pir

-

. AMOUR? RESUME-

ED 151 778'

voc

CS 004 0187

AUTHOR Frederiksen, John R..r,

TITLE A Chronosetric Study of Component Skills in Reading..Technital Report No. 2. ? ,

INSTITUTION- Bolt, Beranek and NevmanvInc.,1 Cambridge, Hass.SPONS AGENCY Office= of Naial Research, Arlington; Va..Personnel

and Training Research.Programt Office. .

PUB,DATE i 15 ,Jan,catta4cT N00014 -76-C -0461NOTE -- , 45p.

--J,,-.

. .

EDRS PRICE HF=$0.83 HC-$2.06 Plus Postage..

. DESCRIPTORS *Cogniilie Processes; ICoaponential Analysis; 'HighSchool Students; measurement TechiligmesiAlodels;

;.. *Reaction Time; *Reading Comprehension; : *Readingprocesses; *Reading Research; Reading Skills;. Tile

BSTRACT 1

A'component skills model of reading is.'presetted. Onthe basis of the model, five, component factors are lypbthesized:grapheme encoding, encoding multiletter units, phonemic translation,'automaticity of articulationevad_depth otprocessing:in word

.

recognition. The'fit of the hipoifiesized,component factor model istested Using coLariance data for 11chrodometric measures; chosen to

;reflect separate stages 'of processing..The fit of the Structuralmodel:is found to be good (p=.2) : Three alternative models raredeveloped, ,each representing a siaplification of thee. general model; .

in each case the alternative structural odel is rejected. The*coaponent skills model accounts. for nearly all ot the variance insubjects' general reading a4lity as ueasured by standard tests ofreading cbmprehension. .(Author) ,

.

... 1

t. i

L

a'

4*********4******?*****************A******************i**************'44* Repodnctions supplied bfEDDS.are the best that-can be made *.

..*- from the original dotnsent. ,-'

.- .

*******************************************************4***************-

I

Ih

dt.

.

1

we.

CO ./.. 1J S DePARTMENT Of HEALTILEllUCATION &WELFARE

,

, NATIONAL INSTITUTE OP, 49 .

BoA Be;ariek and Newman. Inc.14... . -

EDUCATION,.

6"..\. R eport *N9. 3757,

THIS 00CUMENT HAS ISE4N REPRO -DUCEDi ----' EXACTLY AS RECEIVED FROM .. ,

1"4 THE PERiON OR ORGANIZATION OR IGI,N. itin ATING IT POINTS OF VIEW OR OPINIONS A. STATED DO NOT NECESSARILY REPRE

SENT OFFICIAL NATIONAL INSTITUTE OFri EDUCATION POSITION ORPOCICY7

C;)' a# 4La

O

A Chronometric Study tit4"Component Skills, Reaqing L

-

)r-

*IP

'1

r-I

John R. Frederikaen'

Technical Report No. 2

JanuliiAy 1978 .

N.

44

This research was sponsored by tile Personnel and TrainingResearch Programs, Psychological Seiences, Division, Office of

.Naval Research, 'undee' Contract No. N00014 -76 -C -0461, Contract.',Authority Identification /Number NR 154-386.

17,

- ,Reproduction in *hole r in--part is permitted for any purpose ofthe 'United. States Gover ment.

ApprOved fo'r public .re ease: Distribution unliMited.

2

4

A

1

UNCLASSIFIED

SECURITY CLASSIFICATION OF THIS PAGE OW411 Dalea,Entm*9

REPORT DOCUMENTATION PAGE.1. REPORT NUMBER

T'chnical Report4. TITLE Subtitle)

S. GOVT ACCESSION NO

A Chronometric Stu4yof Component Skillsin Reading

r

READ INSTRUCTIONS; BEFORE COMPLETING FORM

3. RECIPIENT'S CATALOG NUMBER

7. AUTHOR(s) / k

John1 Freddriksen

S. PERFORMING,ORGANIZATION NAME AND °ADDRESS

Bolt Beranek,and Newman Inc,50 Moulton StreetCambridge, MA.02138 :

S. TyPEor REPORT & PERIOD COVERED

Technical Report (No. 2)(1 Dec.1,226-1.Jan:1978)

PERFORMING ORG. REPORT Nut4BERTBN Report No. 3757

ON-TRACT OR GRANT publip,ERPO

NO Of4-76-C-6461 -

O. PA

II. CONTROLLING OFFICE NAME AND ADDRESS

Personnel and Training Research*ProgramsOffice of Naval Research (Code 458)ArlingLaa_VA 22217

14. MONITORING AGENCY NAME AODRESS(Ii Offhttont how Contro(lise. °Mot)

GRAM ELEMEN T. PROJECT. TASKA WORK UNIT NOMBERS

'54-386

12. REPO DATE15J nuary 1978

13. NUMBER F PAGES49

IS. SECURITY LASS. Pet Ode report)

Unclass fled

ISA. DECLASSIFI T1ON/DOWNGRADINGSCHEDULE

IC DISTRIBUTION STATEMENT (of this Report)

"Approved for public release; distribution unlimited.

fi

17. otsTRieuirfok STATEMENT (et the Abstract entered le Meek 20, it ttlitereitt beet Report)<

1'111 SUPPLEMENTARY NOTES

IL; KEY WORDS (Contfnr on tome* ottio if 'vireo/0m ow) Manta, y ehtek www*1 -

q .Reading Sk 1s,- Astesement'of,Component Skills, Individual Differences,

%Ireormatio friocessing, Cognitive-Processes; Personnel Assessment.. ,

-

30. ABSTRACT

A componmodel, f(II)EncodAutomatiRecognit

using goseparate

note on mows* elite it newsier, sod lieatIty by Woe* niothat) A

t skins model of reading is'presented.,..On the basis of theve coMpOnentfactors 4 hypothesized: (1) Grapheme Encoding,ng Multi7Ierter, Units,ITII) Phonemic Translation,(IV)it y of ArtiCulation,"alli (V)-Depth of Processing in Wordon. -ir.he fit pfthe,*pbthesized componentfactor model is testedariance:data fOreleven chronometric measures, chosen to refleCt:stages of processing. _The fit of the structural model is (over)

tornON or 1,4iovs4s IS OBSOLETE

SS, UNCLASSIFIEDhim FORMWI. 4

FAN211

SECIIRIlrY CLASSIFICATION OF TINS PAGE MINN/ DM" 1,10.10re-

3-,os

`

L)

UNCLASSIFIEDSECURITY CLASSIFICATION OF THIS PACIE(Whon Dal. Mowed)

L

20. (cont)

found to be good (p=.2). .Three alternative models are devitroped,each.representing a simplification of the general mosleVin each case the .

alternative structural model is rejedted: The component skills modelaceounts for nearly all of the'variance in subjects', general readingability,as measured by standard tests of reading, comprehension.

A

5

r

r

IMO

4-

to

UNCLASSRIED

SECURITY CI. ASSIFICATIO4 01' THIS PACE(Whon Data itnfond)

4

4

1

1

J

4

ac

Report- No, '3,7 57 Bolt Beianek and Newman Inc.

SUMMARY

A component skills model-of reading is presented. On the#-.

basis of the Model., five component fdbtors are hypothesized:

Grapheme . Encoding,. (II) Encoding Multi - letter , Unite, _

(III) Phonemic~ Translation, (IV) AbtoMaticity of A eiculation,

and .(V) Depth of Processing in Word Rebognition. The fit of the.

hypothesized component factor model is tested using covariance

data for eleven ehronomdtric measures, chosen fio reflect separate

stages 'of processing. The` fit of structural model is found, .

to be good (p=.2). Three alternative models,,are developed, each

represdnting a simplification,of the general model; in,each case

the alternative strubturall'model pis rejected. The component,

skills model accounts for nearly all of the variance in. subjects'. #

general reading ab ity, as measured by standard, tests of ; r eading

ibc5MprehensiOn. -4 I°

_\

_

y

. \,

. .

\

p

:.

Report No. 3757 Bolt'Bera60 and, Newman

ACKNOWLEDGEMENTS

7

.,,,-4This research was sponsored by ONR Contract' N00014-76-C-0461.-,

of.,....."

\The support and'encouragement 'Marshall,Farx and Iplity ..lialffi-.

and of Joseph L. Young are gratefully acknowledged: i would like

to thank -Marilyn Adams and Richard Pew for fruitful discussions

and Jessica

.,. - . I '

.

during many phases of the work, and Barbara Freeman

Kdrzon, who implimented the experimental design.

1

4

A

A

4

p.

A

,-

I-

.

t

, )Report No,. 375g

.

SUMMARY

L

4

Bolt Beranek and Newman ,Inc.

a'j

TOLE OF CONTENTS

ACKNOWLEDGEMENTS ',.

r

LIST OF SABLES0

ii

v

LIST OF FIGURES .. tvi.

..

. .,, .

I. INTRODUCTION 1i

Appro.aches to Validation 1 3

* .II. COMPONENT SKILL MEASURES 4

Theoretical Model. 4 ';,. , 7 -

Experimental Tasks :86.

1. Letter Matching p \ 8

242. Bigram Identification ' A 11 .I

.1 p.. ,

3. Psieudoword D oding . '12

4. Word Nading. 13.

Relation'to .Hypothesized Component Skills 15)

*

III. EVA UATION1OF THE COMPONENT SKILLS MODEL 18

.Metho 18.

c,

Sul ect 4 , . 19.N.

w 19Results

\l.. 4%9Ev4 auation erAltnative Structural'Models 22

L

Testing .the Ex ernal iialidity of the Cbmponent-SkillS Model. 25

Chronometr4.c, Measures ..( '27

. =Reading Test Measures. .. 28.4

, 1

IV. SUMMARY AND CONCLUS ONS .;) .

%

0 f

$

c

?9,

ReportReport No. 3757

1_Bolt' Beranek and NeWman Inc." °

LIST OF- TABLES

'table 1., Deqoding under.Two Levels of Perceptual Encoding...,.. .'9,

Table 2. _Variables used in the ComponeAlt Skills Analysis ')\ 1

of Covariancek Structures

Table 3. Definitions ofCompOnent Processes Hypothesized in

the Analysis of Covariance Structures.

Table 4. Maximum Likelihood Estimates of'Factoi Loadings'

ro

\16

and,epiguenes es EXperimentar Variables lt

Table 5.'*Maximum Li lihood Estimates of'Inteycorrelations

7-gable 6. Test of Fit for Three Alternative Hypot

the Covariance Structure.....

Table ..Loadings of Criterion Variables on the Com

among the Fac o-r) 4.1.4 21

II eses about-

Skillp Fac ors

9,

nent

23

2.6.

Report No. 3757'

t *LIST 0F FiGUREB. y, .!

Figure 1. A schematic renaeripg,of the processing model

representih component skis -f in reading.

`Bolt Beranek and Newman -Inc.

Figure 2. An illustr tion.of the structural organization that

is implicit in the perceptual encoding of mufti- .6

letter units

I

5

7

,

.

se

I

Report No.., 37571

Bolt Beranek and Newma'nj Inc.

,

A CHRONOMETRIC ,STUDY OF ';COMPONENT SKILLS IN READING'/

I. INTRODUCTION.1 .\ I li .

\,1

,.

Psychometricians. have long sought to develop skill' easures:

ering the. repertoire of human cognitive abiliti/es cf.

.Thu tone, 1938; 'Thurstone Thurstong, 1941; Frenc 195;

Pren Ekstrom, and Price, 1963; Guilford,'1967;- Cart 11,1974).. -

The goal, has been to build tests of infordation-hancl ng skillthat represent particular methods for processing infor ation, but

: . I

that at theksame time have applicability across a vat-

ety oftask

environments.' .While this early work on cognitivea d percept

abilities is 'in many ways° compatible with mo e,n cognitive. . .

ssychology its effort to, distinquish.componn procews in

human skilled performance, the historical J emphasis upon

cross.asitUational iliformation-processing abilit es has limited.'

the utility of such measures in the analysis, the particular

tcomponent skills that are accjuired in beco g proficient within

\ .

single -task dOmain such -as, that of readieg,

..

4-

.11

In an effort to,de7lop measures that are diagnostic of the,

sources, of. reading-disability among naval recruits, we have been.

engaged in' a feries of 'stud,ies of individual differences rin the

'compon'ent skills Anvolved in reading. The general goal of this 7

has_ _been to develop a set of/component skallmeasur's that

represent the particularinforMation-handling processei.<used in

-1-

1

Repopt'go..3757 Bolt Beranek .and Newman'Inc4.

__.reading..r.. as- they are conceptuadizeOdn tirreilt theories Of the

reading' process: .:These ,include skills involved in the,

tranglation .of orthographic patterns into "sound" patterns and

the accessing of lexical Anformationi as well as perceptual -

skills of pattern recognitionqand\encoaing. .A as

been to explore the potential offered b a chronometrid approach4.

,. .bto the measurement of coMponentaiskill8 in reading-. There are a

number'of reasons why the-lheasurement of 'proces's'ing times, ma

-,provide en important tool for the assessment of skills in young.

.adults..

First, it is difficult to generate errors in such : basic.

. , .

,skills as letter' identification, phonic analysis, and the like ind .\\* .

mature, subjects. Yet, ,individual differences in skill may still.. -J

be apparent in their- processing efficiencies. Second, stu'd'ies of4 la ,

1 gwreaction times in human information ,processing 'have served

.\

.

experimental psychologisks well in their efforts to\buad preciseexperimental

,

models for. reading. In particular, the subtractive method for

Ns analysing reaction times -/-(RTs, hap-? proven its value,- as ab .

technlque- for deriving measurementb,that reflect a-single locus

a43'4information processing. an the'saktractive inethod, the

diffe -nce \in RTs is `calculated far experimental conditions"tha

vary'in t cessingload they place on some single. processingy

subsystem. 'fferenCes (or contrasts) then prov'ide a measure

of the relative-d' \iculty in processing under .the contrasted

conditions. . .With chreful, choiCe

12

of conErasts, among-. 4

I

.

'

Re-pciit NO. 37'57 Bolt Beranek,and Newman'Inc.

experimental conditions, it has teen-our hope that %measurements

of _component processing skills cpn be derived.0

Approaches. to Validation

The assertionj_that a particular RT contragt represents a

designated component skill must in thf firgi case be backed up by

experiments. designed to establish the construct validity of the,/

particular constrasts.- Thus, the first source of'information2

concerning the validity.of component skill measures comes from all

analysis of the indtviduai experimental taSkS from which the RT

contrasts are derived,- In this analysis, variations in

experimental conditions must, be shown to yield the expected

changes in response times as required by theory.

second source of information leading to construct

validation*reSulft frpm a comparison of measures derjyed, from

,

different experimental contexts:* F15om' a set of experimental

tasks, several measures are "derived for each 'hypothesized

component 51ocess, each one based upoh a separaig constast among

Ts for a different set of, experimental codditiens. A

theoretical' prediction can then be made 'about the relationships

1

among these skill measures: alternativd measures of a designated

compOftent_ekill ar&hypothesized to form a Common- factor-that is,0.-:-

digtinct frod the-faatars..formed by other ,component skills. Note- .__ !,- ,-,., , .0- -

that- it is thg:.hig6Aegrep of specificity about the component. . - . ,

,skills,measured by. :the chose'/-.,n

,

RT contrasts that allows us to-1, --,-

'10

4

Report No. 3757

generate

Bolt Beranek and Newman- Inc.r

and test a 'specific hypothetis about the factor

structure_ underlying our set of component ,skill measurements.

And,, verification of this .hypothesis will permit us to conclude

with confidence that-the cOmponerit skills; derived form our model

of reading do in 'fact represent the ,postulated sources of

individual 'differences among readers.

Finally, the role of component skills , in establishing an

indMdual's general level of reading ability can benvestigated,

by using the component skill factors to predict other, moree

.general measures of reading performance. This.provides us with a

third source of 'validating information: the evidence that4

particular component skills coht'ribute to skilled reading as.

Measured by conventional. tests of

comprehension.

reading

II. COMPONENT SKILL MEASURES

Theoretical Modell

ability lnd

The theoretical model guiding the selection of component

skill measures- is illustrated in Figur'd 1. The' model

disinguishes. four main processing levels: I. Vttual Feature

Extraction, II. Perceptual Endodina,' III. Decoding, and

IV. Lexical' Access. Perceptual Encoding is further subdivided. .

, into a component representing the encoding of 'individual

graphemes and a component representing the encoding of visually

familiar, multi-grapheme units- (e.g., :SH, ,ING).r

-4- -

14'

Ui

VISUALDISPLAY

VISUAL FEATURE EXTRACTION

I.

PERCEPTUAL ENCODING

SINGLE MULTI -LETTER UNITS ! LETTER UNITS,.

1. LETTER MATCHING TASK

MATCH VISUALPATTERNS

2. LETTER .IDENTIFICATION TASK

DECODING

PARSING !PHONEMIC IARTICULA-GRAPNEMEITRANSLA-1,4=2:ARRAY I TION I -

I!Nu

RETRIEVE'LETTER NAMES.

3. PRONUNCIATION TASK

LEXICAL ACCESS USINGAVAILABLE CODE S

INITIATEARTICULATION PSEUDOWORD

(WORD1

PRONOUNCE

4. WORD NAMING/LEXICAL DECISION

LEXICAL INFORMA-TION RETRIEVAL

f

USE OFTEXT MODEL

(SEMANTICCONTEXT/

LEXICAL MEMORY

AATICUt.ATORY

SEMANTIC

1 t

7:/

to0rtz0

PRONOUNCEWORD /MAKE to

LEXICAL 0DECISIO 1-1

rt

°

4Fig. 1. A schematic tendering of the processing model representing component skills in reading. Four processing levels

are distinguished: Visual Feature Extraction, Perceptual Encoding, Decoding, and Lexical Access. While these.processes are hierarChically arranged, initiation of higher-level operations does not await completion of prior

' operations in the hierarchy.' Decoding can thus be initiated on the tikas ca (a) independently-encoded graphemed,or (b) multi-grapheme'units. LexiCal access can be baded upon any °Lathe following input' codes: (A) visualfeatures, gl) independently encoded graphemes; (C), multi-grapheme units,(D) a parsed grapheme array, _(E) a ,1

5 k phonOlogical/phonemic translation, (F) a. speech contour, or (b) semantic/syntactic.constraiints on word identity.Experimental tasks 1-4 are thought to require different characteristic depths of processing.

z

sv

H0

. . ,

Report No. 3757Bolt Beranek and Newman Inc.

Decoding is divided into processes of parsing (Spoehr & Smith,.

1.9731, phohemic translation, and articulatory Programming.

A general featui,e of the model is' the notion th4, while.

these processes are hierarchically arranged, the initiation of ..

I..

higher-level operations doeS not necessarily await completion of-.

. ,.

priOr operations in the'hierarchy. Thus, Lexbal Access can be. I

4-* %

initiated on the 'basis of any of the following input

representations: (a) a spatial, distribution of visual features, I

(b) an array of independently encoded graphemes (e.g., T R A I "N

IlI N G), (c) encoded', overlapping multi-letter perceptual units,

"..\

as in ( (TR) (( AI)N)nI(NG)) (see also Figure 2) , (d) a "parsed,,

IA

. .grapheme array. (having a form that may be similar to that

illustrated in Figure 2) , -(e), a phonemic translation of the I

orthographic pattern, as in tner n T0, or Sf) a speech contour,

having assigned stress-and intonation. -Input,representations a-f

represen t differing depths or degrees, of processing} .p(ior to j

lexical access.1 In a similar .fashion, Decoding can take place

on the basis of (a) a set of.independently:encoded graphemes, or

(b) encoded, multi-letter perceptual units. Note that, according

to the model, the,demands placed upon the decoding component are

To handle reader' use of 'context in lexical retrieval, an'additional. input code (g) represents semantic/syntacticconstraints based 'upon a contextually- derived model of discourSe.However, skillsrinvolved in the use of context are not -includedin the present set of experimental .measures and will not beconsidered here. ,

-6-

17 ,

r.

o f a

Report 'sib. 3757-\

0

4

1.,

0 4

Bolt Beranek #nd Newman Inc.

r41,,

R.

Figure 2. An illustration of the structuraIorganizatiOnthat is implicit in the perceptual'encodin of:

the multi- letter units 1TRY,'(NI), (N), (I)

(NG) , (AIN) , (ING) , and (TRAIN).!

-7-/.

ort Na. ,3757

-a

Bolt Beranek and Newman' Inc.

greatly lessened when'the graphethe representation is made up of

multi-letter units having functional utility for decoding,,. such

as affixes, double-vowels, consonant clusters, and the like, as

illustrated in Table 1.

I

I

Experimental Tasks

component skill measures that are referenced to part cular

stages of processing have been derived from four

tasks:

experimental

1. Letter. Matching. .In the letter matching task, the

subject is shown a br4ef (50 msec) display containing a pair of

,

letters that (a) have the same name and fdlm, (AA, aa) , (b) have

similar, names but differ .in form (Aa), or (c) are .totally

different letters (Ad, ad, .AD). The subject's task is to

indicate Whether the letter names ale thesame or, different by, .

pressing an appropriate response key.. Two RT contrasts are

derived from this task: Speed in Letter Encoding (Variable l in

'Table 2) is measured by subtracting the mean RT for phySically

similar letters (AA, aa) from the mean RT for letters' differing

only= in case (Aa, Dd) (Posner & Mitchell, 1967). Facilitation in

Encoding Jointly Occuring Letters (Variable is measured byf

subtracting the RT for letters differ-ins only in 'case (A Dd)

from ,the RT for letters that are completely different (Ad;-)D)..

This RT comparison measures what Posner, in his ,later. work 7has

-8-

19

.1

ReportNo. 3757

TABLE 1

Decoding under'Two Levels of Perceptual Encoding

Bolt Beranek end Newman Inc.

Procesb

I

Perceptual" Encoding

Single-Letter Units Multi-Letter Units

Stimulus SHOOTING SHOOTING

/Encoded Visual Units

Decoding': Parsing

Grapheme Array

Decoding: PhonemicTranslation

Assignment of Stressand Intonation

S/H/0/0/T/I/N/G

I.

SH /OO /T /ING

SH/QO/T/ING

S

-§-20

0

o'

.00

torlatmw. OM.

` Report No. 3757`'

0

,

Boit Beranek and NewmanInc.. 7

,TABLE

ariables Used in

Skills alysis of,C variance

2

the-Component

.Structures*

,

VARIABLE-6

Results'.4.CODE TASKof ANOVA

peed in Letter Encoding:\.. Letter Letter .05'RT for disbimilar cases'(Aa)minus RT for similar cases - .

Encoding Matchihg

(AA, aa)

2. Scanning Speed: ,Increment in Scanning Bigram p<.05RT per letter position. Speed ./Identi 1-

cation

3. Facilitation in encoding jointlyoccuring letters: RT for'dissim-ilar letters (Ad) minus RT forsimilar letters' (Aa).

'Percept. Letter .

Facilita- Matching,tion

4. Bigram Probability Contrast: Bigram Bigram p<.05RT(Low Prob. Bigram) minus Proba- Identif i-RT(High & Middle Prob. Bigrams) bility cation'

'5. Array-Length Contrast: Increa e. Length:, Pse4word-in RT for each added letter. Pseud. Decoding

Syllable.COntrast: RT for Pseudoword2-Syllable minus RT for 1- Syll'. Pseud.Z: Decoding

7. -Vowel Complexity Contrast: Pseudowo drRT for -vv- minus RTfor

Pseud . Decoding

8. Syllable Contrast (as aboVe, but'for vocalization durations.)

Syllable: Pseudoword p<.01Rseud.' Decoding ,(Dur..)"'

.%9. Vowel Complexity Contrast: (as, Vowel : Pseudoword p<.10

'above, but for vocalization Pseud. Decodingdurations).

a0. Percent drop in Decoding Indica-

ur.)

A% eroding Word

u.

tors for HFW and Pseudo:(Sum 5-9 Pse d.r Namingfor Psaud. - Sum 5-9.forHFIWSum .kF5..9 for Pseudowords). /

-0 ,

11. Percent Drop in Decoding.Indica-- ACieCoding Word'tors for HFW and LFW: (Sum 5-9. LFW-HFW 11.' Naming-

for"IFW - Sum 5 -9 for Hill)/1'

-

4- (Sum-5-9 for LFW).-; -

* r

All comparisons are for mean response titles unless.0b4erwise noted.+Values oDthe variablp/d4fessubjects at four ;reading levels atthe indicated sionificance-level.

tg. 4

-10-

Report No. 3757 .

o

Solt Befanek and Newman Inc.

termed -category facilitation (Posner A Snyder, in press). These

two measures

subdivisions

graphemes, an

are thodght to refer respectively to the two

, r

'of .Perceptual -Encoding endaing of individ.tial

d'encodinq o4 multi-grapheme Units,P

NO'

2.. Bigram Identification. : in this

shown 'a' 4-letter array, preceeded

lc,. the. subject is

flowed by a 300 msec,

pattern mask (e.g.,.#04, followed.by'SHOT, and that followed by

####) The actual stimulus array varies from trial to trial: On

a third of the trials, the stimulus items. are familiar English

A words,' while on theremaining trials the items are presented with4

two letters masked so.thatitonly a s7ingle Pair of adjacent letters

bigram) 'is "y0ible (e.g., SH144 *AWL-14TH). Further, the

bigrams are chosen so as to differ in location within the item

(posi ions 1-2, 2-3, or 3-4), frequency of occurence in-English

`(e.g., H. [high] , GA 4middle], i LK [lowl),:and likelihood Irc.

,

()touring in their' presented position-within a fou4-letter word. \

(e.g.,'Tillit high -versus ITHI, [low]) '(cf. Mayzner & Tresselt,'

\ ..,

[ APP

1965). In all cases, the subject's task 'is to report'the letters.

,

that he can s e as quickly and acCuratelY as possible. The,

response measure the RT measured ,from the. onset of the7

stimulus item to th= onset of the subject's vocal report of the

letters (Frederiksen, 1'78). Two measures are derived from Shis.

experiMent. Scanning - Speed (Vatiab4e:. 2) is measured by1.

./subtracting the mean RT for bigrams presented in positions 3-4,

2 A :

Report No.:,j3757 , -:,

a

fromb,the 'Wean fot bigrams presented in pbsitions 1-2 and then

.,.I

diliding by 2. This gives the increment in RT for each shift to-

the right "in letter position. TheBigram Probability Contrast' 1

,\

(Variable 4) is measured by subtradting the Vtli for high. and

. . Imiddle probability bigrams from that for low probability bigrams.

This variable gives the penalty, -in processing time brought by

,reducing the linguistic frequency of a bigram unit by the giVen

Bolt'Optanek and:Newman -Inc.

amount. Variable. 4 provides a second measure of the ability to

orthographically regular multi-grapheme units. Variable 2e,(sca ning speed) is thought to provide a more genera/ measure of

PerceptUal Encoding, andto reflect both the single grapheme and

multi-grapheme subprocetses.

3. Pseudoword Decoding. In the 'pseudoword decoding tam

subjects are asked t6 pronounce pseudoword items that have been

derived from actual'Englis words by changing a ,single vowel.

(e.g., BRENCH, derived from BRANCH). The set of pseudowords

covers a number:obrthographic formt, including variations in

6

length, number of syllables, and type of vowel (Frederiksen, Note

1). We' measure the RT from the presentation of the display to

the onset of the subject's vocalization and the duration of, his

vocal response. Five measures of decoding are derived from this.

experiment: The 'Array Length Contrast {Variable .5) is the

increase in mean RT. for each added letter (e a CCVC-, CCVCC,

CCVCCC). The qyllable Contrast (Variable '6) is measured by6

1.

04

-Report No. 3757 Bolt Beranek and Newman Inc.

subtracting the. .mean onset RT for two-syllable items-from that

for one-syllable item's that are matched on initial phoneme and_\ :

orthographic form (e.g., CVC-CV &rid CVCCV). The Vowel Complexity.

\

Contrast (Variable n-is measured by 'subtracting the ,mean

RT for pseudowords having sequences of bp vowels Ck.g., VVCC)\

\,

from that for pseudowords having single vowels e.g., CVCCC). In

addition, the ,syllable and vowel comPexity contrasts were-

-caluclated using-vocalization durations; forming Variables 8 and. ,

9.. These,_Contrasts in all cases refleCt the increase inw, ....

.

'dedsi Iiccilty occasioned by increasing the orthographic

comple - of a stimulus item in a designated manner, and-are

regarded as,meesures of Decoding. It is thought that measures

based upon. RT to onset of vocalization tap earlier,decoding

processes of parsing and, phonemic' translation, while measures

based .upbn vocalization durations tap_ later processes ..of

articulatory progiamming, stress assignment, - and . the

establishment of 'prosodic features.

4. Word Naming. This task is in every respect similar to

thePseudoword Decoding task, except, for the use of English words

places of'-pseudowords. In addition to Variations in

orthographic form, the stimulus words are chosen to represent two

Linguistlis froequencies,of o4currence, low*requencY words (having.

1.

-13-

24

, ` 1---," s, N

t

.. -.4

Repor.t,No.757,..-

Bolt Beranek and Newman Inc...

-__.\._a mean SFI ,in

.

of 27.0) and high-\,frequencywol-dI-Thiliiii767-------11.'

./

Mean SFI index of 56. Each of the :five ;contrasts descri!bed,

...,above for the Pseudoword 1/ecOding task is also calculated for-khe

. \

Word Naming task, for both high frequendy words (HFWs) and low

a.. I

frequency words (LFWs). Two meaSures e constructed in order;1 .

.

to compare the extent of use of 'Decoding in>proce sing, high and1

low frequency words and pseudowords. The Peice t Dro in

Decoding Indicators, fOt HFWs' and Pseudowords (Vari 1 10) is ,

measured by summing the valUes of the five contrasts for both

IHFWs and Ps udowords, and calcidating the percent drop wing ,theI

formula: \

I\.% Drop = (Sum(Pseudowordsj - Sum(HFWS).) iNSAm(Pseuddwords

The Percent Drop in Decoding Indicators for\ HFWs and,. LFWs -is

measured in a similar manner, by substi utinq LFW'

pseudowords in the above conipar ison . These v. iables ver

developed to measure a fundamental characteristic- LeXAcal

,\Access: the-depth of processing of orthographic information that

characteristically takes place prior to lexical retrieval. Lar

'values for either of these contrasts indicate a decrease in deptho

of processing' when the stimuli are familiar English.words, while

small values incidate that there is a continued use of

word-analysis -skills'in the 4ecognition of common words,. .

2 The SFI or Standard Frequency Index is a logarithmicallytransformed word frequency. Scale (Carroll, Davies, & Richman,1971). High values represent engl'sh words that occur commonlyin text; low values iepreseneuncom on words.

-14=

25

-Report No. 3757., Bolt Berane3 and Newman Inc.,

Relation to Hypothesized Component Skills

It has been our belief that the set of-measures we hive

describecrwill permit us :to, distinguish the five component,

rocesses

components

Perceptual

graphemes

elludd to abUve and listed in Tal?3 The first/two'.

(or -factors) refer to x4e. to . subprocesses of

Encoding, dealing with the encoding of_individual

d with multi -g pheme units. The. this -and fourth

components reer to hierarchically organized 'levels of Decoding:

Phonemic translation includes the parsing of a grapheme array and

the application of orthographic rules to derive a phonemic'

representation. Automaticity of articulation refers to

operations performed on an initial phonemic representation in

deriving an articulatory Or speech representation, including,th'e

assignment Of tress pattern and other- prospdic features. The

last componer4 process refers to what is probdbly the most

fundamental characteristic of Lexical Access, h`amely, the depth

of processing c of the orthogr,aphic. code- prior to lexicalco

retrieval.

The relations we have described 'between component skill

,Imeasures and component processes can be summarized'Compactly in a

.

factor matrix, shown in Table 4. Ignoring for the moment the

numerical value contained in the. table, thehypothesized factor

structure is re resenited by the positions of zero end positive

values in th table. A value (or loading) of zero -for a 'variable

'CZ.1.

-15-

260""

%lb

_ t, - 4 t444

44 ':177

t.,No";73.7517'Repo Bolt BeranekF.,,E.'2.,,,,__"?'-.7---=,..-' . ,,---. ,,,f t ---- ,-x,,-- .-, ..--...- -, "C.";

,

--.. .,..-s, ,.. . , -"N.P 4 ` -/":',1

- -'''':' a'ABLE-,1

_:"......,_*--_,,,t,c \-., -,...7,-; ---.7-_,,--:, ..........,_,.,,,_,,, , : - A

1'.--,477'71i, , 7".,. '-'7=:-.Z., Zi-,4 2,---_-- '-'77' '''''''' r.,.-,---,...

/Deiinition ComPon'et- Pi9ceties,

....;_.-a..!,

-,i--- .."-'--4-liypotheize,.n. the Analysis of Covariance Structures_ ...

:------- -4----1,-----:7-1 -:- , . ... .. (

---.2--. '&,-

and Newman Inc.

.

4,3PAOTOR NAME . DESCRIPTION

.-- r-. k - -,.

:.Grapheme Encoding,

1)..er_peptual. Facilita-

iti-,E.ficbd

Muiti-lette,r Arrays- .

,, -- --- ' .- -- ; -t-,--.-. --7 - _,,---\ --- -,-J,,:

Efficiency in LetterIdentification

Phonemic Translation

-Efficiency in EncodingOrtehograPhicalzy Regularor -Redundant. LetterSequèrces

Efficiency-in Applying'SPelling.Rules to Derive

. ,

V.A e

-

Automaticity of_Articulation

,ree

Depth 9f ProcessingWord Recognition

(

a Phonological/phonemicRepresenismation -3

Efficiency in Articulation;.Syllatication, Assignmentof Stress, Prosodics.

in Use óf Visual,or Whole-WordRec nitfon Strategy inRec, niiing Common-Words.

,

.

.e

Report No. 3757 Bolt Beranek and Newman Inc.

'TABLE 4

Maximum Likelihood Estimates of Factor

Loadings and Uniquenepses for the Experimental Variables*

VARIABLEI

FACTOR:II., III IVY V

Unique -ness

ti

1. Lfter Encoding.

1.00 0.! 0. 0. 0. .00i

1

2. Scanning Speed -..) .64 .53 0. 0. 0. .53

3. Percep. Facilitation 0. .62 0. 0. 0. .624

4. Bj.gram PrObability 0. .54-..0. 'O. 0. .71

5. Length: Pseud. .16 ,U. '.77 0. 0. :36

6. Syll.: Pseud. 0. 0, .80, 0. 40. .37i

1J- ?,-

7. Vowel: Pseud. 0.,/) 0. .55 0. 0. .70t

8. Syn.: Pseud. (Dur.) -0. 0. 0. .96 0. .08

9. Vowel: Pseud. (-Dur.) 0. -0. 0. .36 b. .87

.,

10. A% Decod.:Pseu.-HFW 0. 0. 4. 0. 424 .94 '

11 A% Decod.:LFW-HFW. (t.' 0. 0.. 0. 1.00 ' e

2

00.

.

we*

Zero loadings were fixed by hypothesis; the'goodness-of fit ofthe 'hypothesized structure is measured i x2(32). p

Report No. 3757 Bolt Beranek and Newman Inc.

indicates that that variable is by hypothesis not considerEd a

measure of the particular component process, and it not, expected

to be related to that component except' through possibl%,

Correlations between component processes. A positive loading

indicates that the variable in question,is hypothesized to he a

measure of the particular component process, although t exact

vale of the loading re ins to be estimated*ron the basis of

datA. By reading down olumn of Table 4, one can see which RT,

contrasts have,bee6c:hypothesized to be markers of'a given factor.1

By reading across rOws,.one can see the hypothesized factorialI

composition for each variable.

III. EVALUATION OP THE COMPONENT SKILLS MODEL

Method

So far, this discussion has focused on the nature of

component processes in reading. -and the types of chronometric

measures used in their measurement. Our ability to validate the

component-skills analysis is based upon'an itportant 3eAtelsopmet-.

in the application of statistical.theory to the problem of factor ,,

.ti

analysis, worked )out a few years ago by K*1 agreskog (1970)4.-

Jgreskog's ,technique allows us to , estimate dir.ctly the.4g

parameters of w.,'factor model us:nz the Method of maximumS'

likelihood, provided that ;the number' of parameteF,s, to be

estimated 'does' .not exceed .the degrees. of freedom in the

L

Report No. 3757 Bolt Beranek and Neviman Inc.

,covariance or correlation matrix being factored and that the

-hypothesized factor matr4x is unique in that it precludes ,

rotation of axes, The investigator reduces the 2'umber of

parameters in the'analysis by constraining,the'parameters of the

model (v lues in the.factor matrix, intercorrelations among the

factors, or uniquenesses) to have specified values or to'be equal,Nto other parameters in the et to be estimated. aireskog's (Note

2) program provides a test of fit of the'hypothesiZed factor,

structure, represented by the choice o onstraints on the values

of the parameters. Finally, comparisons a alternative

structural models can be made using a likelihood,ratio test.

Sub'ects

/ Data available for testin the structural model in Table 4

are the scores of 20 subjects who were tested on each of the

tasks 'we have esceibed. The subjects were high school

sophomores, juniors, and seniors, and repregented a wide range of

reading ability levels. Their reading scores on the Nelson-Denny.

Reading Test ranged fr m the ,16th to the 99th percentile..4

ApproxiMately equal num ers of subjects were drawn from a city

and a suburban high school.

Results

The-goodness of fit Of the hypothesized fctor-,,structure is0

Oven in Table 4,, along with estimated values for the- factor

- -19-

30

I

4

Report No. 3757 Bolt Beranek and Newman Inc.

loadings.' The obtained Chi-square of 38.4 (with 32 degrees of

freedom) has a probability of .2, indicating, that the sample

correlation matrix would be obtained with high likelihood given4

that the hypothesized structure is the true factor structure.

Moreover,.the values of the loadings in the factor matrix support

in detail the hypothesized component processes model. Factor\ I,

Grapheme Encoding, is clearly marked by the letter encoding and

scanning speed measures. Factor II , Encoding Multi- Letter

Units, is marked by,the perceptual facilitation contrast derived

from the letter matching.task and the bigram.probability cbtrast

derived from the bigram identification task. The three decoding

indicators calculated from onset RTs in the7.pse4Udoword

pronuncition task load on the'Phonemic Translation factor {III),

and the two' deblErit4: contrasts based upon vocalization dutationS

load on the Articulation fattor (IV). Finally,, the measures of

processing depth in, reading words both load on the last

factor (V), Depth of Processing inWord Recognition.

Estimates of the intercorreIations among the factors are

preSented in Table 5.' A likelihood ratio test of the hypothesis

. that the factors are mutually orthogonal yielded- X2

(10)=20.29,

with p<.05. The factorS can therefore be assumed to be

correlated with one another. SeVeral patterns among these

correlations are of interest. (1) Factors III-Voappear to be

mutually' orthogonal, suggesting that each is. tapping an

-20-,

31

Report No. 3757 r-Bolt-Bdranbk and Newman Inc.

TABLE 5

MaxiMum Likelihood Estimates of Ihtercorrelations

Among the Factors*A

A sit

rilACTOR I II - IV V

I. Grapheme Encoding 1.00

II. Percep. Facilitation -:32' 1.00

III.

Iv.

Phonemic Translation,

Automaticity of

.09-

.58

.41

.24

1.00

-.17 .1.00Articulation

.e-V. Depth of Processing

in Word Recognition-.11 .08 .01 1.00

r

A likelihood ratio test of the hypothesis of orthogonality of

the factors yielded x2(10) = 20.29,with-p < .05.

b

1-

C

M'

Report No. 3757 Bolt Beranek and Newman.InC.

independent aspect of. the reading rocess. Facility in

parsing/phonemic translation appears to be unvrrelated with

processes related to, articulation, and the "extent o decoding in

reading common words is not related to a subject's level of skill

,at the decoding level. (2) The two aspects of Perceptual

Encoding, on the contrary, do appear'to be related to skill in

decoding and lexical access. Subjects who are highly efficient

in encoding multi- letter, graphemic units are faster in phonemic

translation (r=.41) and in Xiculation (r=.24), and teed to use

heir decoding skills in, ccesting common English words in their

exi.con (r=.52). It subjects who are less: proficient in

identifying multi etter units that 'dedreaseiiir 'depth of

processing when reading high frequency words. Interestingly,1

,there appears' to' be a small, reciprocal relationship between

efficiency in single letteY encoding and in encoding multi-letter

units (r=-.32). ,,(3) Fina3ly, it appears that subjects who are

rapid in ,encoding individual .graphemes are also more rapid in

articulatory processes (r= .58).'

Evaluation of Alternative Structural Models

. Three alta4native hypotheses aboutthe factor structure were. \

developed, in order to see if the finer distinctions made between

subprocesses of Perceptuai'Encoding and Decoding are necessary.

The results of these investigationsare present0 =in Tabre:6. In

the first. alternatiye model, we were interested- in. the

1

1

Report No. 3757

1

%TABLE 6

-Bolt Beranek and Newman n .

r

Test of Fit for Three Alternative

Hypotheses about the Covariance Structure

. Alternative Model Effects on Hypothesi2edFactor Structure

1. No distindtionsare made betweenSubclasses ofPerceptual Skills

2. No Distinctionsare made betweenSubclasses ofDecoding Skills.

3. No Distinctionis made betweenthe PerceptualEncoding of Multi-letter Units and

Cithe Parsing of aGrapheme array as aComponent ofDecoding.

Factors I and IIare combined intoa single PerceptualEncoding factors

Factors III aid IVare combined intoa 'tingle Phonemic

Translation factor.

Factors II and IIIare combined intoa single Parsingand Phbnemic Translationfetor.

Number ofFactor, square

d.f.'.

4 54.16 37 .034

4 54.00 36 . 027

51.12 36 .049

cs

)

23- 34

41,

Report No. 375710.

V

a-

Bolt Beranek and:Newman".Inc..,

. ,

encoding'distinction' between. perceptual-encoding' of ,individual graphemes

and multi-grapheme units, represented. by factors I and II. These. ..

.

two were, accordingly, combined into a single Percepttal. .

Encoding factor; in all other respects, the model was similar to

the general model in Table 4. The test of fit yielded'

X2 (37)=54.16 with p=..464, .leading us to reject the firstI

alternatiye model and to conclude that adisiinctiOn must be

maintained between the/two aspects of,Terceptbal Encoding as I

originally hypothesized. . ..

In the second alternative model',.the distinction between

early (pariing, 'phonemic translation)L\eand late (articulatory.

programming) decoding. -processes was dropped. Accordingly,

factors-III and-IV were combined into a single Decoding factor,'.

while in all other respects the model was similar t.p.our'originai!"

model. The test, of fit yielded X2(36)=54.0 with p=.027.-. We

were thus again led to reject the alternatiVe 'model and :tó

conclude that the distinction between levelS of 'analysii within

the decoding process mtst be maintained. t

e In the third alternative model, we were' interested -in

testing the importance of the distinction between the. perceptual

parsing of a 'grapheme array (represented--by factor, II) and-,

parsing cdnceived as a component of decpsling (faCtor III)..

,

Accordinbly,in this model factorsII and III were combined- into

a single factor.. The likelihooeratio'test yielded r(36)=51.12

-4-

Report No. 3757

with p=

Eolt:Beranek and Newman Inc.

.049, and again we were led to reject the alternativ

model. Evidently the_ perceptual grouping ,of,-graphres into

overl pping, multi-grapheme units, ins distinct fromrulerbased

p esses involvedin the translation of an orthographically

regular array.

Testing 'the External Validity of the Component Skills Model

A final source of information concerfting the of.the\

component skill measures lies in their relationship to other,

more general measures of reading profibiency. We are interested

here in establishing what role the component processes .play "

setting levels of reading skill, as fileasied by conventional

tests of reading ability and comprehension.. Two sets of

criterion variables were used: (L) Chronometric Measures

representing overall levels' .of performance in reading

individually presented words, and pseudowords, and'i.(2) Reading'.

Test Scores, including the,Nelson-Denny Total Scoie (the sum of.

. .., \

Vocabulary and Reading Comprehension subiests), Nelson-Denny

Reading Rate, and the Gray Oral:Readingn%est; Total Passage Scorewt

(which includes number of pronunciation errors and reading rate).

The oadings of each. of these criterion eVariablss .on the

nent skill factor* were calculated using a factor extensioncorn

pio edure, and areopreseeteein'Table .

4.

Report No. 3757 Bolt Beranek and Newman Inc.

TABLE 7

Loadings of Criterion Variables

on the Component Skill Factors

CriterionVariable

I II

GRAPHEME PERCEPTUALENCODING FACILITA-

TION

FA

IVPHONEMIC AUTOMA-TRANSLA- TICITYTION OF

ARTICU-LATION

VWORDRECOG-

,NITION

SQUARED1%.MULT.

CORRELA-TION

Chronometric Measures

411.

1. Can Onset .14 .70 .35 .29 '.73Latency:Pseudo.

2. Mean Onset 33 .43 .49 .36 .56Latency:LFW

0

3. Mean Onset . .27 .55 -1.12 .46 .68Latency:HFW . _

4. Word Frequency, .08 .72 .33 . .27 .22 .99Effect. (Onset pT)

Reading Test Measures

5. Nelson -Denny: t -.42, -.59 -.02 -.69 -.35 4 1-.00

. Total Score

Nelson-Denny: -.52 -.23 -.62 -.25 .73

Speed .

7. GrtliirlZal. -.39 -.24 .09 -.43 -.37 .53Rea

(

-26-

37

Report No. 3757 Bolt Beranek and Newman Inc.

Chronometric Measures. Mead onpet latencies for pronouncing

pseudowords and low'or high frequency words (criterion variablesP

'1-3) are highly predictable from the compbnent skill factors,

3with multiple correlations of .85, .75, and .82, respectively.

There is a high degree of consistency in the pattern of loadings

for eachof'these criterion.vatiables: While Grapheme Encoding is.

positively --but not strongly -- related to efficiency in reading'.

words and pseudowords, the ability to encode multi-letter units

is the strongest-predictor .of oral reading latencies. Phonemic

'Translation is related to pse d decoding latencies, lout not

tolatencies sr pronouncing English words, However,

Automaticity of ArticulatiOn does mourn out to be strong

predictor' of reading latencies. Finally,' the loadings pn the

Visual Recognition' factor support our earlier contention

(Frederiksen; 1916) that is the poorer'readers that use a

visual or whole-word basis for recognizing familiar_ words.

The difference in reading latencies- for low and high

frequency woids wax entere as the fourth criterion: variable.

The items contributing to the hig and-low frequency 'scores were

balanced ifi we ind that the grapheme

, 'encoding, component does not predict thisX

iterion. On the,pther

. . - .,,,,,,,-,

hand, high and low frequency words do differ )the populations--.

, ,\.,

The mult pie correlat, ns' are sub3ect td shr,nkbe regarded only as indices of the degree of. shaibetween the component skill-factors and the criteria.

eand,bhould_.variance

-27-4

t No. 3757r

Bolt Beranek and Newman Inc.

of graphem hey contain, and we are thus not surprised to find

that tie multi-let encoding factor .is a strong, predictor of- .

differences in latencies reading low and high frequency

Wards.(Finally, the positive loa on factors III-V 'suggest

again that high and low frequency words a alysed'in different

ways prior to lexical retrieval.

Reading Test Measures. 4 The scores for the three

test measures are highly predictable from the component ski

factors, with multiple correlations of 1.00, .85, and .73 for the

Nelson -Denny Tot41, Reading., Rate,, and Gray Oral Reading Test

s ores, respectively. Again, the strongest predictors appear to

be ncoding Multi-Letter Units d Automaticity of Articulation;

Subjects scoring highly on the re ing tests also. tend tcr/-4be-

efficient in Gephme Encoding and to use thier decoding skills inJ

reco§nizing familiar English wards as well as less familiar

items. Low scoring subjects again are found to be ;ess.efficiept

in encoding individual graphemes, in perceiving multi - grapheme

units-, and in their degree of automaticity in. the final

decoding, and they tend to recognize faMiliar words on

of their visual characteristics.

tag-es of

he basis

The loadings are negative, indic ting that efficiency In

--Proc.spairig within the domairi of each component skill is related:),to,' high Scaies-on-the reading tests.

-""*.;-%' oo

Report No. 3757 Bolt Beranek and_Ne man Inc.

IV. SUMMARY AND CONCLUSIONS

\ The evidence .we have$collectedsupports a component process

model for readihg that "distinguishes at least five 'component

skilld: \I. EffiCiency in perceptuall encoding o

hemes, II. Efficiency in encoding, orthographically regular,

multi apheme\Alnits, III. aficiency in paisAag' an encode

grapheme array and in applying letter -sound corresPCnden0 rules,

\

to deriv a \phonological/. pponemic representation,\

IV. Automatici y in deriving 'a speech` representation, n the

assignment of str s and° other prosodic features, and V. the

proCess ,of 'lexical retrieval, characterized 'by. the'deithof

rocessing (perceptual encoding and decoding) tria---takes place

Prio %.k,tp lexical access. The picture we have, gained of the

patterns of ercorrelation among component skills and' their-.* .

.

relatedness. ds ot reading proficiency permit us to-draw,

y,

two more general ,,, .

1. While com nent c.,_rega -. hierarchicall4

ordered, the' initiati ocesses (e.g., 16d.c612

retrieval)val) Opo not ne\cess

\ 7 14 ' . .

Idx,icai retrievai is seen to, var with the famifiarity of aword..

, \ X,

High\frequency words may be recogn ed on the basie..of -their.

woo* visual\

haracteristibs, without\ th completion of the grapheme

ily await the comple on of earlier

operations. \lhu the depth of processing prior to.

encoding and d coding Rrocesses req red for .recognizing

unfamiliar sds.\:\

1

Regort No. 3757

f

n'e*,

Bolt Beranek and Newman Inc.

'2. .There 'ate interactions (trade-offs)cbetween the use of skills-,

.,. . '. r. . . . .. 4

-At- one level- of. .processing and the mode of processing and,

proceSsing effIdiency,,At-higher levels of-processing. Thus,, an 7 '1_.,,x.,..,...,........

.___\....._ability 'tO.:perceptilally encode__ multi- letter .units reduces the

demands placed on the decoding component, with a consequent -,

4 1

t increase inefficiency of .decoding: Readers who have high, scores

on :factor II (Encoding Muli-letter Units) are also the fastest

decoders,- and -iliellare likely to ''-' apply their efficient,..

word-analysis skills in recognizing--Common as well. as rarcwords., .

On the other hand,, readers'' who, have a low level of still in. -

.

.

percepplly encodin g- ,multi-letter uni ts havee the grel. estI

difficulty-in decoding grapheme art-a.ys-..,..into ."sound," and they are

.-,,the ones who are most . like,lytorecce .the depth'. of -processing,

.. .

-- when Visually familiar wards are encountered. This processing,

-,%.,,

), interaction- 01,911tistrates: how -the- mode - of processing at a highs; ''

,,k

level (here4.. tb, -of evidence,used as` a basis-for performing.t,

illexid dcSsi ..i enced-Wthe_leVel of skill in processing a

' --: .:::: c-_

"-at' a il- ev m6difiCatiO procedures for high-level Iie,

----processiii laeg-Oal--(4 acce,ss)'-..74e4Ves ,40. compensate. for low

.,,_, ..,,-,." ;,.f.,, - ,

ll_.efficiencieSin,_ lowkr-level conIggnept- processes'. Thus,. the` 65

., if : L ' -.' . 4 , s3 ... ... '

system adapts to it' itwn defiCtenoies;And,, is able,--

--, .

to LigroveII

it's overall *foance when the'StiMalus materials permit such, . r4 . ''''N t

an adiUstmAnt: rocessing alaitexistids to take place. In I,- ...,,,_!,..4, i

general, we beive5--,,-that models ,fps- human information processing

A 2. ,

.

).

, j

, ,. ,,t- -30- 1 -1) I

.

11.ki

Report No. 3757 Bolt Beranek and Newman Inc.

within a complex domain such as that of reading will have to

account for individual differences in the procedures used by the

system in allocating its components for _the solution of a

problem, as well as for skill difierenc s among subjects in

processing efficiencies within, the

themselves.

9

-31-

42

componeht/ processes

fr

Report No. 3757

1

"7.

Bolt Beranek and. Newman IdC.

REFERENCE NOTES

Frederiksen, J. R. Decoding skills and lexical retrieval

Paper delivered at the annual meetings of the Psychonomic

Society, November, 1976._ 7

2. Jgresk$og, K. G.,,Van Thillo, M., & Gruvaeus, G. T. ACOVSM:

A General computer program for analysis of"cdvariance

structures including generalized MANOVA. RB-71-1.

Princeton, N.J.: Educational Testing Service, 1971.

-32-

0

C

c

. 43

,

A

Report No. 3757 Bolt Beranek and Newman Inc.

REFERENCES

Carroll, J. B. Psychometric tests as cognitivelasks A new

"structure_of intellect." In Resnick, L. B. (Ed.)

The nature of intelligence. Hillside, N.

Erlbaum,.1916.*

Carroll, J. B., Davies, P., & Richman,.B. The American Heritage,

Word Frequency Book. Boston: Houghtcin M. flin Co:, 1971.

Frederiksen, J. R._. Assessment of perceptual, ecodng, and

lexiwal skills and their relation to reading proficiency.

In Lesgold, A. M.Pellegrino, J. W., Pokkema, S., &

Glaser, R. (Eds.), Covitive Psychology and Instruction.

New York: Plenum, 1977..

French, J. W. The description of aptitude and achievement tests

in terms of rotated factors.. Psyftometric Monograph,

Numbei 5. Chicago: University of Chicago Press, 1951.

French, J. W., Ekstrom,, R. B., & Price, L. A. Manual for kit of

reference tests for cOgnitive factors. Princeton( N. J.:

Educational Testing Service, 1963 (revised):

Guilford, J. P. The nature of human intelligence. New York:

McGraw-Hill, 1967,

J8reskog, K. G. A general.method'foranalysis of covariance

structures. Biometrika, 1970, 57, 2, 239-251.

MaYzner,,M.. S., & Tresselt, M. E. Tables of single-letter and

digram frequency counts for variouOlioid length and letter

-33-

44

2

1

Report No. 3757 Bolt Beranek and_Ne4man Inc.

position combinations. Psychonomic Monograph Supplement,

11965, 1,13-22.

Posner, M. I., & Mitchell, R.F. Chronometric, analysis of . I

Classification. Psychological Review, 1967, 74, 392-409.

Popfter, M. I., & Snyder, C. R. R. Attention and cognitive

control. In R. Solso ('Ed.). Information Processing

and Cogflition: The Loyola Symposium. Potomac, Md.:

Lawrence Erlbaum Associates, in press.

Spoehr, K. T., & Smith, E. E. Thb role of syllables in

perceptual processing. Cognitive Psychology, 1973, 1

5, 71-89.

Thurstone, L. L. Primary mental abilities. Psychometric

Monographs, Number 1. Chicago: University of Chicago Press,

1938. s

ghur tone, L. L., & Thurstone, T. G. Factorial studies of

i itelligence. Psychometric Monographs, Number 2s .

Chicago: University of Chicago Press, 1941.

-34-

45

AN.