DOCUMENT RESUME
ED 138 924 7S- 003 356 .
'AUTEIOR Rigney, Joteph W.; Mun7o, AllenTITLE On Cognitive Strategies for Processing Text.INSTITUTION University of Southern California,.Los Angeles.
BehavLoal Technology Labs.SPONS A2ENCY AdvanCed Research.Projects Agency (DOD) , Wat'lington,
Office,of Naval Research, Arlington; Va.Personnel an Training Research-Programs Office.
REPORT.NO BTL-TF-80PUB DATE Mat 77CONTRACT NO0014-75C-0838NOTE 70p.
EDRS PRICE. MF-$0.83.RC-$3.50 Plut Postage..DESCRIPTORS Artificiad. Intelligence; Association (Psychollogical);
*Dogntiie Processes; E.ducational Theories;*Expectation; Models; *Reading Comprehension;*Reading Processes; *Reading Research; Recall(Psychological).; Semantics;Syntax
IDENTIFIERS *Procedural Semantics; *Schemata
ABSTRACTRecent'deve Iopments in cognitive.psycholOgy and'
artifiCial intelligence have shown that various types of priorknowledge-play important roles\in understanding ,during textprocessing and have,retulted'in a new kindof-MOdel for cdnceptualproCeSsing, "procedural semantict."..LThi:S papei discusses-two typet ofunits, or schemata, which, according to this theory, are responsiblefor Conceptually driven processing- in reading: the form-schema.aceountg for readers' syntactic or formal r-xpectatiohs; the ,
Content-schema accCun.ts for readers' semantic'expectations. Modelsfor specific:forM- and content-schemata are proposei; implicationsfor effectiVe ttraagies fot adult realing aresderived from theptemises of the model; and'four.kinds of reading strategies(single-piss, exhaUst-ive muleti-pass, extractive multi-pass, andsel.eciive multi-pass) -are charactetized in terms of.themodel. Inaddition, potential appliCations, suggested by the consequences of
the theory, ate deAcribed.. These include uses for headings in texts,meant for constructing advance .organizets'..for textt, and.training of.readers to make more effective uSe of texts by being sensitive tbtheir motivating tasks and by exploitiag their capacities,focgenerating expectatioa-s- about the meaning ofthe texts, throughGonceNtually driven- process5ng. (Author/LJR)
**********4***************************1c*********#*******************tc* Documents acquired by ERIC include many informal unpublished *
* materials hot available-from other sburces. ERIC makes every-effort* to obtain the best copy. available. Nevertheless', items Of marginal *
* reproducibility are'often encounterel and this affects the quality *
* 'of the'microfiche and hardcOpy reptoductiont ERIC makes available.* via the ERIC Document Reproduction Service' (EDR5) . EDRS is not, * .
* responsible for the quality Of the original document. Reprodnctions-* supplied by EDRS are the'best that-can be-made.from the `original. *
-*********************************************************************
r 1 a
U S DEPARTMENT OF HEALTH.EpUCATION & WELFARE.NAT'ONAL. INSTITUTE OF
EOuCATION
DOCUMENT +-4AS BEEN REPRO-DuCED ExACTLY AS RECEIVED FROMT.E PERSON OR ORGANIZATION OR!GIN-z.TNG IT POINTS Or VIE& OR OPINIONS:.T;.TED DO NOT NE.CESSAR1LY REPRE-SEN OFF ,CIAL NATiONAL INSTITUTE OpEDUCAT,ON POSITION OR POLICY
BEHAVIORAL
TECHNOLOGY LABORATORIE.S
Pr\DEPARTMENT OF PSYCHOLOGY
1641 UNIVERSITY OF SOUTHERN CALIFORNIA
Technical Report No. 80
ON 0CCNITIVE STRATEGIES FOR .PROCESSING TEXT
'March 107
-Joseph W. Rigney and Allen Munro
Sponsored byx
PerTonnel and Training Research ProgramsPsychological Sciences Division
Office of Naval Research
and
Advanced Research Projects AgencyUnder Contract N9. N00014-75-t=0838
"N.
4
The views and 6onclusions contained i.n this document are those ofthe authOrs-and ?j-lould not be interpreted.as'necessarqyvrepresentingthe official.policies, eithet expressed.or implied, of the Office ofNaval.Research, the Advanced Researdh Projects Agendy, 9r-the U.S.Covernmeut.
Approved far public.telease: D.istribUtion Unlimited.
This document has been approved -for public release and, sale;its distribution iunlimited. Reproduction in whole or in
° part is 'permitted- for any purpospe of the Llnite.d State,s
-Government
\
Unclassified
SECURITY ,!LAssiFICATION or TritS PAGE 'Hhen End;REPORT DOCUMENTATION PAGE
-READ INSTRUCTIONSBEFORE COMPLETING FORM
1 REPORTNUMBER I: GCVT ACCESSION NO.),
Technical Report #801
3 RECIPIENT'S CATALOG NUMBER
5 TYPE cArr REPCRT 6.PERIOD COVERED1
1 Jan. - 30 June - 1977
4 TITLE r'eu2d Subrirta; .
ON.COGNITIVE STRATEGIES FOR .
.
PROCESSING TEXT _ .
., .
6 PERFORMING ORG. REPORT NUMBER
7 AUTHORisr , , . .
J. W. Rigney.and Allen Munro.
8 CONTRACT OR GRANT NUMBER(a)
vN00014-75-C-0838I; %.
.
. .
_
9 PERFORMING ORGANIZATION NAME ANb ADDRESS,
Behavioral Technology Laboratories ,
Tniversity of SOuthern Cafl1
forniaLos Aneles California '.90007
10. PROGRAM ELEMENT. PROjECT, TASKAREA & WORK UNIT NUMBERS
Program elemffit. 61153NProject: -0Task Area:
R0 6RR042-06-01
Work Uait: 154-355I CONTROLLING OFFICE NAME AND ADORES!? ee-
-
Personnel and Training Research ProgramsOffice of Naval Research (Code 458).Arlington, Virginia_ 22217 ,i,
..1:.' REPORT DATE-.\-
March.- 197713. 'NUMBER OF PAGES
--",....,.. 49 + vIA MONITORING AGENCY N AMEfi ADORESSil( dIllerent from Controtlinal Office)
. r.
4.
.
.
15. SECURITY CLASS. (of thug raport),`
Ucr:lassified.
.
15a. DECLASSIFICATION, DOWNGRADINGscHEDULE.r
,
lb oisTR;at..cric;iTATEMENT ((Willie Report), ... .
.
.Approved for public release; distribution unLimited ,
.
.
.
'
.
-N. ..
17. DISTRIBUTION ST ATEM-ENT (of tri abstract entred in Block 20, If different from Report)
.!."-
_
. .
..
18. SUPPLEMENTARY NOTES'
,.
%.
19. K EY WORDS (Continua on reverse fde If neceireary and.identi(y by block number)
Text. Procesaing Strategies.
Procpdural Semantics' .
Conceptually-Driven ProcessingExpectatiOns in Reading
. .
..
.
20. ABSTRACT (Continrwe on reverire Ido If nceeary and Identify by block number)
Gurtrent research on reading is hampered by the Jack oa-framework Within which to study .the effects of areader's priorknowledge on his or her processing of an unfamiliar text. As a
result, most reading teseart.h has.empha'sized perceptual ratheq'than
congeptUal processing during reading. Evidence is cited in support
of the claim.that Various types of prior knowledge play important ,
roles in understanding during text processing.
DD, FORM1 JAN 73 1473 .EDITION OF I NOV 65 IS OBSOLETE
S/N 0102 LF 014.6601Unclassified
SECURITY CLASSIFICATION OF THIS.PAGE (When pedal Entered)
Unclassified
SECURITYCLASSIFJCATiONOPn"WSPAGEOMOnDelsammo.
Recent developments in cognitive psychology an artificialintelligence have resulted in a new kind of model for conceptualprocessing,.called_procedural semantics. In this report,e framework
' is laid for the application,of the procedural semantics formalism'to thenslysis of conceptually-driven'processing in-reading. Accord-ing to this theory', two different types of conceptual processing units(called schemata)' are responsible for conceptually-driven processingin reading. One type iS the form-schema, which accounts for the,SyntacL.
-tic or formal expectations which petiple make use of in text processingThe other type is the content-schemfr7 which accounts for the nature ofreaders' semantic expedtations. Models'for a small number-of-specificfotm- and content-sdlemata are proPosed, and certain experimental.and
- observational evidenceA.s'explained in terms of these models.
Implications for effective strategies for adult reading ar.E2derived from the premises of the medel. Several.di-fferent kinds ofreading strategies are,chatacterized in terms of the model. Whenreaders eMploy Stngle-pass strategies, they process the text in astrictly finest-, left-to,rght fashion. This approach makes minimaluse of-the potentialor conceptually-driven processing that could beachieved through the-activation of some high-level schemata. Inexhaustive multi-pass proceising, the first pass results in thesctiva-tion of a number:Of-form and content-schemata which can serve as anaidin subsequent-:passes, driving expectations about the form andmeaning of what is about.to be read. This technique can often bewasteful of.resources, since it does not 1actively direct ,processing .
toward what is most important or least well Understood. Extractivemulti-pass prOcessing reflectS a more efficient strategy for readingan entire ext. By usingthis technique, a reader "skims" the text ina sq-lecti/e way on repeated.passes, buildihg up such a complete under-standing of the meaning of the text that the final reading of the textis often a process ofmerel§ filling in the gaps in understanding.This technilque is often effectively used by graduates-of adult'readingimprovement classes. .Selective multi--pass strstegies characterize.thereading ofthose who'knoW what it is they want Eo know, and whO arewider no consttaint to learnall that might be learned from a-text.In this type of text yroceesing, the reader beginsthetssk with the
, intention of acqUiring soMe specific information.' As a result, anumber of specific content-schemata are activated and are used.to guide_the order and the selection'of those portiOns of the text to beprocessed.
Several potentialepplications are suggested by-the con-sequences of:the theory forconceptually-driven-Orocessing in reading,presented hereg These include possible uses for headings in texts,0means for constructing advance organizers for texts, and trainingreaders to make more effeCtive use of texts by being sensitive totheir motivting tasks end by exploiting their capacities for generatingexpectations-about the meaning of the texts through Conceptually-driveuprocessing.
-,UnclassifiedSECURITyCLASSIF.ICATION OF THIS PAGE(When Dto Entered)
1
ARpA TECHNICAL REPORT
1. ARPA Order Number2. ONR NR Number3. ProgiaM Code Number4. Name of Contractor
5: Effective Date.of Contract
6. Contract Expiration Date.7. Amount of Contract8. ,Contrpct Number9. Princikpal Investigator
10.. Scientific Officer11. 'Short Title
11.
.2284--
: -154355: 1 B 729: University of Southern CalifOrnia:/ January 1, 1977: *30 September 1977: $150,000
N00014-75-C-0838Joseph W. Rigney (2134 746-2127 .
MarshalljarrCognitive Strategies for 'Reading
This Research Was Supported ,
by,
The Advanced Research Protects Agency
and bi,
The Office of Naval Researgh
And aS Monitored by
The'Office of Naval Research
5
4f
s.
SUMMARY
Current research on reading is hampered by the lack.ofa frameWOrk.within which to study the effects of a reader!s priorknowledge on tis-or her processing of-4n unfamiliar 'text. As aregUlt, most readtng research has emphasized,perceptual rather thanconceptual processing during.reading. Evidence is cited in sUpportof the _claim that:various types of prior-knowledge play importantroles in-understanding during text processing.
Recent developments in cognitivepsychology and artificialintelligence have reSulted ina new kind of model for conceptualprocessing, called procedural aemantics. t'his report, a frameworkis laid for the application of the prodedural semAtics formalismto the analysis of conceptually-dr-b./en-processing in reading. Accord-ing to this theory, two.different types of conceptual proceising units ,
(called schemata) ire reSponsible for conceptually-driven.processingin reading. One type is the formsthema, which accounts for' the syntac-tic or formal expectations.which people make use of in text prgcessing.The:oiher type is the content-schema, which aCcounts for the nature ofreaders' semantic expectations. Models forOmall number of specificform- and content-schemata are proposed; and/certain expe'rimgntal andobservatiodal esvidence is explained in terma of these Models.
. .
"Implications foreffective.str tegies for adult reading arederived from the premises of the model. everal different kinds of .
reading strategies are characterized in ems of.th model. WhenreaderseTploy single-pass strategies, they process t e text in a
.. strict15; linear, left-to-right fashion. This approach makes minimal_
use of the potential for conceptually-driven processing that could beachieVed through the activation of some high-level schemata. In
, exhalistive multi-pass processing, the first pass' results in the active--tionof a number of form- and content-sthemata which can serve as anaid in subsequent passes, drivihg expectations about the form and
' meanimg, of what is about to be read. This technique can often bewasteful of resources, Since-it-does not actively direct processingtoward what is,most important or beast well--understocid. Extractivemulti-pass processing-reflects a more efficient strategy for readingan entire text. By using thia technique, a reader "skims".the text in-a selectiVe way n repeated passes., buildiv4; up such a complete under-standing of the meaning of the -tekt that the final reading of the textis-Often a process of merely filling in the gaps in understanding.This technique is often effectively used by.graduates of adult.readingimprovement classes. Selective multi-pass,strategies characterize thereedill'g of those who know What it.is they want to know, and who areunder no constraint to learn all that might be learned from. a.text.In this type of text'processing, the.reader begins the task with theintention of acquiring some s ecific information. As a result, anumber of specific content-sch ata are actiwated and,are used t.) guidethe ordei and the selection ot rose portions of the text to beprotessed'
-4
Several potential applications are suggested by theconsequence s'. of the _theory for, gonceptually-dri'ven processing inreading presented here. These include posstble °uses for headings in
11texts, means for constructing advance organizers for texts, and train-ing readers to make more effective use of texts y being sensitive
-..
to their motivating tisks and by exploiting their capacities forgenerating expec.tations about the meaning of the texts through concep-tualy-driven processing. .
A
10 0
ta.
6
ACKNOWLEDGEMENTS
The-research desCribed in this report wAs sponsored-byDr. Marshall-Farr and Dr. Henry Halff, Personnel and Training ResearchPrograms, Office of NaCial Research, and by Dr. Harry F..O'Neil, Jr.,'ProgriM.Manager, Cybernetics Technology Office, Defense AdvancedResearth Project.sency. 'Their support and encouragement is gratefullyacknowledged.
te
a
4
-
.
4
TABLE.OF CONTENTS
Section Page',
A1
- II. SCHEMATA IN CONC UALLY-DRLVEN TEXT PRCCESSING 7-
I. INTRODUCTION_
Form-Scriemata
Form.-Schemata in a iiroader Context-
Content-Schemata4
15
24
More Comprehensive Content-Schemata-I
Experimental Evidenck for the Effects of
27
Content-Schemata in\Realing 30
11.
III. APPLICATIONS TO ADULT READING STRATEGIES 35.
.
Text-Processing StrategiesI
, 36,
Single Pass Strategies 37
Multi-Pass Strategies 38
Exhaustive Multi-Pass Strategies 39
Extracti e Multi-Pass Strategies 40'Selecti e Multi-Pass Strategies. 45
REFERENCE1-
43 -iv-
9 .
If
o
47
LIST OF FIGUIRES
'4
Figure
A Stage Representati of An Interectivg. Model
9f Reading
A Two .Dimensiona 1 S li.ce of the Messeie 'Centqr
A Scene\
Examples of Different Schem'Idte
Page
1.
2 .
3.
4
4.
5
28
fi
e
(
r
1 0
. a
C,
ON COGNITIVE STRATEGIES FORPROCESSING TEXT
Introduction
W. feel a strong need for some kind of model, or theory, oZ
learning from text, to counter the current sterile empiricism in this
field, by giving direction to research, Lnd to deal systematf.cally With-
the characteristics of different kinds of information in the text
itself.
Text is a linear string of symbols organized, bTspacing,
into characters, words, sentences, paragraphs, and'higher order sub-._
1 -
ivisions. A small set of characters can form an enormous number of
.w\ds, but only a small set of these is used in any one text. Words
can be combined to fordan infinite number of physically different sen-,
tences, but there are only'a,few commonly used rules for determining sen-
tence structure. Patterns Of characters and patterns of words recur
over and over in words and in Sentences. But, every piece of text isP
different from every-other' piece, making it difficult to specify and
to control stimulus parameters.
There is litt)e doubt that the familiar 15atterns of letters
in words (Alderuidn & Smith, 1971), and of wards in sentences facilitate
text-processing.taSks. -Faliar structure allows prediction of elementi
in the structure. Letters of familiar,words_can be read out of memory
after a glance at the printed words. Travers (1973) showed that "...words,
or large segments of words,/are habitually prptessed in parallel, While
random strings are processed as a series of individual letters or small
chunks." .-(p. 109). The familiar structure(of sentences permits pre.-
diction of classes of,words-coming next; nouns, adjectives, verbs,
connectives, and the like (see Kolers, 1970; StevenS. & Rumelhart, 1975;
1
Weber, 1970). The readability of the poem "Jabberwocky" is a tribute
to the facilitative effect of normal_sentence- structure, communicated
in this case by the presence of particular "function Words"-(such as
prepoStions and determiners) and by morphological affixes (such as
tense endings, plural markers, and the like). ParagraWi structures _
are much looser, indeed, almost arbitrary in comparision to sentence
structures, which, in turn, ard more flexible than word structures.
Loose though such structures may be, however, readers are.sensitive
to aberrations in paragra-A §tructures. Meyers & Boldrick (1975), for
example, showed that randomly rearranging the order of half of the- -
sentence§ Tn a text resulted ina severe deierioration of.subjects',
abilities to recall the stories later. But, Suppose that som e,. sentences
were anomalous-in termsof meaning; the wrong subjects for the verbs
-
and objects--the locomotive drew a picture of a cow. This introduces
semantic "sErambling" without topographic disorder. Marslen-Wilson &
Tvlet.-(1976) describe tlie:.effects of this treatment on recall. Or,r J
suppose one.sentence.in:the paragraph was not related to the topic of
the paragraph. Bransfopei & Johnson (1973) demonstrated the effects of_
this kind of seMantic scramblfng on memory for the information in the
paragraph.
e
-
We-are saying there must be some kind of top-downprocessing- _
that is predicting detailed levels from highen structure, and that4
occbrs in parallel with the bottom-upPrOcessing that synthesizes words
from letters, sentences from words, and paragraphs from sentences.
Bottom-up Processing has beefi almost excluA.vely, emphastzed in,theories
,. about reading.
If the levels of te)cf prbcessing task6,-from proceSsing,
letters, to proCessing words, to srpcessing sentences---werecombined
in appropriate ways with the levels of scrambling text--from letters.
in words to words in -sentences. r, senronces in paragraphs, then it is:
likely that discoutinuit n dependent measures that woHld
irAdicate that at least, ( Lccesses.Were running in paral'
. Rumelhart (in.presS) has advanced am interactive modej of
reading that describes in some detail mechanisms for.interactive top-dow'4/-
bottom-up processing.- Noteworthy are examples he cites of how top-down_
proce.3sing can influence bottom-up processing, his-parallel,pr ocessing,
Mechanisms, and his-Oantification of the interaction between the two
-sets-of parallel proCeases as a.multiplicative relationship between
direct eVidence and contextual evidence for hypotheses that drive pro-.,
- ceasing:
ahe values Of:Vi and Bi)aredeterMined in termsCT
of mixtures of higher0
level (parent)4,-same level (right and left sisters ) and lower-levelN
(daughter) hYpOtheses prevailing=at any one time.
,The'following figures from Rumelhart suggest some of the
features of his model.
The pattern'synthesizer in Figure 1 contains-a mesSage center. 2
tha
"Keeps'a running liSt of'hypotheses about the.,nature of the,input sting. Eaph knowledge sourceConstantly acans the message center,for the appearanceof-hypotheses relevant tOjts own sphere of knowledge.
-Whenever sucha hypothesis Oters themessage center:the 4nowledgesourcein .question evaluates the hypo7thes in light of itS own specilized knowledge.
13
I.
14
'-FigUre 1,
,A sta0 representation of an interactive model of reading,
(From Ru elhart,in Friss)
!OST
FRO LE .
INTERPRETATION
, Phrase
Level
Syntactic
Level
Categorical
Level
Lexical,
Level
Letter
Cluster
Zavel
Letter44 Level
cl
0
4
0
a Feature,
'Level
'NP'
16
Figure 2,
No dimensional slice Of ihe message center.
(FroM kumelhart, in Press)'; '
As a result of its'analytis the hypothesis may geconfirmee,.dis'cOnfirmed and removed from the 'messagecenter, or e new hypothesis.canwes'sage center:. This procesC;cOntinues until, somechcition can be reeChed; At:qhet point the-mostprobable hypothetis is determined te be the ,correctone. To facilitate the process, the message centeris highly, structured to t.-,hav the knowledge sourcesknow'exactly where toJindrelevant hy.P75Theses and '
so that.dependencies amonglypotheses are easily.'determined.
/
T/
The meesege.center can-.be represented as a three74dimehtional 'e.pace./,,Dne dimension representing theposition along the/lihe of:text, one.diMensionre-,preseptinethe.level-of theilypothesis (word level,letter levelj pUrase'leVel, ane dimensipnrepresenting aiternetive:hypotheses at't.he -same
Astociated With each hypothesis is e running'estimate oEthe *obebility'thatjit is the correcthypothesis: MOreover, hypothetet et each level mayhave potnters tO hypotheseS at higherHor'lower4levels,on which theyAaredependehtThue, for exampfe, thehypothesisehat the first word.in,e .string is theword THEit'suliportgd by the hypothests' thatlthefirst-letter df the strAng Is 'Tt anci sUpPorts the
,
hypothesis thet the string begtnt with a noun phrase.-'Figure 2 illuttratesA twd-dimefsiOnal slieemessage cedtek et some,point during the.reading It thephreee THE CAR. The figUre,ilfustrates hYpothesee atfive different levels.(feeture level, letter, leVelletter cluster level-, lextcal_leVel.and 4ynteeti.C'.level). The ciiiagram a,twaildimensionalsiiceinaemuch as rc) alternatiVe hypotheset are-illuStrated.
;In practice, lof course, many alternative h'Ypotheses- ,
would be conSidered and evaluatedzin the,.,eur.se of-:i.eadtae thisiphrate. It shoutd be popited outthatthe tNie,like etrhetw:e ehouldndt beitaked.tO:fmean'that the.tree wae.eonstructed -either from:a purely"bottom-up" procets (starting witk the features, then,hypothesfzinethe letters, then the letter.ciutteiLetc.), nor from a. pUrOYs"top-doWn" analysis (startingwith the,viewthat''ve,have anoun phrase and.that.noun -phrates are made up of,detetminers folloWed, bynouns, etc:)J Rather, the llyPotheses e'en be
.generated.etjeny..level. -If it_is.fikely that a linebegins- with 0 noun:,phiase, then- we poStulaté a nounphraseYand..lOok.for evidence.±If me see featUree
18
\
that suggest a "t"'as the first letter wepostulate a "t" in.the first position andcontinue processing. If we later have ioieject ei:her or both of these hypotheseslittle is lost. The system makes the' bestguesses nd checks out their imgications.If these guesses are wrong. it will'take abit'longer, but'the system will eventuallyfind some hypothtses at some level 'that it
, can accept." (Rumelhart,..in'press)
Schetata in Conceptually-DriVen Text-Proce-
In his proposal.for/aninteractivaModel of ru ,ulue_
(in press))makes it clear that there is otdina ily more than one source
.4 .oi conceptually-drrn,kroceSsing during reading., One type of soUrce
s.for top-down ana ysis is the reader's hypothesis, about the general'kind
f situation or nteXt with.which the text deals. 'A differenebtype of
Source fjr tbsp-,dow processing is the presence of speciiA "triggere'
*.in the s C -Pion or context, (as perceived-:hy the reader) which result,
in more lOcal or specific expectations than those of the first type.
Let us illustrate the-difference between these two types ofw
ir
sJurces for top-town processing by.bOtrowingan extended example of
. .
Rumelhart's. IfeSubject is first shown a picture of a scene (not,
let us agree for the present, of an activjty); and is.then"presented
with a tachistoscopic expbsure of a brief phrase that, according to the
instructions, describes some asPect or element of-the picture, then w6
di.4terent types of top-down processing are probabiy brought to bear in...
,
4, ., the subject's reading of the. phrase. Thelfirst type of top-down
processi is that which has to do With the subject's expectatiOnsN.
. .
created by the instructions: That is*, the subject understands.that the,
. .
alphabetic string to be presented.will have something to do with th
.
li.. ..
. .
scene previously predinted; tEat At will' be, in fact, a partiar'''. ,
.
description Of that,scene.- Thissort of expectation is very different
19,
20
.Figure 3.
,
A, scene, Figure prCvided by jean Mandler.
(From Ruielhart, in press)
# .4 4
,
0
I
,
$
from many or' 'ch the subject sow lave, were 11-? intri ons
;different. i.nLltructions,
expect to see a phrasc± which was an,bmpianation of'Some aspect of the
ample, the subject. might
picture If this were the case,othe subject woUld expect a liwer.
tmph.-tse than he does in the case of a partial description, and a phrase
4
wfth specifically relational or causative predication. Another exaMple
4of a .different sort of..expectation which different instructions cquld
(0'
, ..
. . - ._.,_ ,
arouSe is that of a prediction related to some. aspect 'Of-the picture.-_ -
In this case, the subject would presumably expect that the phrase
would contain some actional or change- state pred\icate.
The second type.of top-down or conceptual,y-driven processing
. ,
dn the context of this experimental example.of RumeIhart!s is the .
roUp of expectatigns.arouSed by specific details in'the_presented4rt
scene. Thus, a'prcture of a Wakswagen parked at a gas statiodwith a0 .
lake and mountains in' the background (see Figure 3) mabrespons1b1e .
forAkies of expbctations, such as that the phrase might be "A".
Volkswagen," or "A gas station," or YThe Mountain," or "Talake, and
so on. In fact, of ,5ourse, this is only a small fractiOn of the pos-,
sible specific e4pectations in,such a context; since many synonyms or
conte*tually equivalent fexPressions are possible, such as car,
"The 'service station," and so on.
Blow, it is all very well to specifythat there are at least
these twoo4fferent types of conceptUally-driven processing which must'
^ J
be a part of the-reading process, but to do this
-
same thing as showing how such c,yoceptually-driven rocessing is1
actuallyaCcoMplished. We think.that thdre is a great need for models---
of specific conceptuai/prOcessing entities 'which are responsible for1 \
not at all .the
7-
r
,
theSe top7down_eflects.- In wfiat'folloWs we sketch an approach to,
such a model.
1
- A/natural framework f4or the construction of a. model ofconbeptual entities' which have both structural and procesSingaspects
4
is that of procedural semantics (Minsky, 1975; Norman, Rumelhart, &
LNR, rg*5;.Winogre, 1975). 'Much of what follows is.based on a par-
rticular proce4,urar semaTtic0 model, naTely thatlifthe LNit research
...group
,- .. .0 4 -1C,,
'Above we mentioned twetypeS of
4,
,Conceptuaildriven processing
hich, might be expected in the paradigM.in which a presentation of a
4 INstopic preseneation of afi alphabetic
a
picture is followed by the:tace
string. The effects ofhe two typ s of proessing carjbe modeled.
, 3 ')
withIn.a procedural semantics system by4two,different'types of schemata
/
which are activated in this ekperichental context. The first tyPe of
conceptuaily:dri.,7en processing is that.which involves expecta0ions or
. .
I*
predictions about the form Of the String, which appears brieffY after. ,
the ilicture. This iolves a.number of fairly explicitprediction
aboUt syntax,,and a 'few expettationa about pertions of the semantic and
lexical content of the strinss. -WeProposethattfie effects of these0
-4. :.
kinds of ekpectatiou be modelk,by a type of schema, which, for want'
ki
of a'beciker wile, we will tall formrschemata for the present. -kamp,les,-
of form-Sch mata, explaineld in greater detail,below, include a bescrip-7
,
:option-Schema an-Explanati.op-SChemand a.Prediction-Schenia.'
, ,
The second type of!conceptualij=driven-processin-g Which--should
,
....:...
.take place in-the paradigm of.picture-thentphrase'involves expectations,
., ,
- 10_, \
\--
,;- .. .
t, ,abOut tfie content of the briefly-presented alphabetic stringsz... Thi s' : ,\ L.).-
\-4.
ii
.
. .
7rocessi= cau.be thbusht of as,_a_number of fairly explicit expectations \, ,.. _. .
-*-)
C.
I.
-10-,
with respect to the seMal:tics and to'the lexical components of an
Utterance, In only a few cases will this type of processing result,
,in syntactic expectations:, and thoae cases will ordinarily involve
.
"idiomatic" expressions, which are themselves much like lexical items.
Within a procedural semantics model, the prodpction of these types of
expectation i. can also be mOdeled, by the actions of another type of
schema, ihe content-schema. Examples.of content-schemata, discussed
detail below, include a'Service-Station-Schema, (.4 Volkswagon-Schema,'4
a Mountain-Schema, and so on.
,
In the next two sections, sime illustrative exampIeS of form-
and-content,,schemata are presented.
6Form-Schemata
, In the experimental context which,we have been uSing ns the
t ,
-basis fon organizing this,disCussion, there are a' number of possible,
form-schemata which mighE be activated depending on the natur of the
'instructions given to the subject, as well as on othet-factors0
ttie sake of the p e eni discussion, let us es"sume that there are four
;
possible types of phrases which subjects Coulp be led to expect to see
follow the pictUre. They are descriptions, explanations, predictions,A
and histories. In the example given in Rumelhart (in press)the sort
of phrases used are descriptions.
, 'Other factors which might be expected to 'contribnte tOthe activa- .
tiOn of form;schata must-inc de thesSubjectts previous experience with :this:experimental.paradigm, a _the nature of his onher. perceptions.of thepreceding phrase stimuli in the'cürrent experimental block; For example,
'if the instructions informed subjects that they would be presented withpattial.descriptions of the- picture,'we would expect the Description-Schema
, 9
to hecipMe activated,'and Co affeOt the ...iubject's -interpretations of thefirst phrase presented. If, however, the presented strings were in fact
,. predictions rather than,descriptions, then to the extenA that. subjectscorrectly manage to perceive.the,truematgre of,the.OrerenteAphrases, their,activations of-the DescriptionrSchema should be replaced b3Pa:Ctivatiods of-
. . :,
the Prediction-Schema. ,
. .2 4-11-
if-
Possible degcription's(which'Cou/d follow a presentation of Iigurel
include "The car," "A service station," "The car is at the service
station " "The lake in fron't of the mountain," ahd so:6n.- What kind of,
explanations might follow a presentation of Figure 1? "The car is stop- ,
ped in the road because it ran out of gas," and so forthie _4edictions
which-tould be presented after the subject-!sees the ,picture in Figure 1
include "The car will leave soon," "The driver will -get back in tfie car,""
'"The lake will dry up," etc. Histories which could be Aresented 'after
titlio picture include phrases, such as "The driver of the car went into the -
gas station," "The car just ran out of gas," and "The lake was dry last :
summer."
Let's consider the structure of some of these form-s,chemata.
We have.chosen to represent these schemata.in the format of pvdicaie
calculus, in order that the Predleate-argument or procedureparameter
relationships (the scoPe relationships, inOther words) should.--be clear.
In order to make the, se schemata7active Procedures 'in a computer simulation,"
tfiey would have tb be.integrated with a pre-existing data base of lower-
level.schemata, such as thbse named as sub-schemata of these schemata.
It is our intentIon that the proedures here described cobld beninteL,.\\\
grated:into 'One of the MEMOD datahases. (Norman, Rumelhart, & LNR, 1975),
such as NOUNWORLD or STORYWORLD.
First, consider the form of a Description-Schema.2
4s2iThe use of-Pdotible curlY brackets"' .in this schema and in:
/ those below is intended to connote that only one of the elements enclosed...,- ..
-.in these brackets will appear in .any given instance of.the scher. Thus
means that either a, or b-or c. can-be used to convey whatever schema thiscurly bracketpair ocCurs in. .All.three Are expeCted to some extent whenthe schema they are part of has been activated. But.= more, than one..of them is possible.as an instantiation of the calling schema.
,12-
2 5
ba
DESCRIPTION (OBJECT).
IEXIST (OBJECT, LOCATIVE-PREPOSITION (OBJECT, LOCATION))
.is when
NOUNPHRASE (OBJECT),e
POSSESS ,(OBJECT, QUALITY)
PAR01'..(OBJECT, PREVIOUS-SCENE)
end. ,
Whatie the meaning of the Description-Schema? It means thee
a Description May be conveyed by any of three syntactic-semantic devices.
The DescriptionLSchema4is "satiafied" or "activated" if any one of these
three subscheMata are activated. The 11/-st of these syntactic devices is
simply a noun phrase whose referent is an object,, (There is, 'in addition,-
a restriction placed on this objectsee the last line of the Description-
Schemanamely, that the object be part-Of the preyious scene shoWn the
sUbjectireader.) The NOUNPHRASt schema has an internal sttucture of its
own, which is not discussed here but which,would,presmably, include
the full range of syntactic possibilities for noun. phrases.
The second possible syntactic-semantic'device for conveying a '
description, according to the Description-Schema, is a locative sentence,
which predicates a locative relationship between the object and some
loCation. The third device is a qUalitative sentence (such la "The lake
has an islanA" or "The car is small"). Again, the Description-Schema
can be aatisfied if the subschemata for either one of these devices is
activated. In any case, the restriction that the OBJECT mentioned in
these sentences must be part of the just-presented scene still holds.
Consider now the structure of another form-schema, the Predic
.tion-Schema.
-13-
2 6
$
.
. PREDICTION (OBJECT1, [OBJECT )3
is when
FUTURE (Proposition (OBJECT1,... OBJECT2, OBJECT3,.. .:.))
PART-OF (OBJECTn, PREVIOUS,SCENE)
4
end.
When t e Prediction-Schema is activated, the reader expects that the/cl
phrase to be presented will be in the future tense. The FUTURE schema
.must, then, provide for the fact that the sentende can use either the
will construction or the be going to construction to convey tne future
tense. The Prediction-Schema requireS that at least oneof the noun
Thrases of the sentence (or, rather at least one of the arguments of the'
proposition) must pave for a referent an object from the previously pre-.
sented scene.
Thus far we have considered only the-top-dOwn processing.,
.effects of these form-schemata. It should be pointed out that these-
schemata are also open to the effects of bottom-up activation. We have
discussed the fact that these schemata excite or activate their cdMponent
.sub-schemata when they/theMselves are activated (as a result of, say,
the set provided bytheinstructiOns). It should also be clear, however,
that the form-schemata can themselves be activated in A bottom-up; data-,
driyen fashion, when their own subschemata happen to become activated
independently. For example, if the FUTURE schema were to become activated
4as a resblt of .syntactic or lexical processing on a phrase ( ,g., the
.7)
phrase ia going to wOuld ord4 sult in the-activation of the
3The use .of square brackets to enclose some of the param,-c2rs
of a schema indicates :that those ParameterS have optional status. This
means that only bne of the objacta from the scene need be mentioned inthe prediction, but that*other objects can optionally be included. ,
FUTURE schema), then this alone would be enough to at least partially
activate the Prediction-Schema. Mutual activation works.both ways -(see
,Levin, 1976, for a discussion).
Form-Schemata in a Broader Context, Although it is undouttedly
an instructive exercise to develop atheory of form-schemata within the
restricted experimental context we have teen using, we would do well to1.
.remembef that real reading does not conaist of a sequence of tachistos-
copilly presented flasi-es; each one preceded by a context-setting picture.
In naturalistic reading. context is not ordinarily established by preceding
--pictures. Natural texts gre ordinarily long sequences of sentences, with
only occasional graphic supplements in certain types of texts. The differ-
ences between,ordinary reading and'the sort of reading which subjects do
in the experimeht we have been.discussing have several consequences fon
the application of the,schema theory of conceptually-driven processing
.to ordinary reading. One of these consequences has to do with the co-
herence or unity which is an important aspect of any well-written
lengthy text. Another consequence is related to the fact that the con-
textual effects which result in the activation of particular content-'
schemata must be ascribed not to the presentation of a picture, but
rather to the reader's ,pmderstanding of the-preceding textual material,
Considef first the cOheren& or ulity of most natural texts.
,The form.-schemata which were discussed above (such as Description and
Prediction) were designed to account in part for the order of the indi-
vidual words of the presented phrases, In real texts, however, there is
akpo-a partiaIly,predictable ordering of the phrases and sentences Of
the-body of the tekt. Tie intenled meaning of almost any naturally7
oCcuring te: c cetLuyb
-15--
2 8
grossly violated if, for example,'
the sentences c were to be put into random order. Rumelharr
(1975) has tr.:. ount for these facts about the high:Er-order
structure of leveloping,a special kind of form-sc lama to account
for the epiz., dcture of narratives which are characterized by
having primar lists. His"schema for stories" can be thoughi of
as a form-schi- guides readers expectations about the sequence
of ideas in a Hi text.
Ther presumably,be other types oL high-order form-
schemata for pnceptually-driven processing in the reading of
texts. For ex ___gney (1976) has suggested that textual materials
be classified f the four following types: narrative, explanation,
representation, ascription. Acceptingthis classification system,0
the only type o for which a high order text-proceasing schema has
been developed -c'atives XMandler & Johnson, 1977; Rumelhart, 1975;
Thornd3,4e, 197 -2e the processing of the other text types is pro-
bably of even ;!.tte _ significance for pedagogical applicationS, this
is obviously :d important field for future model-building.
Rigney propc :hese four kinds of information may be found separ-
1
ately or .int -'w' 1 in one text, that the knowledge derived from each.
-'.
. may contribu: .nceptually-driven motor performance, and hat dif-I ,
ferent kinds cf L.Jtor performance may depend upoh different amounts and
mixtures of .this knowledge. At issue here are the effects of,these
..- different kinds orinformation on text-processing strategies. To examine
this issue, it is necessary to characterize these four categories of
information in More del%Ail.
As
currently is
pointed out above, the interest of procedural semanticists
narratives. The ou,tstanding features of narrative
Information:- , Jy-related episodes, characters, and plot, 'set t1,
-16-
2 9
of informatiin,apart from Che other t:tree categories. The, reader
martative cam identify with one or mc: 2 characters anc vicariously
:,Ernce their (fmotions a tempo=lly-orz. 7-nd 77.Tents
smulations of :pf, epfdic flo,,J of lif fr-om whicb IL-ae readeT an,r
l'impressfon of I and persc7,l __rr,o1;Imerit. Nariati-/es
are written with :_:increte nouns.abot. :oncrete episodes, with
;flt Jrnical Vocabularies, and the narrativ .71eS us=ally are designed
sy reading. ReadinT, narratt-e text is , s ar. to 7.ne data-limited
r:f processing discsed by Norman and Ez11,row (1975) than to re-,
processing. That is, words easily understood; sen-,
ten::: structure is easily followed by Most ri..aders Speed and accuracy
of 77.pcessing are not normally limited by competition for the reader's
F=-,ssing resources. The presence of protazonists in the nartative,
-ne Disodic nature df tilt,: text, and the plot must facilitate-concep-,-
-driven.processing. These. features, combtned with high imagery-
descriptions. probabl:- facilitate retention. These characteristics
n:-...17rative also make this type of information attractive for research.
.Houlir-er, narrative information is infrequently encountered in.technical
=aiming or in the sciences, where explanatory, representational, and
>rescriptive information are more common.
We have jusC characterized the reading of some texts as data-
imited processing and the reading of other, more diffiCUlt texts as
: rce-limited processing, using the terms introduced by Norman and ,.
Bobr7w (1975)., Tosome.it may,seem,that we violate the spirit of these.,
r. since Norman and- BobrOw apply 'the 'data-limited' designation
to simplesensory' processes, such as hearing a signal in
mr_csi.,,;i115Wever, Norman 3obrow make clear that the termS
3 0
resource-e- 1 a-Limited re absolute, 'apply only to pL:J( sses
within a 7 :Llan- rl.age of resource alloclon. Thus, if we wE:re to
..conEider ion I, which people are sitc.i: usly doing oth,.r
resource-den1L::ling taisks, such as shadowing o :ape7recorded m==ssage,
as,well as text., then we would ex,De.: al: types of
Processing, ttIudiLr narrtI've reading, td ap73ear resource limit C..
What we are c.-..x;! 4ith here, however, is n=mad reading, fr.: whf.ch
subjects.are rac:: .:aneously_trving to oar:- out other complex :ogni-
rive tasks. st:. n in passing. that sate seen to be canable
of knitti tng nove7i but not while reading technical materials.)
In normal _-practicad ai.:21t readers can be said "nehave ifl .
a data-limited ff:a37 . while proces_ lig most nm-rative texts; (There are,
exceptions, pethhaps, Or the wo=k ct au:hors, such a, Tames Joyce
'but it is r_om. ur-7..aL- :hat such texts ran be sJmrly classifi as narratives.).
Performance_ par!-7......L:Larly if measured y self-ra,4ngs of undrstanding ,
or grasp of bc r.lot structure fr iikel tc be uniformiL7 aigh for
such texts: xas oE performance similar to thoSe menticue.:d above
for narratis Nrot." re2veal muCM poorer results for the same reeaders
with more te7.7 Lica: rEaterials, if, say, the same amount of reing time
per word were ed to the subject.
Explanian, a form of infcrmation Al,countered in E,I1 science,
is characlérizaC IT _exiCal items with highl- specialized referents,
often abstract ccmceTts,.. which are related IT causality more often than,
4BosTEi, biographer=f Samuel j7nnson, provides mi extreme
example of data-tImij'ed processingin7. readin:,;-. He claims to.ha7e foundJohnson weepiTlg bec,suse he could n turn tsi T'ages of,a book as quicklyas.he could r.2.J (Boswell%seetals to f. this wa.s.a tribute to.Johnson!s amacri:41 mencal faculties, '-ut it 0,.--ts possible to us rt it
:.was .-f:mtended ..T.CfcalHon more as a cimment o 17e dearth,of.Contz inhis 7eaaIng ml=er.,)
a_
by time to other abstract concepts. Rather
high imagery episcdes, there is a causally-b:
based on abstractions. .The vocabularies used
td flow of
--lence of descripticn;
anat_ms contain
technical terms that may be unamiliai to the and the.expository
style often is forbidding.
ProceSsing for coMprehension of lex- l.s.ems LTA explanatory .
infortation interaCts with processing for highe-1 --mantic structure.. .
-The reader has to work barder.to translate t:72 Into a paraphrase
equivalent rhat he feels he comprehends. ThansL, technical termss
muSt engage a different set of schemata thar reac. -T- a narrative, if
the reader decides to try to, infer the'defi :he term from the
context, or ifile decides to look in another ,Jur7-1 for its definition.'
Either of these activities becomes a major of the processing
strategy. Some relationships among explana :oncepts tend to be
confusing ;:o keep track of and to remember. .'!red opposites are
ekamples; positiVe-negative, D-1, high-low, 717-:]= carriers and
. majority'carriefs., forward biased-reverse Verbal explanation
in which obieots or.events:are.changingatat : one.npposite to
another as a. conseqUeaceof Other objects or .7e.nts changing state Cfom
one opposite to another often.are very-confsl-'ig en a, first reading.
Processing explanatory inforMatic: ean be mmre resource-.
limited than data-limited. Processing is me- lfkaly to be pushed into, -
a degraded,,mode, in which the reader does no f. fully urnderstand what he
is Teading.
'Information in theform of represtation describes features
of objects. Representational,information temds '.13 contain a high pro-
port n of graphic Material' o supplement tex:- Its semantic structuYe
3 2
,
:c,I1Ja. co be closely reLatec , the ',truccure of the c...) ing des-
so ahat in a pa-srti r tenic:.. domain, say LLc equip-
MEIT: amd svstems in the Nav . represet: ILtional-informatic Jr:: be organized
in cn,e a few standard fo:-mats. r some experia :. e these, the
reac -r ca develop schemata the strt.ctures that will :calitate
subeatuent processin mar_rfiel f.ais domain. 7..hea --hit the
object that is -he subject :he renres_entation k = well
defined physical and fumcticaL 1 stru:cure must also fai1 concep-
tualiydriven prccessim. O ct:s can be classifieL int: factilies based
on similar structures, and higer- eve: schemata cam be 1.arrw.J that
. allow predictions based on.names f objects and their or.
functional featLlres. Furthermore_ many of the schema orae object
)
be highly useful for proces,--limE representational =formation about
another -similar object, at the conceptually-driven level. Graphic
material in representational information would influenc.--., ;,-.rocesming
stn.tegies. Good pictorial information can quickiy onvq Foys_cal and
functional siructure. Block diagrams are presumably easily aomrerted to
higt:-level schemata. Examination of ItlocAtiagrams p-2bably-should
occur early in'a text-nrocessing stragegy.
Like explanaition, repnr.zaentation is likely to tmvo:ve abstrac
co=epts represented by lexical iIens linfamiliar to :he reade7r, and ti,au:s
temn-processing is likely to be 'rced toward t.reurce- Lid mo6Y.
Hcv7.-ar, representation, invclirih: as it does descriprion c- .:oncrete
obje=mS, allows the uS*e of conren.-schemata develope:L: from e!xver_iendirt
'tharaa ab4---cts, Which 'sho\uld familf..-tare conceptually-dr=ert
as, f-Or example, for Telar_ing fumatinnal to physical '-stm-cre,of.an
elec. tronic, device.
Prescriptive info7Etion i3 concern-led with procedures.
,a=r_fhes how to do E,T,
rules for &it::
It
iing. We use a broac. de_Anition, to inclu e
10.ng as well as step-by-:step instructi s_
comsider genaral ;--7..:es of nLis type,
7-aevel schemata. 7._ can- d Lreht beha
tnis wiew-pofnL th_e ime-being,
nfze tuat Scandura (297- makes a different
===tured learning.
The structure
Eltan-_ ',performance is comt,
crgand.red by goal structu
King ,ezci Langston, 1972).
-.2reseript-__ve
3d cf strings
once learned, as sources
ior in certain circumstances,
least, ilthdugh we recog-,.
of "rulet" In his thecty4
=ration is heavily tiMe-basa12.
relatively simple actions
(Rigney -19-59; Rigney, Towne,
Textual :nior tin of this tYpe may
desert:he goal srructures, ,:77 it may descri-L. ;:nysical actions, or it may
4
e a TmLxture. It, too, s often supplemenrndby picturesand diagrams,-
and ..tr may ccmtain, as -=;tatistical algorithms, worked out examples.
It :Is often the ca naz, fn follewing.prscriptions, it is difficult .
to ta..._1'wham an intrmeciate goal 17as been attned, or if some error
has T=eVented its arl-,414mmellt. Man.'machine in-,:arfaces,upon which 'al
taslis are performad n do not.- p=vide this :_nd of detailed feedback.i
For- eimple, wf7- ,I=ss-necks, aln error a long statistical compu-
tarfo= nay mot tmmedia-ely.
p-re:-..,/e information often.assumes tnat the reader already
th
/I
f-.-- at he can use tc complete the prescription
Lisuai: applies to ietails ._ procedures. Brown (l976
inted out exampIas of implieIj procec-ireS in prescriptions used
to tich imathematics.
The, goal r tures .tot :7:escr1p .rls can be ca:Isally as u'elL
3 4
as sequentially.relate... ';.:oal is to operatr, a device, say a ,
radar repea=er, the power MalE ,-Je turned on and otLix ccntrois mdst be. - .
set to put the repeater in =e 7trect mode. If tha ta7a.-=oal is t71-...,
2ompute the standard d-e,:Lat7-11 :7-77mm a set of observatioms, certain inter-?
mEldiate r,?sults must b, _irst. Different patieins o sequence
relatiomships within goal s=ct.:_res were described i igne7 et. al.,
(1971) for serial task:3 tc D: -,e.rformed on ml=m/machine interfaces.,
4
Simiiapatternsu o doubt exist In other goal structures in prescriptions.
aoththe secAlential and the caLz)al struc=ures of presortmtions should aid,t
conceptually-driven proessia.: of prescrirtive te=t- 1Procedural.schemata
from prior experience with otner prescri?tionS of a similar 'kind would-,
ugcl. Resource n1iitatio =,, for processing prescriritive text could
relate to lack of undrstanding'of technical telnology, t from gaps
in the irstructLms, .e., ±mpflait procedurelk The reao(2r mighl mot
detect these gaps u=7,-..1 tried to use the iritructions to Co sometthing:
Ance prescriptiOns c:,,i1O vary frmm iists of generi rii1.s -o expli 'it
itep-h57-step instr, t' processing them cc21.1d be _itherf primaril
Tasour-a-lirited or pr-. nartly
La summary, e have att,limted to di,stingul_sh am: four t es
tax=aal ; narrativ- exppnation, repreety=tion, and
7reorip=don, on the iasis of fea: ees that: would i=::-.;enc. conceptually-
l-:iven processing. These different kinds of informa:tiem contain dif-
.erent featu.es that impose processing. loads ,:ft ''tffererat :itructural.
Levels of And tliAt- dri proc:essineith t:.-c+iwzree. ).eing
1..mited (yr being :.:Aleolarce77-71-- i;rocessing narTatives would
rend, a= De 12ta.-11mted. is, all levels of tqe text are asily
understood 'processing can proceed at top speec to fill in s:czs of
`t?, 5
-22-
.1%
-existing schematic structures. Tte Otber three kind's of in:formation
y contain technical terms at the lexical level whce meaning.is
unknown to the reader- Learning the meaning of.Ulese terms by looking
them up outside the texr-takes time from processing the texr. If the
meaning cannot be found, comprehension of the text will be degraded. ,
The other three categories of text also of--:-.en are incomplete representa7
tions, explanations, or prescrfptions, because the wthor assumed Ale
goadeAmuld complete them by inferen,re or alreadr.,- possessed the know-
ledge or procedu*;-9.the or left cut. If these resources are not
possessed by the reader, text-processing will be resource-limited.
Strategies are likely tc be different for data71imited than for resource-.
limited processing. The latter is Ifkely to be more heavily data-driven,
to be slower, and to he operating more often In a degraded mode
ctually, is probably nor thc case that natural texts can,
in general, he classiVA simply as ar. example c.f cme or another of the
above four text tFpes. Mostnaturai texta of mc.rE -.7,an a couple of pages
probably toniist of a sequence mixture f these 1:xt types. For
example, am electronic trouble-slimoting manua be.expected to
include explanations of the functions of certain oil-cuits, representationa
. b
of the layout of the cfrcuits, prescriptions for actual troubleshooting
(e.g., niV7ic.i.ay:: difconnect the power while removing Cle case of this
instrumnt%" and 1).)s,sih1y even pihrratives "fl. there was a young
repairtor oarc,d Franl. whc 7hought Le would save time by not disconnecting
the liower befare removing the case of theAG-34', Of course,.it in
not always true that-natural texts ere a mixture c.-1 these text types. _
Nariative texts, in particular, may .6ometimes be ver7 extensive, even of
book- length, without ever being interrupted with prescriptions or explana-
tions. Nonetheless, pedagogica: materials in particular are probably
3 6-23-
e
-chz,..1racterized by mixtures of the foutstexttypeS. Schemata for,the
Ci-iiree text types which have not yet been modeled need -to be developed.
Cantent -Schemata
W1thin the simple and not-wholly-natural context of the
'tachistosC,:. 'c experiment we have4een using to discuss conceptually- (-
) 4. _
'driven sehemata, content'schethata are fairly straightforward. Content-.
-
schiemata can vary in tomplexity, just as form-schemata can. The simplest
sort of ccntent-schemata are those which are closely bound to singleA.
'lexical items (words). In the-experimental-paradigm, these.are the
concept sc-rlemata which are responsible for the subject's reccignition
that one object in a picture is a car, anotlier is a lake,'and so on.
1
Once these concepts have been activated (the idea of a car or Ihe idea
oE a lake), then the'names. for these concepts, "carorielakt," are_
.also activated, and are thus, to some extent, expected in the,subse-,
cimm7.t tachistoscopically 4esented phrase.4r)
We will present as examples of the-class.oP lexical-level
co7.:...:ent-schemata two ilistances, one fot the 'concept lake and one for the
concept Volkswagen. Consider first t e schemat.M4for Volkswagens.
VOLKSWAGEN,( )
iswhen
nME .,(xlidelkswagen")
CLASS'(x, CAR) . 4
CLASS (x, VEHICLE)
CLASS (x, OBJECT)
SHAPE. (x, ....)
SIZE (x,. ....)
COLOR..(x, variable)
etc.
Here the visual properties of .
Volkswagens are spe5ified
7
-24-
ft
DRIVE (PERSON, x)
SOMETIMES (NEED (x, GAS))
end.
,Here.the functional properties
of Volkswagen are specified
Presumably, it'is the visual properties af Volkswagens, as specifi
subjects' Nolkswagen-schemata,'which enable subjects to recognize he
approp,riate portion of the Figure asla Volkswagen: (It s not our
s.3
concern here to specify howlthis is accomplished; see Wi stbn, 1975 for
some recent work on a frames or schemata approach to visual recognition
, , - . _
.problems.) Once these visual information subschemata hav-E been activated,
.g
( .,.
,...
,
as a result'of visuarprocessing of the picture, they cause the schema
for'Volkswagens to become activated. This attivation re'sults in certain
expectations with respect to some of the lexical items in the subsequently
presented
/hrase, The only direct lexical activation as a resuit of the
Volkswagen-Schema i "Vorkswagen." (Perhaps we should have also included.
t °
in the schema the alternate name "Bug.") In fact, however, the activa-
V(.
v..
tion of the Volkswagen-Schema not only causes th'e word "Volkswagen" to
\ be expected, but also, indirectly, other lexical items. It does this byx
\
.
causing activation of other, lexical-level content schemata, which, of ,
4 . .
..
\
..
t tDurse also have their own namcs. For example, the schema VOLKSWAGEN\\
.
activates the schema CAR,- which has the,name "car." Therefore, "car" is
another,expected word in the phrase which 'the subject will seL after
the presentation of the picture.
In some of the example sentences and phrases mentioned earlier,
reference was.sometimes made to object& which were not pictured in the
figure shown to the subjects. For example, '.'the driver"-IC'an be-referred
ta. '
'to in a post-picture phrase. It is possible that such references are
kao '
33-25-
.
.
. more difficult for subjects to perceive in a tachistoscopically pre-
sented phraSe thau-refeiences to-objects which were seen. However, it
seems likely that they are easier td perceive than are references to
irrelevant objects, stich as "the popCorn." One possible answer to how
this mechanism of relevance is achieved is to be found in the content-
schemata. -Because'the Volkswagen-Schema includes a subschema which has0
a PERSON who DRIVES the car, there is some expectation for a reerence
to this'person, although this Maybe on1S- weakly activated.
It should befremembered that thenature of the "expectations"
we have been discussing for these Particular lexical items is necessarily
somewhat weak. There are a large number of possible lexical items,
.baed on the schemata activated by the depictions of objects in the plc-.
ture presented.to the subjects, so not a great deal of activation can
be alloted to any one lexical- "hypothesis."
Another activated lexical-level content-schema is the Lake-
Schema, which is respOnsible for the subject's expectation that the word
"like" may appear in the tachistoscopically presented phrase. Here is
the schema.
LAKE (x)
IS when
NAME (x, "lake")
CLASS (x, BODY-OF-WATER)
CLASS .(x, LANDSCAPE-FEATURE)
CLASS .(x, OBJECT)
.SHAPE,(x, irregular & variable)
SIZE (x, large & variable)
'COLOR (x, BLUE, GREEN, GRAY)
-26-
3 9
;
PoSSIBLE (EXIST (ISLAND, in x))
etc.
POSSIBLE (SAIL (BOAT, on x))
POSSIBLE (DRY-UP (x))
=-,
4
end.
Conent-schemata can, of course, be activated by other means
than the perceptilrn of possible referents of these schemata in a picture.
In normal re.e. ng of a sentence like "John asked his mother if he could
sail the bOat," the.Lexical-level content-schema SAIL will ordinarily
begin to be activated slightly before the schema BOAT (simply because
reading is normally a-left-to-right process). To.some extent, the prior
activation of SAIL.will facilitate the activation of BOAT, because the
schema BOAT is a possible component of the'schema SAIL. If the sentence
were "John asked his mother to pass the gravy boat," the lexical-level
,Content-scheinaGRAVY'would lick activate the schema BOAT, although it
would presumably actiLmtethe schema for GRAVY-DISH, which would,.in turn,
result in an expeciatton for the, lexical item. "boat," a possible name
for a gravy container. Anderson & Orrolky, (1975) and Anderson, Pichert,
Goetz, Schallert, Stevens-& Trolllp (1976) have demonstrated that,contexts
have the property of selecting'specific interpretations for the lexical
items that appear in them. They show further that it is these Interpre-.
tations (or instantiations of the relevant schemata) which seem igrbe
remembered, rather than the lexical items themselves.
k,More Comprehensive Content-Schemata. People reading texts
11
undoubtedly have other, more complex or more integrative, content-schemata
at ;:ktrk when they.are reading texts. Text understanding consist; of
-27-
4 0
MoatabstraCt
::Top-dovin,Schemata in reading 2-10-77
cr-
COntent Form
content-form dimension
Narrative schema
Psychopiment-report schema
0riSperling paradigm
article schema
44
Boats-sail-on.44 -4.)cn lakes schema
. Nostconcrete
Volkswagen schema
Figure 4
-Description schema
Examples of different schemata
4 1
-28-
more than stringing together lexical-level concepts. _Some simPle
examples of supra-lexica]. schemata are included in the above schema
for the Concept lake.. For example, the "POSSIBLE (SAIL (BOAT, on x))"
subschema conveys the notion that boats can sail'on lakes. There must
be a large number Of simple schemata like-this,. which,reflect people's
knowledge about possible relationships between objects in the world.-
Such schemata Can be activat'ed in reading in a number Of different ways.
The perception of a lexical item such as "boat" or "lake" whose concept-
schema participates in suCh a relational schema (1fte "POSSIBLE (SAIL.
'(BOAT,:on LAKE))") results in-the:activation of the relational achema,
to some extent.
Many.more abstract content-schemata are possibl.e. For example
4<-
a psychologist who is familiar, with the Sperling Paradigm of short-term
memory experiments.will experiente an actiVation of his or her Sperling--
Paradigm-Article-Schema when reading a report of.such an experiment.
At the level of greatest abstractness, the distinctions 'between form
and content-schemata become less absolute. The Sperling-Paradigm7ArtiC1e-
.
,Schema just mentioned surely has form characteristics as well as content
.characteristics (e.g., the activation of such- a schema should mean1
(:expectations for,a certain-format -- introduction,section<,.followed by.
Methodology,,followed by Results, etc.). The extremely abstract, compre-
hensive schemata proposed for narratives (Rumelhart,'1975; Thorndyke,
1977;'Mandlet & Johnson, 1977) have content aspects,as well as form.
Figure 4 gives some.examples of different types of schemata-
varying along two dimansions: abstractness and the content-form dimension.
4 2
-29-
1..
Experimental Evidence for the. Effects of Content-Schemata in leading.
What kind of experimental or observational evidende can be
.cited in support for the kind of content-schemata We have just sUggested?
One type Of experimental evidence, which supports the existence of the
sort ofccntext effec predicted by the lexicalLlevel content-schemata
is provided by Swinney and Hakes (1976). Earlier research (Foss; 1970;_
Foss and Jenkins, 1973) Showedthat the presence of a lexical ambiguity
in a neutral sentence resulted in a momentary increase in processing '
complexity, as measured-by-reactien time in,a phoneme monitoring task.
Swinney and Hakes showed that "at least some types of,,prior disambiguating
contexts 'can eliminate the processing load effect typically obtained:
following an ambiguity." (p. 688). Bedause one meaning for a poten-
tially ambiguous word was activated by the cd-atext, apparently only that
meaning of the word was activated when it was read: -This seems to support
the claim inherent in the model tor lexical-level content-schemata that
access to a concept for a lexical item results in access to concepts for
related words:
There is also evidence for effects, dde to content-schemata
with.large'r Scope thanlexical-level schemata. One of the:most'coMpelling
ltdemonstration experiments" that comes.to mind,is a group of reading
experiments done by Bransford and Johnson .(1973) and some of their co-
workers. In one experiment, for example, all mbjects were reqUired to
read a'brief passage (about 100 words), Half of the subjects,were given
the passage with ona.title;'half saw,it with.a,different title. The
title.was."A $pace.TriP to an Inhabited.Flanet." ..The.iecond was
,
"Watching a. Peace March from the 40th Floor.." This.was Od text which
followed the title::
4 3
-30
"The view was breathtaking. From the window onecould see the crowd below. Everything lookedextremely small from such a'distance, but the color- .
ful costumes could still be seen.. Everyone seemedto be moving in one direction in an orderly fashion
,and tpere seemed to be little children as well asadults. The landing was gentle, and luckily theatmosphere was such that no special suits had.tobe worn.- At first there was a great deal ofactivity. Later, when the speeches started, thecrowd quieted down. The man, with the televisioncamera took many 'shots of the setting and the crowd.Everyone was very friendly and seemed to be gladwhen the music started." (Bransfore Johnson,1973, p.412)
-As can be seen, the text makes'quite good sense when.understo from
the Niewpoint pf either title.5 But what does it mean t "make gpod
sense from the viewpoint of either title"? What does .this mean in(3
.
terms --7F the.processing the reader is doing as he.reads the text? From
the pori2-b\pof view provided by the theory of content-schemata the title
has th... effects of activating different cohtent-schemata, one haying
4to do with peaee marches the other with spacethips or science-fiction.
These.schemata are mid-level content schemata. They have much larger-
scope than simple lexical-level schemata, but they do not have the dePth
and complexity .(or many of the form characteristics) that larger schemata.
such as an episode-schema might have.
According to our theory; when one of these sChemata is acti-
yated, it guides processing of the text. Many of the sentences in the
text are evidently ambiguous. Yet the feeling one has in reading the
text, after first having read one of ttie titles, is not one Of ambiguity
sr All. Because one has formed a conceptual "set" which guides processing,.
.' 5n enenc
.
I fact the.5th ste, "the' landing was gentle and luckily *.. ,.
) the atmosphere was such that to special suits had to be worn," does not-
.'make very good sense fromtbe viewpoint of a peach march... This sentence's:Significance in: the experiment is diseussed-below. ',
-31-
A A
.each of the conc. Lntroduced by the sentences already bos a "slot"
to fit into in a ,xisting mental structUre, the activated Peace-March-
Schemaor Spacesh:1,7- chéma.
What cau,d be the format of suCh mid-level,content-schemata
in the schema framework-we have been using? .Here are possible structures
for a Peace-March-Schema and a'Spaceshi,07Schema.6
PEACE-MARCH
is when
LARGE (GATHERING .(PEOPLE))
ORDERL: (M)VE (GATHERING (PEOPLE down STREETS))
INTEND (PEOPLE, DEMONSTRATE (PEOPLE FAVOR (PEOPLE;PEACE).. to POLITLCIANS))
POSSIBIE (SING (ETOPLE))
(SPEAk (LEADER, to PEOPLE))
POSS:7LE (NEWS-COVER (MEDIA, 9)7.
. enc.
SPACESA -LANDING -
is
LAND (J;22ESHIPS, on ALIEN-PLANET)
P IBLIL (EXIST (ALIEN-CREATURES, on ALIEN-PLANET)).
PC:%.::113, (SIMILAR (ALIEN-;CREATURES,. HUMANS))
,POSSIBLEJBREAMABLE (ALIEN-ATMOSPHERE))
shoulti be remembered'that schemata sUch as thOse followingare Aesigned o conVey not Oe.autbors' conCepts for th particular ideas,
:but rather tha concepts-of. some Mythical'"average man:" hprahave beenpreVious attempts to aescriSe.-sOine imbortint mid-level co ent schemataby psychologists. Heider'a (1958) attempt to formulate a7"naive psycho=logy" is:.one example of this: The.prese t work, however, has the benefitof'a More explircit Model framework (nam y4A)rocedural.aemantics) than .
'did Heider'S, .
- The symbol i is used to refer to the activation of the_callingSchema itself, in this case,.the activated Peacp-liarch-Schema.
-
'
POSSIBLE (FRIENDLY (ALIEN-CREATURES, to HUMANS))
end.
What eVidence is there that the two different titles resulted
In different procesaingof the2text, aside from the introspective reports
. of readers that they havedifferent mental experiences when reading. the
same text with different titleS? Recall sentence mentioned in Foot-,
note 5, "The landing was gentle and luckily' -Lie atmosphere was Such that .
no special suits had to be worn." The contep introduced by this sen-
tence do not seem to fit naturally into a Pear.:-1,1archs-Schema, although'
/ .
they work well in a Spaceship-Landing7Schema. 717 a post-treatment test
of memory for the'text passage, those suhjemts wmotead the passage
under th 'peace march" title had significantly 7=rer memory tor this
sentemce than.did those subjects who read the pom7Txp.undet'the "space-
ship' title.8
In other Work (Barclay, 197-3; Bra:1st-D.3rd, Barclay & Franks,
1972; Bransford.& Frank's, 1971; Johnson, Eransford, Nyberg, & Cleary,.
1972)'Bransford'and his associates demonst,7ated that memory and depth
of understanding are closely related. We may consider It likely that
subjects who.tead the i)asdage under the 1=..arade" title'did not have as'
complete an understandingof the andmalous sentence and its significance,
in the framework.of:the narrative as did the subjects.in the othet grOup.
From theviewpoint of our theory of reading, tile reason fpr thiSdiffetence
. is,that the4ctivations ofthe'Peace-Match-SChema did not make any con- 7,
1ceptually-driven processing. contributiwto the underatanding,of-this
. ,
4"tifter hearing the passage, Ss weie asked to recall it. Most
sentences were recalled well except for.the'oneabout "the landing."
here was e*tremely low tecall.for this sentence,.and Ss noted that there
was one sentence-(i:e., about,a landing) that they could not understand.
Even when,presented with a "cue outline"-(e.g.,,Luckily the landing
and the atmosphere .), Ss exhibited very low ability to
remember what the sentence was about." --Bransford & McCarrell, 1974,
p.297.
46 ,
-33-
1
...particular-sentence during reading. For those sL..-bjectS who experienced
an activation of theeSpaceship-Landing-Schéma, however,.there was an
important conceptually-driven contribution to the understanding of this
-sentence.
. Schallert's (1976) work can be viewed a a replication.of the
.
underlying findings of Bransford & Johnson (1973) . 14th more nearly
precise control of the structure of the texts. Subjects read Aragraphs
which were Specifically conStructed to be ambiguous, each having two'. ,
coherent and distinct semantic readings. The two titles prepared for
each text determined which 'reading was appropriate. The results' show(I
that when subjects prncesed a 'text Iprmeaning, the nat.= of their
mornnries for the texrlicre partiadetermimed by the preceding title..
What possib1 -:. signifLcanc,., for'application to adult reading
strategies do the results of BransfDrd and Johnson'S experiment have,.
when viewed from the.perspective of our theory? They suggest that
maximum understanding, memory and efficiency in processing obtain
whenreaders have experienced prior activationof the.appropriate
'mid-level content-schema. One prosaic application of thia suggestion'
is that texts should be written with qpropriateheadings, that is, ,
those which will cause the'activation oT schemata which.have slots for.,
the major concepts presented in the portion.of.text following that
heading. Another possible applidation would be io train adult readers-
to make more effective uSe of headings. .For example, if a reader
thought briefly about each heading before actually reading a text, he
.'might begin the reading task with a very good idea.of the meaning of
the6text, due to the activation of a nuiber of content-schemata of
varying scapes. This seems to be an important,aspect of a puMber of
47
reading improvement courses, such as the Evelyn Wood Reading Dynamica.
program. This kind of application is discussed in detail below;
Still another way in which readers :an be brought to experience the
activations of facilitative c.tent-schemata is for them to read some
simple, short, prepared sl_mmary of the text first. Educational psycho-
logists have exi)lored te u e of such summaries ad rela.ted preparatory
maLterials, they call "advance organizers." For recent reviews of
this'literamure, see Paw & Waller (1976).and Hartley & Davies (1976).
It has beem noted .t:J.:at ne Of.the greatest problerx.With this area.of
resear , is the lack of a cohesive theoretical'framework
fur characterizini.; r comparing,particular advance organizers. It may
he :hat the theory .'or content7schemata we have described, when coupled
with a framework fo: producing summaries such as that proposed in Rumel-
hart (1975), could crovide Such a framework for analyzing advance
or7anizers. 4
Application to Adult Reading_ Strategies.
Rumelhart does.not deal with our major intetest here;.intentional
control oVer the reaC(ing processor. -.He has,described an automatic, pro-7
babilistic processor. It seems clear to us that, superimposed on this,
th6re must be a conttolstructure, an executive level, that Can assign .
'different Jcinds Of prOcessing taska to,the reading processor: Most" text
contains a.,variety of different kinds of information; it is &land of
1.z...formation fruit cake. The resources of, the reading proceSsar, can be
Ideriloyed to satisfy aVariety of different kinda of goals asSociated
with sUrface structure and_deep structure, and to satisfy one goal there
may be more than'orle processing strategy: For example, f we are read-
,.
ing for meaning, WE can continhe to.pick upinformation on successive
4 8
passes (many experimental studies of reading allow only one,pass),
,We cap set our reading .processor for different subsoals on each pass,
or for a maximum com:oreliension on ea:ch paSs. We can skip around, not
following the serial topography, Or we can follow it. We can review
our retention of what we just 'read, or nct, betwen passes. We can
regulate the Amount of processing capacity -we exT nd, from low-concen-
tration skimming to llighreoncentration reading. We can read with the
intention to remember, or with some short-term goal in nind that does
not require.long-term retennlon (e.g.,.looking up a teIephOne number).
The common denominator of al_ text procesaing strategies seems to be
limitedinput capacity. It is not ar accident that the information.
in texts is'packed in serial strings.
SuppOsinethat the theories odtlined above offer a reasonable
way for us to to.understanesome of the effects oE conceptually-.
driven proCessint in reading. Do they hav; anypplicability to adult ,
. reading strategies? We think that they do, and that, to _some extent,--
they help to explAin,theeffectiveness of the commercial reading
effectiveness courses approach, discuSsed below under Extractive-Multi-
Pass Strategies.
Text-Processing Strate'gies
We will group these as single-pass strategies and multi-pass
strategieS. In what follows, we attempt to delineate basic strategies
for understanding text. 'That, PS, we'assume that the reader has,the
task of understanding all or a selected part of a text passage. We will
leave for subsequent consideration the influences,of other orienting
tasksi noting in passing that the literature contains many examples of
the ef(ects of different'orienting tasks on word, sentence, and text
processing. One of the most sttiking is the.demonstration by AAronson
°-dbnd,Scarborough (1976) thattsubjccts who were told they Would.h6Ve,ta-
recall sentences word-for-word used different word-by-word.reading-
times.and'patterns than subjects instructed simply:to tomprehend the
sentences. During reading', recall subjects spent an average of 181 msec
more per word than comprehension subjects, and.their phrase structure
.
processing patterns were,markedly different. We also recognize that text
in textbooks often is supplementeE, by review questions, exerdises, or.
problems; and that the reader can choose to,use or be taught to use .
rehearsal, or mental imagery, or other learning. strategies.. For a
,pre4minary discussion ofthese issues, see Rigney (1976).
Single-Pass Strategies. It might be possible for a reader to
Start at.the,beginning of a-text, with the first word, and read it word-,
by-word straight through to the end, without looki-w ahead or looking
back. The reader.would attemPt tc comprehend the temt on one trip
through. However, eye-movement studies indicate thLt sf,:ricti word-by- .
word processing seldom occurs. Remders may look bac at parts of sen-
tences, or look ahead, or cross sentence and paragraph boundaries in.the
search for meaning. For experimental purposes, it m.._v be necessary to
force an appr'xiMation of striet 1ass processfng. A necessary
tondition is segmented PreSentation -of the text, -word-by-word, sentenee-'
bysentence, or fraie-by-frame, to prevent non-sequential vis451
'scanning strabegies. 'Auditory presentation often. is used, aithough this',
adds the requirement for phonemic processing, and drops the realuirement
for graphemic.procesSing. .
Strict single-pass processing canbe. forced, .
as in Thorndyke's study, by paCed presentation, e.g.c on,le line at a time
for 5 seconds. Even here, itseemS likely that studefs might san
.- back and forth on the line during the 5 seConds. Lets strict single-paSs
-177
50
, strategies are allowed by sequentiaL programmed instruction frames
which may contain more visible information, and. may require additional
processing in a frame. The student processes the module, answers a
question about it, and goes.on if cdrredt or reprocesses it if not. The
frame could be shown briefly and then reMoved.
There seem to btseveral things wrong atb singre-pass proces-
sing strategies. For eicample, they reduce tbe scope of the input for top-..
down procesSing, using the topography of the pl , and skipping about,
across sentence and paragraph boundaries to pick up clues to higher-level
structure.. Clues to this structure must be accumulated word-by-word
and sentence-by-sentence, before it can be.predicted from prior.experience
with.similar material.' This could-limit the contributiOns of conceptually-
driven Processing.
.Some kinds of...single-pass strategies.may be effective for pro- 1)
cessing some kinds of text. In conceptually.dense, technical material,
there can be many7technial terms that are not explained.
in the text...1 . .
The student MaY have.to search for explanations of these terms and re-
write.the sentences,.one by one, substituting'the explanations, befoie
.he can:comprehend the passage made up of the sentendes. This seems to
be a matter of translating the data into equivalent, analogical or meta-
phOrical forms, that can be used by-conceptually-driven processing . Thib
is one, way, of achieving understanding of conceptually denst text,
'Multi-Pass Strategies. Single-pass strategies discuSsed above
are *based on the idea that 'all the, information in the text can be ex-
tracted on one,trip through. It is the acquisition of ideas, rather
than tht acquisition of words that distinguishes%single-pas:
'multiple-pass strategies. The basis for multi-pass strategies is that
-38-
several passes could be more effective than a single-pass. Putting
text on pages and Ordering pages in sequence allows the reader to
select. apy part of the.text he pleases to process next. He can
cross over sentence, line, paragraph, and page boundaries any time heeP to
chooses, de can go back and reread a phrase or sentence he does not
'understand, or look ahead tO get some ideaof higher.levels of organi=
zktion, or ski& over material he.already understands or that is not
relevant to hs goal. He can selectively prOcess-FO satisfy some
non-semantic objective, ignoring the information in the text. How
should he use this poWer and flexibility of his text processor?
It seems to us that these strategies should utilize the
'power bf conceptually-dtiven piocessing in,the top-down, bottom-up
model. HeretOfOre, the implications of conceptually-driven processing',e
in text comprehension seem to have.bben overlooked. Yet, it is clear
,
that it is of overwhelming importance. Most of the time, when we are
e.
reading, we pre acquiring new information from data that cOnsists of
utterly familiar patterns of letters in words, words in sentences, and
sentences in paragraphs. Fiom whence-comes the pew information?
It may be useful to'distinguish:among the different kinds of
.multi-pass strategies on.the basis of how the text processoV is Used in'
each pass. Tentatively, N5e identify ekhaustive, ektractive, and selec-.
time types:
Exhaustive Multi-Pass'Strategies. The:exhaustive'Multi-paa'S .
sttategy concentrates on full comptehension on,each pa'ss.. We ail have
had the experience oT reading a passage, say a chapter in a textbook,
for a second or thirdrtime and eagh time realizing that we are acquiring
- more information. The-exhaustiVe.multi,pass strategy probably is fairly. ,
'5 2
1.
commonVused by students. It is simple to apply. ThS reader finds a
quiet environment, aits down with the text, andreads it through with
,\
as much concentration as possible, and then reads it again wiph deep
concentration, and so on. Management pf intervals between readings is
important...A-low long,should'they be? What should.go on in these inter als?
"Evidence in the literature is equivOcal, although the reader probably
'should not read similar material during the interval and the interval,
probably:should not.be.too long. The interval could be used,for a review
of what was read. This will revealjrregUlarities in recall. Recall is
likely to be good for top,level truèure ; not so good for details, at0
least for narrative Information (Thornd ke, 1977; Mandler & Jbhnson, 1977).
Review might uncdver gaps in knowledge that otherwise would not be remember-,
*eAfrom the first pass, but the review itself is likely to be selectiye.
Whether review is done or not done, the next reading is likely_.to be
different than the first. It will be.more selective and less compee-.,.
hensive, concentrating on parts of,the text reMembered to beAifficult,.4 .i. , .
..
,
,:
skimming parts remembered'to be of lAttle importance, or already known.i "
' What seems to happen asa cOnsequence of top-down, bottomLup'intar--
actionaduring.sthe preceding passja that confusion and some anxiety
will be generated over passages that were not, well understood: These
feelings7Will motivate the reader to direct his text.processor to Con-,.
centrat& on these confusing passages the next time. There iS a tendency,
then, for the exhaustive multi-pass, strategy to be conceptually-driven
'toward the extractive. multi-pass strategy on successive passes.
Extractive Multi-Pass Strategies. For non-trival passages,
ay a chapter in a textbook, onereading will not Suffice to store all. ,
0.N,
the information in,the shapter in LTM, or to make.all of it retrievable..
-40-4f,
5 3
from LTM. If interactions between top-down and bottom-up processing
tend.to 'make successive passes extractive, why not teach extractive
multi-pass strategies? The studenTght be taught to look for overall
structure, for terms he does not Imo for.summarizing sentences, for
style of exposition, for differentlands of processing,tasks, etc.
The assumption $s that relatively quick, extractive passes might accumu-.
jate contextual Information that mnuld be extremely useful Tor the top-
down part.of tho text processor to use in guiding procesAini. If the
, context could be established early in LTM, could the acquisition of
4etailed information then proceed on an extractiv'e basis? Multilass
extractive processing seems to work well for reading the-journals.
Studies n Cognitive Psychology,. Memory and Cognition, or any other
cientific journal haVe a standard format dictated by the editorial
,
poftcy. One multi-pass extractive processing Strategy used to under-,-
stand articles in these journals could be:
1. Read the title and sumMary.- Dothese descr,ibe what theeXperiment is about? If nOlt, read,atew sentences of'the,introduction for additiOnal clues.
2. 'Look for tables *or figures that summarize result§ andgive labels of main variables. .Try-to get some idea 404-Whet happened'.
. Read the final,discussionnr Summary. These,give more'clues'to what the author thought he was doing..
- 4. Reid part or all of the,thethods and procedures sectiontciget some.idea.of_theepisodic'strUcture bf theexperimeAtal design.
5, Look for answers.to unanswered questions or.passageA thatwill reduce confusiOn.
,Mentally reVieWryour understandint of what waS done,
found, And condluded.
The Evelyn Wood reading dynamics approaCh seems tobe the
leading example of an extractive multi-pass strategy. Successful'graduates
'
-
of the EW classes "read" texts atLa much faster rate than.they did
before ,training. (We use scare quotes here only because some might
object to the use of the term "read" whe:. the reader does not actua4.447
see every word in the text. For our part, we feel that this is reading
if the reader gets just as complete an understanding as would be the case
if every word had been read.) In addition, they usually seem to compre-
hend or remember more of what they read, based on Multiple-choice measures
of grasp-of-content developed by the EW staff. This seems like an anoma-
lous result, since we in psychology ordinarily expect to find speed-
accuracy trade-offs. Yet it is explicable, we think, if we understand
some of the EW techniques in termsbof the schemata theory outlined above.
One of the most important rules that EW graduates are supposed to follow,
particularly when they are reading technical material is to preview
the material. The previewing is done in several passes. First, the
reader pages through the text looking at major headings, picture captions, ,
and diagrams. During this process, he or she is to try to form hypotheses
about the major points made by the text. Then the reader makes a second
pass through the material, this tilAe-a slower and more detailed one.
Using the,hypotheses formed in the first pass, the reader tries to form
a number of partia4r independent hypotheses about the detailed points
made by ehe text. On this pass, he or she tries to notice vOcabulary
items or concepts which are unfamiliar, With the intentiOn of seeking the
meaning of these items during the actual reading of the text. If the
reader feels uncertain about the overall'structure of the text or a
the purpoae of some subsection of the text, a third pass or A parti
pass fs also performed.- Only when the major ideas are.firmly fixed
the mind of the reader does he or she finally "read" the text. As a
result of teaining in self-ferced pacing of eye movements, this final
5 5-42-
reading is done quite quickly, but it is not our interest here to
deal with the nature of the speed training, but rather only the compre-
hension training.
The effectiveness of the EW previewing techniques can be-`
understood in terms of thetheory of forM-and content-schemata discussed
above. The?-first pass can be thought of as an attempt to activate an
appropriate high-level textual schema, or, rather, a number of high-level
schemata of both types (form and content). If the reader was not aware
of the top-level form-schema approprix to the text (i.e., whether the
text was primarily a narrative, a p escription, a representation, or an
explanation), this pass should'activate the appropriate schema. In
addition, some top-level content-schema should be activated by the first
pass--the reader should experience an activation-of his psychology-
Experiment-Report-Schema When he looka through a short text and dis-
covers headings like "Methodology.," "Results," "Discussion," and recog-
-nizable psychological terms in the title and figure.captions. (This
example iS" true, of course, only in the case of those readers who are
familar with articles on experimental psychology). In those cases in
which knowledge of the topic matter can be extremely specialized, an-
even more detailed and explicit'top-level content-schema can be activated
on the first pass. Thus,'some psychologists might preview a certain
article and experience an activation of a "Sperling-Paradigm-Arti6le-
Schema," rather that the more general "Psychology-Experiment-Report-
Schema,". The important notion provided by the schema-theOry approach
to reading is that previewing can activate schemata-Which might have bene-
ficial conceptually-driven processing effects in reading.
The second pass, which is guided, to some extent, by the schema
-43-
56
activated by the first pass, shoulV result in the activation of some
more explicit, lower-level content-schemata. Some of these schemata
will be lexical-level content-schemata, activated by the presence of
particular words in the bits of text read by the previewer. The reader
naturally tries to fit these activated schemata into the structure pro-,
vided by the top-level schema.
During these passes through the text, the reader, is building
up a conceptual representation of,,the entire textin a sort of outline
form. When the actual reading of the teXt begins,he or she does not
need to see every word, because most of what the'text says is already
4
known. aeading is,then more a process of filling in the gaps in an
established\knowledge structure, rather than a process of creating an
entire knowledge structure from scratch, essentially in a bottom-up,-
data-driven manner.
Why is it, then, ,that stUdents of the EW method of reading
appear to be capable'of processing more information per unit time? We ,
would prefer not to believe that EW graduates have significantly larger
1-STM buffers or significantly faster-read-in and read-out of STM. What
other options are there? Perhaps the nature.of what is stored in STM
is different. Specifically, perhaps the EW graduate processes the text
in significantly larger chunks. Because the EW grad has deliberately
:ought about the activation of a large number of top-level and mid-level
schemata, the text can be understood in terms of these complex, integra-
tive concepts as it is being read. Whef/Fhis works, it is not necessary
for the reader to first store in memory each of the low-level, lexically-
bound concepts in STM in sequence, and only 'then to figure out what the
ounifying, integrative concept is that captures all of these.. Instead,
the higher-level concept schema is already activated, and the lexical-
5 744
level units need only be superficially checked to ensure that they fill
their expected roles in the integrative schema. -
Selective Multi-Pass Stragegies. Obviously, the type of
multi-pass strategy hat is most effective will be strdngly influenced
by the objectives of the reader and by the type of text to be processed.
No one'ever needs to learn absOlutely everything in a passage of text,
from physical topography to deepest deep structure. Less exhaustive
objectives ar the rule*. It is essential that the objectives be clearly.
speci;ried sinc they will determine the processing strategies that could
be useful.
In all text processing strategies'discussed so far,-the top
level goal was assumed to be achievement of some reasonable level of
comprehension of all the teXt. Of course, these strategies,could be
applied to a page, a chapter, A book. The segment of text has td be
defined.
A,more limited objective, is very common, and is particularly
important in relation to job'performance requirements. The reader's
needs for information from processing a text are established by a re-
quirement. to perform some task or function.. He must find information
in the text.that will help him meet bis requirment, and he.wiShes to
,learn only that information. He must be able. to:identify information
he does not want to learn aS well as information he needs. In snch cases,
it is particularly important that the learner not adopt'an exhaustive
reading strategy; since this is extremely inefficient when only a small
part of the total inf6rmation in the text is needed.
One.strategy which may work well when the reader has a limited
,and specific objective with respect to selecting information fromithe
-45-
.58
text is for the reader to begin, not by_reading some portion of the
text, but by thinking deeply about his objective before using the text.
By thinking about thE., -_,bjetive and relatinvittc what he already knows,
the reader activates numper:of content-schemata which should be.rele-
vant to.his objective. These activations may.include the activationof
some specific lexical items for which the cOntent-schemata are incoMplete.
If this is so, these lexical items-can serve to initi4te the fiist use of...
the index or table of contents of the text, if the text is provided with
these self-directional-aids. Even if the text is not so provided, the
-reader can begin his processing of the text by skimming, searching for
instances of these poorly, understood a'rms or semantically related terms
in the text. Hopefully, anlarea'of the text rich in such terms might
contain the solution to his problem.
...This type ofmulti-pass strategy is heavily conterned with
locating apPropriate information. By its continual use, the reader`.
would be expected to learn schemata relatil to text-searching strate-
gies. In the professions, the volume of the literature often forces
the use of this sJe,tive strategy, with the consequence that profes-
sionals are likely to have a good store of knowledge about where infor-
mation about topics is located in scientific journals and books, and
about who did the studies.k4-
Since the selective multi-pass strategy is a-valuablc tool in
job environments in which the reader must direct his.own text-processing
to aChieve learning objectives that will allow him to accomplish specific
performanCe requirements, this strategy is the subject of a more intensive ,
treatment elsewhere.
t./
-46-
REFERENCES
AarorOon, D. and Scarborough, H.S. Performance theories for sentencecoding: Some quantitiative evidence. Journal of ExperimentalPsychology: Human Perception and Performance, 1976, 2, 56-70.
Alderman, D. and Smith, E.E. Expectancy as a determinant of functiOnalunits in perceptual recognition. Cognitive Paychology, 1971,2, 117-129.
Anderson, R.C. .and OrtunY, A. On putting apples into bottles: A
problem of polysemy. Cognitive PsychologY, 1975, 7, 167-180.
Anderson, R.C., Pichert, j.W. Goetz, E.T., Schallert,, D.L.,.Stevens., K.V.and Trollip, S,R. Instantiation of general terms. Journal ofVerbal Learning and Verbal Behavior,'1976, 15, 667-679.
Barclay, J-.R. The role of comprehension in remembering sentences.Cognitive PsychologY, 1973, 4, 229-254:
Bransford, J.D., Barclay, J.R..; and Franks, J.J. Sentence memory: A
constructive versus interpretive.approach. Cogniti-ve Psychology,
1972,.3,. 193-209. V
Bransford, J.D. and Franks,..J.J. The'abstraction'of linguistic ideas:-Cognitive Psyehology; 1971, 2, 3317350.
Bransford, J.D: and Jbhnson, M.K. Consideration'of some probleMs of
comprehension. In W.G.. Chase (Ed.) Visual Information Processing.New Nork: Academic Preas, '1973.
Bransford, J.D. : and McCarrell, N.S. A sketch of a ognitive approach-
. to comprehension: Some thoughts about under ing what it means
to. Understand. In W.B. Weimer &,D:S.'Palermo s.), Cognitionand. the SymboliC Processes. Jiillsdale,-New Je Lawrence
ErlbaumASseciates, 1974:
iBrown; J.S. PerSonal,communication. ARPA Contracters eting, Boston
Mass., Novembr'1976.,- -
*
H.W. and Waller, T.G. ,Mathemagenic behaviors and efficiency ining from prOseaterials: Review, critique and recommendations.
Vey w of Educational Research. 1976, 46, 691-720.1V
Hartley,'.7.-and Davies, I.K. Preinstructional strategies: The role of'pretests, behavioral objectives, overviews and advance organizers.Aeview of Educational Research, 1976, 46, 239-265.
Heider, F. 'The psychology of interpersonal relations. New York: JohnWiley &;.-Sons, 1958.
Johnson, M4.;.;.Bransford, J.D., Nyberg, S., and Cleary, Comprehension.factors.in interpreting memory.for abstract and concrete sentences.Journal of Verbal Learning and Verbal Behavior, 1972, 11,'451-454..
-
-47--
6 0
References - continued.
Kolers, P.A. Three stages in reading. In H. Levin ana J.T. Williams
(Eds.), Basic Studies in Reading, New York: Basic Books, 1970.
Levin, J.A. Proteus: An activation framework for cognitive processmodels. (Working papers ISI/WP-2). Los Angeles: Information
Scrtnces Institute; 1976.
Handler, J.M. and JohnSon, N.S. Rememberances of things parsed: Story
structure and recall. Cognitive Psychology, 1977, 9, 111-151'.
Marslen-Wilson, W. and Tyler, L.K.- Memory and levels of processing ina psycholinguistic context. Journal and Experimental Psychology:Human Jearning and memory, 1975, 1, 584-591.
Minsky, M. A framework for representing knowledge. In P.t Winston (Ed.),
The Psychology of Computer Vision. New York: McGraw-Hill, 1975.
Norman, D.A. and Bobrow, D.G. On 4ata-limited and resource-limitedprocesses. Cognitive Psychology, 1975, 7, 44-64.
NOrman, D.A,, Rumelhart, D.E. and The LNR Group, E;cplorationS inCognition. SAy-Francisco: Freeman, 1975.
Rigney, J.W. Cfo, cognitive strategies for facilitating acquisition,retention, and retrieval in training and education (Tech..Rep.No. 78), Los Angeles: University of Southern California, BehavioralTeChnology Laboratoriei, May 1976.
Rigney, J.W. and Towne, D.M. Computer techniques for analyzing themicrostructure of.serial-action work in industry. Human Factors,
1969, 11, 113-122.
Rigney, J.W., Towne, D.M., King,,C.A. and Langston, E.T. Computer-Aided
Performance Training for Diagnostic Procedural Tasks. (Tech.Rep.No.70)
Los Angeles: University of Southern California, Behavioral TechnologyLaboratories, October 1972.
Rumelhart, D.E. Notes on a schema for stories. In D.G. Bobrow and A.1
Collins (Eds.), Representation and Understanding: Studies in Coroi.tve
Science. New York: Academic Press, 1975.
Rumelhart, D.E. Toward an interactive model of reading. In S. Dornic (Ed.)
Attention'and Performance VI. Hillsdale, New Jersey: Lawrence Erlbaum
Associates, in preds._
Scandura; ,J.M.. Structural approaCh td instructional problems. American -
ilblf
Psychologist, 1977, '32, 33-53.
..
challert, D.L.. Improving memory for prose:- The relationship between
'depth of processing and context. Journal'of-Verbal Learning and'
Verbal Behavior, 1976, 15, 621-632.
61-48-
References - continued.
Stevens, A.L. and Rumelhart, D.E. Errors in reading: Analysis usingan augmented network moeel of grammar. In D.A. Norman, D.E.Rumelhart, and the LNR Research Group, Explorations in Cognition.
San Fram-isco: Freeman,,1975.
Swinney, D.A. and Hakes, D.T. Effects of prior'context upon lexiCalaccess during sentence comprehension. Journal of Verbal Learning andVerbal Behavior, 1976, 15, 681-689;
Thorndyke, P.W. Cognitive structures in comprehension and memory of-narrative discourse. Cognitive Psychology, 1977, 9, 77-110./
Travers, J.R. The effects of forced serial prOcessing on identificationof words and random letter strings. Covitive P4chology, 1973,5, 109-137.
Weber, R.M. First graders use of grammatical context in reading. In H.
Levin and J.T. Williams (Eds.), Basic studies in reading. New York:
Basic Books, 1970.
Winograd, T. _Frame representations and the declarative-proceduralcontroversy. In D.G. Bobrow and A.M. Collins (Eds.), Representation.and Understa'nding: Studies in COgnitive Science. -NeW'York: _Academic
Press, 1975.
Winaton, P. The sychology of computer vision. New York: McGraw-Hill, 1975.
6 2
749-
ONR DISTRIBUTION LIST
Navy
4 Dr. Marshall J. Farr, DirectorPerSonnel & Trairking Research
ProgramsOffice of Naval Research (Code 458)Arlington, VA 22217
1 ONR Branch Office495 Summer St.Boston, MA 02210Attn: Dr. James Lester
1 ONR Branch Office1030 East green St.Pasadena, CA 91101Attn: Dr. Eugene Gloye
ONR Branch Office536 S. Clark StreetChicago, Il 60605Attn: Dr. Charles E. Davis
1 Dr. M. A. BertinScientific DirectorOffice of Naval ResearchScientific Liaison Group/TokyoAmerican EmbassyAPO San'Francisco 96503
1 Office of Naval ResearchCode 200Arlington, VA 22217
Commanding OfficerNaval Research LaboratoryCode 2627Washington, DC 20390
1 LCDR Charles J. Theisen, Jr.,MSC, USN4024Naval Air DevelopmentCenterWarminster, PA 18974
-Commanding OfficerU.S. Nal,ral Amphibious School
Coronado, CA 92155
63
-1-
Commanding OfficerNaval Health Research CenterSan Diego, GA 92152Attn: Library
1 Scientific Advisor to the Chiefof Naval Personnel (Pers Or)
Naval Bureau of Personnel400m 4410, Arlington AnnexWashington, DC 20370
Dr. Jack R. BorstingPrevost 8. Academic DeanU.S. Naval Postgraduate SchoolMonterey, CA 93940
1 Mr. Maurice CallahanNODAC (Code 2)Dept. of the NavyBldg. 2, Washington Navy Yard(Anacostia)
Washington, DC 20374
1 Office of Civilian PersonnelCode 263Washington, DC 20390
Superitendent (Code 1424).Naval Postgraduate SchoolMonterey, CA 93940
Mr. George N. GrAineNaval Sea Systems CommandSEA 047C12Washingt65 DC 23062
1 Chief of,Naval.Technical TrainingNaval Air Station Memphis (75)Millington, TN 38054Attn: Dr. Norman J.,Kerr
1 Principal Civilian Advisor for ,
Education and TrainingNaval Training Command, Code 00APensacOla, FL 32508
Attn: Dr. William L. Maloy
1 Dr. Alfred F. Smode, DirectorTraining AnalysiS & Evaluation GrodpDepartment of the NavyOrlando, FL 32813
Chief of Naval Education andTraining Support (01A)
Penszcola, FL 32509
1 Capt. H. J. Connery, USNNavy Medical R&D-CommandNNMC, Bethesda, MD 20014
1 Navy Personnel R&D Center-Code 01
, San Diego, CA 92152
2 .Navy Personnel R&D CenterCode 306San Diego, CA 92152Attn: Dr. James McGrath
A. AYSjoho.im, HeadTechnical SupportNavy Personnel R&D CenterCode 201San Diego, CA 92152
Navy Personnel 11C.D Center
San Diego, CA 92152Attn: Library.
-1 Navy Personnel R&D CEmterSan Diego, CA WL,2Attn: Dr. J. D. Fletcher
, 1 Capt. D. M. Gragg,.MC, USNHead, Section on Medical EducationUniformed Services Univ. ofthe Health.Sciences.
6917.Arlington RoadBethesda, MD 20014
Officer-in-ChargeNavy,Occupational Development &= AnalYsis,Center (NODAC)Building 150, Washington Navy YardCAnacostita)WashingtohimDC 20374
LCDR J. W. Snyder, Jr%F-14 Training Model ManagerVF-124-San Diego, CA 92025 6 4
-2-
1 Dr. John FordNavy Personnel R&D Center.San Diego, CA 92152
1 Dr. Worth-Scanland,Chief of'Raval Education & TrainingRAS, Pensacola, FL 32508
Army
1 Technical DirectorU.S. Arm,; 7:arc!- for
_:;ciences
1300 Wilson Blvd.Arlington, VA" 22209
At
1 Armed Forces Staff CollegeNorfolk, VA 23511Attn: Library
1 CommandatU.S. Army Infantry SdhoolFort Benning, GA" 31905Attn: ATSH-I-V-IT
1 CommandantU.S. ,t0-7:,y
ATTN: EA
Fort BenjaMim Harrison, F 46216--
1 Beatrit7e
Army k4Aeatch Institute1300 Wilson Blvd.
n 22209
1 Dr. Frank J. HarrisU.S. Army Research Institute1300 Wilson Blvd.Arlington, VA 22209
1 Dr. Ralph DusekU.S. Army Research In4itute1300 Wilson Bivd.Arlington, VA 22209
1 Dr. Leon'NawrockiU.S.- Army Reseatch Institute1300 Wilson Blvd.Arlington, VA 22209.
1 Dr. Joseph WardU.S. Army Research Institute1300 Wilson Blvd.
"Arlington, VA 22209
1 Dr. Milton S. Katz, Chief .
Individual Training & PerformanceEvaluation Technical Area
U.S. Army Research Institute1300 WilsOn Blvd:,rlington, VA 22209
I Col. G. B. HowardU.S. ArmyTraining Support ActivityFort Eustis, VA 23604
1 Col. Frank Hart, Directa;."Training Management InstituteU.S. Army, Bldg. 1725Fort Eustis, VA 23604
.1 HQ USAREUE & 7th Army-)CSOPS
.
-::AREUR Director of G.:ED
rU New York 09403
A.P.I Field Unit - LeavenworthP.O. Box 3122Ft. Leavenworth, KS 66027
Air Force
116.- Research Branch
AFMIT/DPMYPRandOlph AFB, TX 78148
1 AFHRL/AS (Dr. G. A. Eckstrand)Wright-Patterson AFBOhio 45433
1 Dr. Ross L. Morgam (AFHRL/ASR)Wright-Patterson AFBOhio 45433
//.
1 Dr. Marty Rockway (AFHRL/TT)Lowry AFBColorado 80230
,DCDR USAADMINCENBldg. #1, A310'Attn: AT21-0ED Library'Ft. Benjamin Harrison, IN 46216
r Dr. Edgar JohnsonU.S. Army Research Institute1300 Wilson Blvd.Arlington, VA 22209
Dr, James BakerU.S. Army Research Institute1300 Wilson Blvd.Arlington, VA 22209
-3-
6 5
1 Instructional Technology BranchAFHRLLowry AFB, CO- 80230
1 Dr. Alfred R. FreglyAFOSR/NL, Building 410Bolling AFB, DC 20332
1 Dr. Sylvia R. Mayer IT)
HQ Electronic Systems' DivisionLO Hanscom FieldBedford, MA OlTiO
Capt. Jack Thorpe, USAFAFHRL/FTSWilliams AFB, AZ 85224
1 Air University LibraryAUL/LSE 76-443Maxwell AFB, AL 36112
1 Dr. T. E. CottermanAFTWASRWright Patterson AFBOhio 45433
1 Dr. Donald'E. Meyer'U.S. Air Force
. ATC/XPTDRandolph AFB, TX 78148
1 Dr. Wilson A. JuddMcDonnell-Douglas Astronaucics
Co. EastLowry AFBDenver,, CO 80230
1 Dr. Willian(StrobieMcDonnell Duuglas Astronautics'Co. EastLowry AFBDenver0 80230
Marine Corps
Director, Office of Manpower.Ptilization
HOS-e Marine Corps (Code"MPU)BCB, Building 2009Quantico, VA 22134
Dr. A. L. Slafkosky. Scientific Advisor (Code R.D.-1)
HQ, U.S. Marine.CorpsWashington, DC 20380.
AC/S, Education PrpgramsEducation'Center, MCDECQuantico, VA 22134
Coast Guard
Mr..Joseph J. Cowan, Chief, Psychological Research Branch
(G-P-1/62)U.S. Coast Guard HeadquartersWashington, DC 20590
Other DoD
1 Advanced Research Projects AgencyAdministrative Services1400 Wilson Blvd.Arlington, VA 22209Attn: _Ardella Holloway
_1 Dr. Harold F. O'Neil, Jr.Advanced ReSearch Projects AgencyCybernetics Technology, Room 6231400 Wilson Blvd.Arlington, VA 22209
1 Dr. Robert YoungAdvanced Research Projects Agency1400 Wilson Blvd.Arlington, VA 22209
12 Defense Documentation CenterCameron Station, Bldg. 5Alexandria, VA 22314Attn: TC
1 Military Assistant.for Human ResourcesOffice of the Director of Defense-Research and Engineering
Room 3Di29, The PentagonWashington, DC 20301
1 DirectorManagement Information Systems OfficeOSD, M&RA A
Room 3B917, The PentagonWashington, DC 20301
Other Govenment
1 Dr. Vern UtlryPersonnel R&D CenterU.S. Civil Service Commission1900 E Street NWWashington, DC 20415
1 Dr. Andrews R. MolnarScience Education Dev. & Res.National Science FoundationWashington, DC 2.0550
Dr. Marshall S. SmithhisIociate DirectorNIE/OPEPA.National Institute of EducationWashington, DC 20208
1 Dr. Joseph L. Young, DirectorMemory & Cognitive ProcessesNationaL Science FoundationWashington, DC 20550
1 Dr. H. Wallace Sinr..iko, DirectorManpower Research & Advisory ServicesSmithsonian Institution801 N. Pitt StreetAlexandria, VA 22314
-4--
6 6
1 Dr. James M. FerstlEmployee Development: Training
TechnologistBureau of TrafhingU.S. Civil Service,CommissionWashington, DC 20415
1 William J. McLaurinRoom 301Internal Revenue Service2221 Jefferson David Hwy.Arlington, VA 22202
Miscellaneous
, 1 Dr. John R.-AndersonDept. of PsychologyYale UniversityNew Haven, CT 06520
Dr. Scarvia B. AndersonEducational Testing ServiceSuite 10403445 Peachtree Road NEAtlanta, GA 30326
1 Professor Earl A. AlluisiCode 287bept. of PsychologyOld Dominion UniversityNorfolk, VA 2308
1 Dr. Daniel AlpertComputer-Based EducationResearch.LaboratoryUniversity of IllinoisUrbana, IL 61801
1 , Ms. Carole A, BagleyApplication AnalystMinnesota E ucatiormi Computing
Consortium1925 Sather Ave.Lauderdale, MN 55113
Dr. Gerald V. BarrettUniversity of AkronDept. of PsychologyAkron, OH 44325
6 7
-5-
1 Dr. John BrackettSo4ech460-Ibtten Pond RoadWalthl, MA 02154
1 Dr. Robert K. BransonLA Tully Bldg.Florida State UniversityTallahassee, FL 32306
1 Dr. John Seeley, BrownBolt Beranek and Newman, 'Inc.50 Moulton St.*Cambridge, MA 02138
1 Dr.-Victer. BundersonInstitut9 for Computer Uses in
Education3,55 EDLC
Brigham Young UniversityProvo, UT 84601
1 Dr. Ronald P. CarverSchool of Education 4p,
University of MissOuri-Kansas City5100 Rockhill RoadKansas City, MO 64110
Jacklyn CaselliERIC Clearinghouse on Information
ResourcesStanford UniversitySchool of Education - SCRDTStanford, CA 94305
1 Century Research Corporation4113 Lee HighwayArlington, VA 22207
1 Dr'. Kenneth E. ClarkCollege of Arts & SciencesUniversity of RochesterRiver Campus StationRochester, NY 14627
1 Dr. Allan M. CollinsBolt Beranek & Newman, Inc.50 Moulton St.',Cambridge, NA 02138
Dr. John J. CollinsEssex Corporation6305 Caminito EstielladoSan Diego, CA 92120
Dr. Donald DansereauDept. of Psychology .
Texas Christian University -
Fort Wroth,"TX 76129
1 Dr. Rene V. DawisDept. of PsychologyUniversity of MinnesotaMinneapolis, MN 55455
1 Dr. Ruth DayDept. of PsychologyYale Univeesity
Hillhouse Avenue-New,Haven, CT 06520
1 ERIC Facility-Acquisitions4833 Rugby Ave.Bethesda, Mb 20014
1 Dr. John EschenbrennerMcDonnell DouglaS Astronautics
Company-EastP.O. BoY. 30204,
St. Louis, MO, 80230
Major I. N. EvpnicCanadian,Forces Personnel
Applied Research Unit1107, Avenue RoadTor14to, Ontario, CANADA
1. Dr.'Victor FieldsDept. -of Psychology'Montgomeri CollegeRockville, MD 20850
4
1 Dr. Edwin A. Fleishman,Advanced,Research Resources
Organization8555 Sixteenth St.Silver Spring, MD 20910
Dr. Larry FrancisUniversiey of IllinoisComputer-Based Educational
Research Lab.Champaign, IL 61801.
.A1 -Dr Frederick C. .Frick
MIT Lincoln LaboratoRooth D. 268
P.O. BoA 73 .
Lexington, MA 02171 6 8
-6-
1 Dr. Joha R. FrederiksenBolt Beranek.& Newman, Inc.50.Moulton StreetCambridge, MA 02138
1 Dr. Vernon S. Gerlach. College of Education
146 Payne Bldg. BArizona State UniversityTempe, AZ 85281
1 Dr. Robert Glaser, Co-birectorUniversity of Pittsburgh3939 O'Hara St.Pittsburgh, PA 15213
1 Dr. M. D. HavronHuman Sciences Research, Inc.7710 Old Spring House RoadWest Gate Industrial ParkMcLean, VA 22101
1 Dr. Dun-can HansenSchool of:EducationMemphis State UniversityMemphis, TN 38118
1 HumRleestern Division278$7 Berwick DriveCarmel, CA 93921Attn: Library
1 HumRRO/Columbus OfficeSuite 21's
2601 Cro3s Country DriveColumbus, CA 31906
'1 HumRRO/PZ.'Knox OfficeP.O. Box 293 .
Fort Knox, KY 401214
1 Dr. Lawrence B. JohnsonLawrence Johnson & Associates, Inc.Suite.5022001 S Street NWWashington, DC 2019
1 Dr. Arnold F. KanarickHoneywell, Inc.2600 Ridgeway,Pkwy.Minneapolis, MN 55413
,Dr. Roger A. Kaufman203 Dodd HallFlorida State UniversityTaltahassee, FL 32306
Dr.. Steven.W. KeeleDept. of'PsychojogyUniversityof: Oregon.Eugene, OR 97403,
Dr. Davis KlahrDept. of PsychologyCarnegie-Mellon UniversityPittsburgh, PA 15413
1 Dr. Robert R. MackieHuman Factors Research, Inc.6780 Corton Drive
Goleta,,'CA 930i7
1 Dr. William C. Mann 4-
1Triversity of So. Californiainformation Sciences Institute4676 Admiralty WayMarina del Rey, CA 90291
1 ,Dr. Leo MundayHoughton Mifflin Co.P.O. Box-1970Iowa City, LA 52240
Dr. Donald.A. NormanDept. f Psychology C-009University of California, San Diego,La Jolla, CA 92093
Mr. A. J. Pesch, PresidentEclectech Associates, Inc.,P.O.'Box 178N. Stonington, CT 06359
1 R. Dir. M. RauchP II 4Bundesministerium der VerteidigungPostfaCh 16153 Bonn 1GERMANY
1 Dr. Andrew M. RoseAmerican Institutes for Research1055 Thomas Jefferson St. NWWashington, DC 20007
- 1 Dr. Leonard L. Rosenbaum, ChairmanDept. of PsychologyMontgomery CollegeRockville, MD 2650.
Petrullo2431 N., Edgewood St..Arlington, VA .22207
Dr.,kenneth A. TolycynPCR Information Scientea Co..CoMmunication.Satellite Applications7600 Old'Springhouse Rd.McLean, VA 22101
Dr. Steven M. PineN 660 Elliott HallUniversity of Minnesota
75 East River Rd.
Minneapolis, MN 55455
6 9
-7-
1 Dr. Mark D. ReckaseEducational Psychology Dept.University of Missouri-Columbia12 Hill HallColumbia, MD. 65201
Dr. Robert J. SeidelInstructional Technology Group
HumRRO300-N. Washington.St.Alexandria, VA 22314
Dr: Richard SnowStanford Univers,itySchool Of. EducationStanford, CA 94305
1 T. Persis'SturgisDept. of PsychologyCalifornia'State University-ChicoChico, CA 95926
1 Mr. Dennis J. SLILivanc/o Canyon Research Group, Inc.32107'Lindero Canyon RoadWestlake Village, CA 91360
1 Mr. Walt W. TornowControl Data CorporationCorporate Personnel ResearchP.O. Box 0 -HQN060Minneapolis, MN 55440
1 Dr. Benton J. UnderwoodDept. of PsycholoeNorthwestern UniversityEvanston, IL 60201
1 Dr. Carl R. VestBattelle Memorial InstituteWashington Operations2030 M Street NWWashington, DC 20036
1 Dr. David J, Weiss \Dept. of PsychologyN660 Elliott HallUniversity of MinnesotaMinneapolis, MN 55455
1 Dr. Keith WescourtDept. of PsychologyStanford UniversityStanford, CA 94305
Dr. Claire E. WeinsteinEducational Psychology Dept.University of Texas at AustinAustin,.TX 78712
Dr. Anita WestDenver Research InstituteUniversity of Denver,Denver, CO 80201
1 Mr. Thomas C. O'SullivanTRAC1220 Sunset Plaza DriveLos Angeles, CA 90069
1 Dt. .Earl Hunt' .
Dept. of PsychologYUniversity. of WashingtorSeattle, WA 98105
Dr. ThomAs G. StichtAssoc. Director? Basic SkillsNational InstitUte of Education1200 19th Street NWWashington, DC 20208
1. Prof. Fumiko SamejimaDept. of PsYchology..Austin.Peay Hall 304CUniversity of TennesseeKnoxville, IN 37916
7 0
-8-