than 50 milesup the Ohio from Cairo thk"Tennessee -,f

.1

I

and Cumberland : rivers offered pArallel routes into},a.

Tennessee, Alabama, Ind Mississipni.Grant observed

that Forts Heprt, and Dondlson,!the Confederate earthworks

ti

....II

J. B

. Ear

lsw

as e

ngin

eer

of S

t.L

ouis

7. C

ofile

dera

tes

bega

n to

thro

wup

ear

thw

orks

alon

g M

issi

ssip

pi

Dat

e.._

.,--

..

6. B

egan

insu

mm

er o

f 186

1

ilkoL

ight

18. G

rant

dec

ided

to c

aptu

re F

orts

Hen

rya

,Don

elso

m

fl. N

orth

wa

to g

et p

ast

eart

hwor

ks

fhou

9ht

132.

Nor

th,

deci

ded

to'

use

gunb

oats

10. E

a.ds

-con

strq

cted

Fle

et o

fgu

nboa

t's

Exp

bnsi

on

Exp

ansi

on

8."E

arth

wor

ksw

ere,

loca

ted

whe

re o

ld-

.o

Spa

nish

fort

sch

oked

dow

nriv

ertr

ade

Eve

nt

0 bj

Situ

atio

ns

Act

iCan

s'rn

.

-21.

Gra

nt fi

rst

cons

ulte

d w

ith'

Com

mod

ore

And

rew

H. F

oote

Obj

)

4

22. F

oote

cdf

rem

ande

dth

e gu

nboa

t flp

tilla

Figure 1.

A.pOiton of the story grammar structure for the

passage.

12. G

unbo

ats

had

parr

rally

arm

ored

cas

emat

elik

e M

ansa

rd r

oof

Obj

.

Gun

boat

s ha

d.fl

citb

otto

med

hul

l'

4. 0

I

O

r

A

' which closed thes e rivers, were the win keys to the rebel1

. -

West, Their capturewould open a navigable waterway into the

enemy's.cen ter and drdve in his flanks. 0.n 30 January 18624

Grant: after consulting with Commodore Andrew H. Foote,

commanding the gunboat flotilla, obtained*Halleck's.

reluctant consent)

to try, and was furnished with the

5

necessary transports and gunboats. (Morison, 1965, 'page 639.),

The e6mplete story grammar structure for the passage contemns1 .

facts interconnected with about 200 predica tes or relations: 131 ofI .

these facts are explicitly contained within'ttle text, and 12e"impl2cit

-facts, " while not explicitly mentioned in the passage, were redulred

''by the story gr ammar. For example,,the story grammar requires that

all activities, such as "Fads Constructed a fleet of gunboats," be

.

,motiyated by plans. I therefore added the plan,'"North decid ed to -use

.gunboats,"'es,an 4licit fact even though it was'not explicitly

mentioned in the Motiscin text (implicit facts are shown in parentheses

in.

Figure 1). For any structure such as the story greilimer structure,

we can count the facts which are neighbors di any given fact. A

neighboring fact is one which is connectfdto the%iven factlby a

0,

1 . ' - .- , :

sipgle relation or predicate. For example in the structure shown in

d

Figurp,l, fact -7 has three neigh14,(rs (facts 6, 8, and 11), while fa,ct 8

4has onlyjone neigh67O'r (fact 7).

r.

,

.

r

. 0

---,

The Serial Structure/

1. ,

-.

,.

..

. .

. .i

The 143 facts were also structured in a simple linear chain according

to the serial orderrder in which the facts occurred in the original passage..

.

This is referred to as the serial structure. Except for the initial. ,

and final facts, all facts in the.serial structure have two neighborp.

Noe that the serial sltructure contains the same, implicit and explicit

facet included in t'he story grammar analysis. The implicit facts were

1 added in a positiop to produce the most accerfable.narrative.

Scoring of Recalls I

For each .recall of each subj , I noted whether each fact was

absent, part- correct, or correct. n the,recall. A fact was scored as

4

correct if the recall contained subs antially correct paraphrase of

the fact. A. fact was scored as part- rect if ,some of the material

in the fact was mentioned in the recall (i.e., the recall-might contain=

a partially c orrect or incorrect sta tement of the faCb. If the

information in the fact was not Mentioned at all, it was-Scoredlas.

absent

As an example consider the fa-t: "The'Confederates began tp. ,

4 . 4

throw up . earthworks along the Mispissippie" Theifact was scored as.

correct in this recalli "...it talked about hew the Confederate

forces had taken over most of the Mississippi y sort of throwing up f

earthworks...." -The fact was scored as part-c rrect in this recall:

".!. and the Confederates were starting to then alOng the Mississippi.,.."

P

I thus obtained theAollpowing,data for analysis:

a) For each subject on each of four trials, each,fact scored.

aq absent, part-correct, or correct.

.b) A 'story grammar structure interconnecting the 143 facts.

' "c) A serial structure interconnectigg' the 143 facts.,

Recalls

The.average performance f subjbcts shows\a steady, almost constant

1improvement un successivetri,1 (Figure 2). the variation among

\

,z:.

subjects, however, was surprisdngly large. 'Combining the four trialS,\ ,

theWorst subject got 14 facts \correct and 56 laces part-correct,

.

a'while'the best subject got 25.1 facts. correct and 90 facts part-correct.?

., . 1 ,

A fact was gived a score of two if it was.correct, one if it was

part-correbt, and zero otherwise. Thus the range of scares (combining

ally trials) was from 84 to 592.\

To get 'a detailedlcook at how 'neighbors, in the story grammar

structure affdct the.reall of facts, I 'looked at each of the first

three trials of\a subject, classified each fact as absent, part-correct,. -

or correct, and then noted the score for that same fact on the subject -'s

next trial. Al.]. 'scores were combined to yield a'scol -e on trial 114-1

for facts which were absent, part-coyrect, or correct On trial n,

averaged over values of n.from-one to three. 4

SS

0.5

0.4

0:1

0

Gorfrect or

POO-Correct

-.r

r

L2

Trial

4

Figure 2. Overall performance of the subjects on the recall task.

CZ*

r'

e

0,1

00.

.First consider the effect of the complete story grammar structure.

Netghhors in dhis analysis are called 'a priori story grammar neighbors,

since the number of neighborsAn the complete story grammar structure

8

. re counted whether or not the subject has actually recalled.them on

atny particular trial.) The number.,of a priori, stOwjrammar neighbors

appears to havelittt. any,'.effect on the subsequent recalL of

facts (tile left graph of Figure the'slopes of! the regression lines

P .

shown in Figure3 aFe givenin Table 1).2

In contrast to the number of a priori neighbors in the complete

a.

-story grammar structure, the remaining graphs in Figure 3 show

-the effect of die numbek ofineighbOrs which the subject actually mentioned

' . Os part-correct or correct-)" on trial n. The canter graph'of Figure S`A

'shows the averagescores on trial for facts which 'were absent,

part-correct, or,correct on trial: n as a function of the number of

, .

story g tcamEar neighbors of thA,fact which were mentioned on trial n.

There is a clear positive effect. No matter what the initial status -

of the fact as the number of its story grammar neighbors mentioned

on a trial increases, the fact,is more Likely to be remembered on the

next trial.

One question'is whether alternaive,structurestmight also show

an effect comparable to that of the story grammar structure.' A simple

alternative is the serial structure; linking the facts together in

a linear chain according to dheir serial order) in the Morison text:

Here, of course, each fact'has two neighbors, except for the initial.

13

1

f

4.

'N.

Z /

2.0

"C

=C

2 0

1

1.5

P C

n

1.0

1

An.

E.

13

45

+

025

An

21/0

1,5

r.0

tiso.

;4,

015

0

-01.

2+

,Num

ber

of a

prio

riN

umbe

(Of S

tory

Giii

mm

arc

Sto

ry G

rani

mar

Nei

ghbo

rs.

Nei

ghbd

rs,M

entio

nid.

on

Tria

k n.

%

Figure 3.

The effect of neighbors (op trial n) in the storygrammar and serial

structures On the average score for..a fact ,(on the:trial n+1).

..

Scoring: Absent = 0, Part-correct = 1,, Correct =.2."

Cn'

PLCn!

and A

.,..

n.

indicate facts which werl respectively cortect4

part-correct, and

k

absent gtrialen. The vertical bars

indicate + 1 standard error.

'

-7-

/

A .

1.2

Num

ber

of S

eria

lI

Nei

ghbo

r6 M

entio

ned

on T

rial n

..

Effect

fable 1

of Neighbors (on Trial n) on Average Score(on Trial n+1)

Linear Regression Analysis4

Statud'zo.f Fact on Tfia1 n

Ind4endentAbsent Port- Correct Correct

Variable Slope to Slope t %Slope ts.

Nur4ber'of a priori Story

Grammar Neighbors -011 1.53

.

.032 1.64*.

.024 1.34

Number of Story Grammar

NeighPor§ Antioned

on Trial n .'08 6.54***

s

-

.173, 4.20*** .105 2.69**

Number of Seri'all Neighbors

.

Nentionpq n .172' 8..86*** ,.125 2.94**

NuMber of Both Story, .

Grammar and Serial

,Neighbors Mentioned.-

'on Trial n, :2*30 7.42 * ** .148 2.42** ,.159 3.16***

Number of Story grammar.

but.not Serial

NeighboYs Mentioned

on Trial n p.061 .3.38*** .141 3.30*** .043 1.18

Number of Serial but not

Story Grammar Nekghborse't

iMentionqd on Trial n .131 5.36 ** .069 1.42 -.018 -7 43

. . ',, . .

aThe talue test the hypothesis that the slope is not greater than zero..

*E <001

r

O

1

0.0

c

and final facts. .The graph on the right side of Figure 3 shows the

Of

effect of the number of serial neighbors-mentioned on.trial n, on

the average score for a fact on trial n+1. There is also'a clear

effect here, comparable to that.for the story grammar structure. Of-.

course there is considerable overlap between the si9ry grammar*and

9

serial structures, as we 'should expect if the syntax of the passage

(the serial structure) reflects the sementics'of the, passage (the

story.grammar structure). In this particttlar case, 30% of the facts

.which are neighi3ors in the 'story. Fammar structure are alsoineighbors

in the serial structure, while 38% of the facts which'are neighbors

in the serial structure are also neighbors in the story grammar

structure./ The. question still remains, however, as to, what extent

our effects, reflect the story grammar structure, and to what extent

they simply'reflect the serial or42r of thefacts in. the passage.

To separate theteffect of the story grammar and serial structures,

a multiple linear regression analysis of the data was carried out.

The results of'-this analysis are shown in Table 2. When a fact is

absent on.trial n, serial neighbors mentioned on trial n are almost,

three times at effective as story grammar neighbors in improving the

recall for that fac.t on triarn+1. Wh4n a fact is pert- correct or

..

. .\

-I

Correct on trial n, however, story grammar neighbors'mentioned on,

\ *

trial n are more effective than serial neighbors in improving're all. . N/

on trial n+1/

' ..... f.

Tire effects of the story grammar and serial structures we ek.,

also compared in a somewhat-different manner by separating the, .

r ,

. neishbOrs of each fact into three groups: first,those which were.

r

i 6

4

4

I

Table 2 , 13- , ..,

Comparison of Effect of.Story.Grammar and Serial Neighbors

(on Trial n) on Average §core'(on Teialn+l)

Multiple Linear Regression Analysis

0

Status of Fact on Trial n

Absent Part:Uri-bet tependent #

Variable Slope -ta Slope t Slope t

Number of Story Grammar

Neighbors Mentioned` on. .

.Trial n .056 3.10*** .149 3-.45*** .094 2.36**A

Number, of Serial Neigl4ors

Mentioned on Trial n .143 .69***-

.077% 1.75* .063 1.58,

- aThet values test the hypothesis that the slope is not greater than zero.

**g<.01

* * *p <.001

1

v

t, .

neighbors in both. the story grammar and serialistructures;, second,

'. .,

hose which were neighbors in the' story grammar structure but TA)

in the.seriAl structure.

(story grammar,-only); third, those which- .

t

were neighhors.in the serial structure but not in, the story grammar

4'1

structure (serial-only). The results of'thls analysis are shown in ,rp . N

i il.Table 1. Again we Ide that the serial-only neighbors have a stronger,

Er

lo

.

PoSs'tiN'ie eff.. ea on the su4se4uent recall of absent facts. However,V /-"'' .----w both serial-only and story grlmar-only neighbors have positive

.

'effects on theceubsequ'ent recall.ofq.part-correct facts, the story44 \ I -

grapmar-only neighbors have the stronger effect. Finally, while the

story g ammar-only neighbors have a Knitive effect on the subsequent.....t_ ; e

\ .

irecaTI of correct facts, serial -only neighbors have no effect.3

4

DISCUSSION 4

In this itudy I have, analyzed a natural narrative prose passage

into two structures: a serial structure based on serial order in the

passage, and a story grammar structure based pLmarily on causal'

relations within the passage. The results show that these structures

are important for the memory and recall of the passage.

If we loofi at some particular fact on a given trial, the number

7of neighbors of that fact (according to either the serial or story

grammar structure) which the subject has alSo mentioned will influence

the recall of that,fact on the next trial. In generA, as a fact

has more neighbors mentioned, it is more likely that the fact will

be remembered correctly, and less likely that it will be forgotten)

on the next trial. However,'there4are important differences to the

- I

,

effects of the serial and stork grammar structures.. When a fact, is

'absent on trial n, the number of neighbors in both the serial and

story,gramffiar structures mentioned on trial n affects the, recall of

that fact on trial n+1, but the number of neighbprs in the serial

1,

structure has the gterer effect. When a-fact is partially-correct

on trial n, again neighbors in both structures mentioned on trial n

affect the recall of that fact on trial n+l, buS now the number nE,

story grammar neighbors mentioned has the greater effect, Finafly,

if a fact is correct on recall n, its recall on trial 114-3 is

1,

influencbd only by the number of its neighbors in the story grammar

structure also mentioned on trial n; the nurdb.er of its neighborsoin.

the serial structure mentioned on trial n does not have any effect.

This pattern of results has a simple expla4nation./ On first

bearing the tape recording, tlie.subjects perceive the passage as a

64colIectionof sentences or facts,strang together ih serial order,

but as portions of the passage begin to "make sense," they perceive

and organize the passage in a manner closer to its undeTlying meaning

Atructure: the serial order loses its importance. I found that,only

the neighbors in the story grammar structure,actuall mentioned by

the Asubject on one .recall will affect performance on the subsequent

recall. The structure inherent in a prose p age alas no effect

unless it is present in the subject's memory\ four that passage.

4)

1

Tr

0.

12

References

Crothers, E. J. Memory structure and the recall of disbourse. In-3

R. 0. Freedle & J. B. Carroll (Eds.)s Language comprehension and

the acquisition of knowledge. Washington, D,C.: Winston, 1972.

Frederiksen, C. H. Effects of task-induced cognitive operations on

comprehension and memory proccesses. In R. 0. Freedle & S. B. Carroll

(Eds.), Language comprehension and the acquisition of knowledge.

Washington, D.C.: Winston, 1972.

Frederikkra, C. H. Acquisition of semantic information from4discourse:

Effects of repeated. exposures. Journal of Verbal Learning an d

Verbal Beliaviot, 1975, 14, 158 -169.

Monk, D., & Kintsch, W. Memory'search 1: Paragraph memory and the

retripval of inforthation. In W. Kintsch <Ed.), The representation

a 1

of - meaning in meinory. Hillsdale, N.J.: Lawrence Drlbaum Associates,

© 1974. ,

. ,

neyer, B. J. F. T he organization oft

prose and its effects on memory.I.

2

% '

0'

Amsterdam: North-4011and Publishing Co., 1975.

Meyer, *B. J. F:,& McConie, G. W. What.is recalled after hearing a'

passage? Journal of Educational Psychology, 1973, 65, 109-117.

Morison, S. E. The Oxford History of the American people. New York:

&ford 41niver4ity Press, 1965. eft.

N man', D. A., Rumelhart, D. E., & the LNR Research Group. Explorations,

in Cognition. San Francisco: W.H. Freeman, 1975.

Rumeart, D. E. 'Notes on a schema for stories, In D. G. Botirow &

M. Collins (Eds.),, Representation and understanding: Studies

incognitive science, New York: Academic Pregs, 1975.0

a

11

. .

Footnotes

1The straight lines shown in Figure 3 are the least-squares fit. .

to ate data, 'with the points weighted in proportion to the number of

,

cases they represent. For example in Figure 3'theretwere 1891 cases

I ,

:E`-of an absent fact with one neighbor in the story grammar structure,

but only 623 cases'of an absent'fact with two neighbors. 'Therefore

these data wee weighted inthe ratio of 1891 to 623 in determining

th# least-squares-fit line.

While the maximum number of.neighbors in the serial structure is

two, facts can have up to seventeen neighbors in the story grammar

, .0

structure. .1n order to,simpaify ,the graphs and make a fairer comparison

between story grammar and serial neighbors,,in the center graph of

Figure 3 data for more than two story grammar neighbors mentioned (5%

of the total dgta) have been grouped With data for two neighbors mentioned.

Similarly, in the left 'graph in Figure 3, d t& for morethan five a prioriI

story grammar neighbqrs have been groupe'd with data for five neighborsd ,'

.This procedure does It materially affect the resultg or conclusions

in either ease.

2Ot course, the'itumber of a priori neighbOrs isflot# completely

independent of the number of neighbors actually mentioned by a subject,0

since for instance a fact with five neighbors 0 the complete structure

has. a greater potential, number of mentioned neighbors than a fact with

only one,neighboLin the complete structure.

ijof mentioned neighbors3This paper reports the effect of the number

(on trial n) on the averege score for a fact (on trial nig). I have

also analyzed the data using other mods of scoring facts; looking

at the effect of the number of correct neighbors, and assigning daferent. ,

weights to correct and,patt-correct neighbors. These analyses all yield

results essentially identical to those presented here.

2 ;