ED 050 925

DOCOINEIT RESUME

24 RE 003 649

AUTHOR Chapman, Robin S.; Ting, Ai ChenTITLE The Effect of Mode of Elicitation in Articulation

Testing.INSTITUTION Wisconsin Univ., Madison. Research and Development

Center for Cognitive Learning.SPONS AGENCY Office of Education (DREW), Washington, D.C. Bureau

of Research.REPORT NO TR-154BUREAU NO BR-5-0216PUB DATE Mar 71CONTRACT OEC-5-10-154NOTE 16p.

EDRS PRICE EDRS Price MF-S0.65 HC-$3.29DESCRIPTORS *Articulation (Speech), College Students,

Consonants, *Error Patterns, Grade 1, PictorialStimuli, Preschool Children, *Reading Research,*Reading Skills, Visual Stimuli, *Word Recognition

ABSTRACTDifferences in articulation error rates and error

patterns as a function of five elicitation modes (picture, picturewith pretraining, word repetition, sentence repetition, and nonsenseword repetition) were examined. The same 15 words (or picturesrepresenting them) were stimuli in four real word conditions;nonsense words were formed by recombining the vowels and finalconsonants of the real words. Subjects were 22 preschoolers, 8kindergarteners, 10 first graders, and 24 college students. Eachsubject was tested in each mode with order of testingcounterbalanced. Errors in initial consonants, initial consonantclusters, vowels, final consonants, and final consonant clusters werescored separately. Error rates and error patterns were similar forthe four real word modes. The nonsense words showed an error ratethree times higher than that of the real word modes and, in constrastwith the real word modes, increased rather than decreased frominitial to final position. The effect of test items which areeffectively nonsense as well as additional considerations bearing onthe choice of real word testing modes are also considered. Tables andreferences are included. (Author/VJ)

A P LA

'

zgte", .

SCOPE OF INTEREST NOTICE

The ERIC Facility has assignedthis documeJ for processing

\to:

In our judgement, this documentis also of interest to the clearing- Lishouses noted to the right. Index- /ing should reflecr them specialpoints of

U. S. DEPARTMENT OF HEALTH, EDUCATION & WELFAREOFFICE OF EDUCATION

THIS DOCUMENT HAS SEEN REPRODUCED EXACTLY AS RECEIVED FROM THE

PERSON OR ORGANIZATION ORIGINATING IT. POINTS OF VIEW OR OPIN.OTTSSTALL) DO NUT NECESSARILY REPRESENT OFFICIAL OFFICE OF EDUCATIONPOSITION OR POLICY.

Technical Report No. 154

THE EFFECT OF MODE OF ELICITATION IN ARTICULATION TESTING

Robin S. Chapman and Ai chen Ting

Report from the Project on Reading and Related Language ArtsBasic Prereading Skills: Identification and Improvement

R. L. Venezky, Principal Investigator

Wisconsin Research and DevelopmentCenter for Cognitive LearniirgThe University of Wisconsin

Madison, Wisconsin

March 1971

Published by the Wisconsin Research and Development Center for Cognitive Learning, supportedin part as a research and development center by funds from the United States Office of Education,Department of Health, Education, and Welfare. The opinions expressed herein do not necessarilyreflect the position or policy of the Office of Education and no official endorsement by the Officeof Education should be inferred.

Center No. C-03 / Contract OE 5-10-154

2

NATIONAL EVALUATION COMMITTEESamuel Brownell Henry Chauncey Elizabeth KoontzProfessor of Urban Education President Wage and Labor StandardsGraduate School Educational Testing Service Administration, U.S.Yale University Department of Labor,

Washington

Launor F. Carter Martin Deutsch Roderick McPheeSenior Vice President on Director, Institute for President

Technology and Development Developmental Studies Punahou School, HonoluluSystem Development Corporation New York Medical College

Francis S. Chase Jack EdlingProfessor ::',.ector, Teaching ResearchDepartment of Education DivisionUniversity of Chicago Cregon Stute System of Higher

Education

G. Wesley SowardsDirector, Elementary EducationFlorida State University

Patrick SuppesProfessorDepartment of MathematicsStanford University

*Benton J. UnderwoodProfessorDepartment of PsychologyNorthwestern University

RESEARCH AND DEVELOPMENT CENTER POLICY REVIEW BOARD

Leonard BerkowitzChairmanDepartment of Psychology

Archie A. BuchmillerDeputy State SuperintendentDepartment of Public 1 struction

Robert E. GrinderChairmanDepartment of Educational

Psychology

Russell J. HosierProfessor, Curriculum

and Instruction

Clauston JenkinsAssistant DirectorCoordinating Committee for

Higher Education

Herbert J. KlausmeierDirector, R & D CenterProfessor of Educational

Psychology

Stephen C. KleeneDean, College of

Letters and Science

Donald J. McCartyDeanSchool of Education

Ira SharkanskyAssociate Professor of Political

Science

B. Robert TabachnickChairman, Department

of Curriculum andInstruction

Henry C. Wein lickExecutive &.. toryWisconsin Education Association

M. Crawford YoungAssociate DeanThe Graduate School

EXECUTIVE COMMITTEE

Edgar F. BorgattaBrining'. Professor of

Sociology

Anne E. BuchananProject SpecialistR & D Center

Robin S. ChapmanResearch AssociateR & D Center

Robert E. DavidsonAssistant Professor,

Educational Psychology

Frank H. FarleyAssociate Professor,

Educational Psychology

Rinse!! J. Hosier Wayne OttoProfessor of Curriculum and Professor of Curriculum and

Instruction and of Business Instruction (Reading)

*Herbert J. Klausmeier Robert G. Petro IdDirector, R & D Center Associate Dean of the SchoolProfessor of Educational of Education

Psychology Professor or Curriculum andInstruction and of Music

FACULTY OF PRINCIPAL INVESTIGATORS

Vernon L. Allen Frank H. FarleyProfessor of Psychology

Ted CzajkowskiAssistant Professor of Curriculum

and Instruction

Robert E. DavidsonAssistant Professor of

Educational Psychology

Gary A. DavisAssociate Professor of

Educational Psychology

M. Vere De VaultProfessor of Curriculum and

Instruction (Mathematics)

Associate Professor of EducationalPsychology

Lester S. Golubiacturer in Curriculum and

Instruction and in English

John G. HarveyAssociate "rofessor of

Mother atics and of CurriculumInstruction

Herbert J. KlausmeierDirector, R & D Center

Professor of EducationalPsychology

Donald LangeAssistant Professor of Curriculum

and Instruction

James MoserAssistant Professor of Mathematics

Education; Visiting Scholar

Wayne OttoProfessor of Curriculum and

Instruction (Reading)

Milton O. PellaProfessor of Curriculum and

Instruction (Science)

Thomas A. RombergAssociate Director, R & D Center

Professor of Mathematics and ofCurriculum and Instruction

B. Robert TabachnickChairman, Department

of Curriculum andInstruction

Richard L. VenezkyAssistant Professor of English

and of Computer Sciences

Alan VoelkerAssistant Professor of Curriculum

and Instruction

Larry WilderAssistant Professor of Curriculum

and Instruction

Peter WolffAssistant Professor of Educational

Psychology

MANAGEMENT COUNCIL

Herbert J. KlausmeierDirector, R & D Center

V.A.C. Henmon Professor ofEducational Pxydiciogy

Mary R. QuillingDirector

Technical Development Program

Thomas A. RombergAssociate Director

James WalterDirector

Dissemination Program

Dan G. WooipertDirector

Operations and Business

COMMITTEE CHAIRMAN

STATEMENT OF FOCUS

The Wisconsin Research and Development Center for Cognitive Learningfocuses on contributing to a better understanding of cognitive learning by chil-dren and youth and to the improvement of related educational practices. Thestrategy for research and development is comprehensive. It includes basic re-search to generate new knowledge about the conditions and processes of learn-ing and about the processes of instruction, and the subsequent development ofresearch-based instructional materials, many of which are designed for use byteachers and others for use by students. These materials are tested and refinedin school settings. Throughout these operations behavioral scientists, curricu-lum experts, academic scholars, and school People interact, insuring that theresults of Center activities are based soundly on knowledge of subject matterand cognitive learning and that they are applied to the improvement of educa-tional practice.

This Technical Report is from the Basic Prereading Skills: Identificationand Improvement element of the Reading and Related Language Arts Project inProgram 2, Processes and Programs of Instruction. General objectives of theProgram are to develop curriculum materials for elementary and preschool chil-dren, to develop related instructional procedures, and to test and refine theinstructional programs incorporating the curriculum materials and instructionalprocedures. Contributing to these Program objectives, this element has twogeneral objectives: (1) to investigate ways to test for skill deficits and toovercome them and (2) to develop a kinde7garten-level program, including diag-nostic tests and instructional procedures, for teaching basic prereading skills.Tests and instructional programs will be developed for visual and acousticskills, including letter and letter-string matching with attention to order, orien-tation and detail, and acoustical matching, segmentation, and blending.

iii

4

r.

CONTENTS

Abstract

Page

vii

I. Introduction 1

II. Method 3

Design 3

Stimulus Materials 3

Procedures 4Subjects 4Data Transcription 4Error Classification 4

III. ResultsFailure to Elicit Test Words 6

Nonsense Word Errors 8

IV. Summary and Conclusions 9

References 11

LIST OF TABLES AND FIGURES

Table Page

1 Stimulus Materials 3

2 Mean Per Cent Error by Sound Class and ElicitationMode for Each Age Group 5

3 Mean Per Cent Error by Sound Class and ElicitationMode, Averaged over 40 Children 6

4 Per Cent Failure to Give Test Word, by Group andMode of Elicitation 7

Figure

1 Difference between Error Rate for Sound Class andAverage Error Rate for Each Mode 7

ABSTRACT

Differences in articulation error rates and error patterns were examined asa function of five elicitatio.i modes: picture, picture with pretraining, wordrepetition, sentence repetition, and nonsense word repetition. The same 15words (or pictures representing them) were used as stimuli in the four real wordconditions; nonsense words were formed by recombining the vowels and finalconsonants of the real words.

Subjects (Ss) were 22 preschoolers, 8 kindergartners, 10 first graders, and24 college students. Each S was tested in each mode with order of testingcounterbalanced. Errors in initial consonants, initial consonant clusters,vowels, final consonants, and final consonant clusters were scored separately.

Error rates and error patterns were similar for the four real word modes.The nonsense words showed an error rate three times higher than that of thereal word modes and, in contrast with the real word modes, increased ratherthan decreased from initial to final position.

The effect of test items which are effectively nonsense is discussed fornormative and correlative studies. Additional considerations bearing on thechoice of real word testing modes are presented.

vii

6

INTRODUCTION

In testing articulation, the experimenteror clinician muFt decide how speech soundsare to be elicited from the child; this paperis concerned with some of the consequencesof that decision.

Speech sounds may be elicited spontane-ously by pictures, incomplete sentences,printed words or sentences, or conversation.The spontaneous mode may have the advantageof eliciting normal, or habitual, speech pat-terns, if this is the experimental goal; it hasthe disadvantage, in younger children andpoor readers, of failing to sample phonemesoccurring in non-picturable or hard-to-readwords.

Speech sounds may also be elicited byasking the child to repeat words, phrases,sentences, or nonsense words. The imitationmode permits the testing of a wider range ofphonemes; however, it has been argued thatthe child may indeed imitate the model, re-vealing what he can say rather than what henormally says.

Evidence that imitative ellcitation tech-ques yield lower error rates than spontane-

r AS is mixeL:. In a study of 100 normal chil-dren with a mean age of 3 years 9 months,Templin (1947) found no differences betweenspontaneous (picture) and imitation modes fwords. Siegel, Winitz, and Conkey (1963), .a study of 100 normal kindergartners, reporteda slightly lower error rate on imitation foronly eight out of 40 speech sounds or clusters(/-1, -r-, sk-, r-, tr-, -z/).

Ln studies of young, functionally speech-defective children (second grade or below),Morrison (1914) and Burkhard (1965) reportedno difference in spontaneous (picture) andimitative modes; Fitschen (1965) and Snowand Milisen (1954), however, found fewererrors in imitation. Snow and Milisen alsotested seventh and eighth grade children withfunctional speech defects and found the dif-

ference in spontaneous and imitation modesto be far greater for older children than younger.

Several other variations in test materialsand procedure may affect error rates. Foryoung speech-defective children, Fitschen(19651 and Burkhard (1965) found that the imi-tation of phrases or sentences produced highererror rates than imitation of single words.Scott and Milisen (1954), in a study of 64functionally speech-defective children rangingin age from 4 to 14, found imitation of non-sense syllables (CV) to yield lower error ratesthan spontaneous, elicitation of words, ardfurthei, that imitation error rates were lowerwhen the child could see the experimentersaying the nonsense syllable than when thechild could only hear the syllable. The ef-fect of nonsense (or unfamiliar) words on nor-mal children's articulation errors has not beeninvestigated, although makers of imitationtests have attempted to use only high-frequencywords.

Venezky, Chapman, and Calfee (in prep-aration), in a study of the articulation errorsof normal kindergarten and first grade children,found error rates higher for repetition of wordspresented on tape than for repetition of wordsspoken by an experimenter facing the child.Higher error rates were observed for onespeaker (female) on tape than the other (male).

It would appear that the following factorsaffecting error rate must be taken into accountin interpreting differences in mode of elicita-tion: population from which sample is drawn(age; normal versus functionally speech de-fective); tasI (spontaneous production versusimitation); stimulus characteristics (real word,sentence, nonsense word); presentation, forimitation tasks (tape; live, experimenter'sface concealed or visible); the environmentof the speech sound tested (word or utteranceinitial, medial, and final; occurrence in diph-thong or cluster); and the phonemes tested.

7

1

In the present study, the particular pho-nemes tested and their word positions andcluster environments were held constant; theeffect on error rate of picture, word, sentence,and nonsense word elicitation modes was

2

studied for four age groups of normal, mono-lingual speakers (preschool, kinde:- irten,first grade, and college). Of particular inter-est were possible differences in imitationmode condiLlons.

METHOD

DESIGN

A 4 x 5 x 5 factorial design was uo.,ri:group x mode x sound class, with ropeatecimeasures on the last two factors. The fourgroups of Ss participating in the study werepreschoolers, kindergartners, first graders,and college students. Each S was tested infive elicitation mode conditions: picture,picture again, real word, nonsense word, andsentence. Five classes of sounds were de-fined for each stimulus list: initial conso-nant, initial consonant cluster, final conso-nant, final consonant cluster, and vowel.

STIMULUS MATERIALS

Fifteen common nouns representing con-crete objects were drawn from articulation

test lists used In a previous study; all wereof intermediate pronunciation difficulty forkindergarten and first grade children (Venezky,Chapman, & Calfee, in preparation). Five setsof stimulus materials were derive0 from thesewords: two sets of slides which were colorphotographs of the 15 objects named (pictureconditions 1 and 2); the word list itself; 15nonsense words formed by permuting thevowels and final consonants or consonantclusters of the original test items; and 15sentences of 6 words or less each containingan embedded test word) The real words,nonsense words, and sentences used arelisted in Table 1.

'One final consonalit in the nonsenselist was incorrectly recorded as /1/ ratherthan /r /.

Table 1

Stimulus Materials

SentencesWords and Nonsense

Pictures Words

I drank a glass of milk.The long string broke.The king's crown is silver.The frog Jumped into the water.Grandma gave me a scarf today.His new shirt is. blue.Susie's mouth is open.My truck has four wheels.They built a church over there.A star twinkled in the sky.My thumb hurts very much.Daddy brought flowers home.Sandy got a blouse for Christmas.My tooth fell out.The grass grows fast in summer.

glass /gla-t/string /streee/crown /kn1/frog /free o/scarf /skaus/shirt /gals/mouth /marf/truck /tram/church /6os/star /stave's/thumb /eauk/flowers /flaug/blouse /bluC/tooth /tas/grass /gran/

3

Each of the five sets of stimuli were in-dependently random-ordered. The verbal Ma-terials were recorded on an Ampex 1100 taperecorder with a Shure lavaliere microphone at3 1/2 ips by a male speaker of MidwesternEnglish with previo as recording experience.Two practice items preceded each recordedlist. Inter-item intervals were approximately3 sec. for the word and nonsense conditions;enough time to repeat the sentence twice(approximately 5 sec.) was left between sen-tences.

PROCEDURES

Groups of Ss were pretrained at least aday before testing on the slides to insure theelicitation of the appropriate labels. Theexperimenter (E) presented slides with aCarousel projector and provided correctivefeedback to the labels chorused by the group.

In individual testing sessions, slideswere presented in a hand-held battery-operatedviewer and taped materials were played throughstereo earphones by a Uher 5000 tape recorder.Responses were recorded on a second Uher5000 at 3 1/2 ips using a Shure lavalieremicrophone.

Brief instructions to name or repeat pre-ceded each condition. The order of conditionsfor a S withi.i an age group was determined bysequential assignment of Ss to rows in Latin-square blocks of the 74 possible orderings ofpicture, real word, nonsense word, and sen-tence. In the picture conditions, each Snamed the slides in two different randomorders, with an intervening presentation inwhich E named the slides.

SUBJECTS

The 64 Ss were drawn from University ofWisconsin students, first graders and kinder-gartners attending Nakoma Elementary school(an upper-middle class school in Madison,Wisconsin), and four-to-five year old pre-schoolers attending Child Development, Inc.,day-care centers in Madison and Middleton,Wisconsin. The original design called for24 Ss in each age group; mechanical recordingdifficulties resulted in the loss of data for32 Ss. For the remaining Ss in each agegroup, order of conditions was approximatelycounterbalanced. In the data reported here,24 adults, 10 first graders, 8 kindergartners,and 22 preschoolers are represented.

DATA TRANSCRIPTION

Errors of pronunciation were independentlytranscribed in broad phonetic notation (IPA) bytwo experienced linguists; a third transcriberresolved any disagreements between the twotranscribers .2

Pronunciations of real words attributableto dialect variation were treated as correct.These included post-vocalic /r/-deletion instar, shirt, flowers, church, and scarf; sub-stitution of /a/ for /3/ in frog or /3/ for /a/in star and scarf; and substitution of /a/ for/m/ in glass and grass. The same criteriawere applied to the nonsense list in order toprovide comparable error scoring; post-vocalic /r/-deletion was accepted in /stay a-z/,/gl eV, /g40/, and /marf/; /a/ or /3/ wasaccepted in /aos/, /tram/, and /marf/; /a/was accepted in /str.W and /fr0/.

ERROR CLASSIFICATION

Substitution or deletion errors were clas-sified according to whether they occurred inan initial consonant (IC) or initial consonantcluster (ICL); the vowel (V), including /al asa single vowel phoneme; or final consonant(FC) or final consonant cluster (FCL). An in-sertion error was scored as a cluster error ifit occurred medially in the cluster. A pre-vocalic insertion was classified as an IC (orICL) error if the insertion was consonantal;if a vowel, it was classified as a V error.Similarly, a post-vocalic insertion was clas-sified as a V or FC (FCL) error depending onwhether it was a vowel or a consonant.

For each S, a percentage score for eachof the five error types was derived by dividingthe number of phoneme errors by the totalnumber of phonerrE.s for that classificationand then multiplying by 100. Thus a scorerepresented per cent error of total possibledeletion and substitution errors. Insertionerrors, which are not taken account of incomputing the total possible error, were rare.The total number of phonemes for each errortype in each stimulus list were IC, 5; ICL,21; V, 16; FC, 14; FCL, 2. When S failedto produce the expected word or nonsense syl-lable, or when noise made the response unin-telligible, the number of phonemes in eachdivisor was accordingly reduced.

aIPA, or the International Phonetic Alpha-bet, is one of several systems for representingspeech sounds in any language.

10

RESULTS

The mean per cent error for each agegroup, mode, and sound class in the 4 x 5 x5 design is shown in Table 2. No errors weremade on the two phonemes exemplifying theFCL sound class in any condition except thenonsense mode. It Is clear, then, that FCLerror rates we id be significantly lower thanall other sound class error rates in real wordmodes, and that the nonsense mode woulddiffer significantly from the real word modesin FCL error rate. Thus a significant modeby sound class interaction could arise from

Table 2

FCL error scores alone. It was decided to testfor additional sources of main effect and in-teraction by deleting the FCL sound class fromthe analysis of variance; college students'scores were also deleted.3

A 3 x 5 x 4 unequal-n analysis of vari-ance, age group by mode by the remaining

3College Ss were omitted from the analysissince it was pccsible that their relatively error-free performance might attenuate effects ofmode and sound class present for the children.

Mean Per Cent Error by Sound Class andElicitation Mode for Each Age Group

Group ModeSound Class

IC ICL V FC FCL

Preschool Picture 1 9.8 6.3 5.4 3.1 0.0(n=22) Picture 2 10.8 6.3 4.6 3.9 0.0

Word 10.0 6.3 4.3 5,2 0,0Sentence 5.4 4.6 2.6 4.5 0.0Nonsense 7.3 5.4 12.0 16.9 20.4

Kinder- Picture 1 15.4 1.3 2.4 4.1 0.0yarten Picture 2 5.0 4.7 .9 4.7 0.0(n =8) Word 10.0 1.2 0.0 .9 0.0

Sentence 2.5 .6 .8 0.0 0.0Nonsense 7.5 3.6 4.7 13.4 12.5

First Picture 1 4.0 1.4 1.2 .7 0.0grade Picture 2 4.0 1.4 0.0 0.0 0.0(n=10) Word 6.0 1.4 1.3 2.1 0.0

Sen. ze 2.0 .5 1.2 .7 0.0Nonsense 4.0 6.7 3.1 8.6 10.0

College Picture 1 0.0 .6 0.0 0.0 0.0(n=24) Picture 2 0.0 0.0 0.0 0.0 0.0

Word 0.0 .2 0.0 0.0 0.0Sentence 0.0 0.0 0.0 0.0 0.0Nonsense .8 .2 1.3 4.2 8.3

5

11

Table 3

Mean Per Cent Error by Sound Class and Elicitation Mode,Averaged over 40 Children

Sound Class Averagewithout

Mode IC ICL V FC FCL Average FCL

Picture 1 9.5 4.1 2.7 3.8 0.0 4.0 5.0Picture 2 8.0 4.8 3.1 2.7 0.0 3.7 4.6Word 9.0 4.0 3.6 2.7 0.0 3.9 4.8Nonsense 6.5 5.4 14.1 8.3 16.2 10.1 8.6Sentence 4.0 2.7 2.7 1.9 0.0 2.3 2.8

Average 7.4 4.2 5.2 3.9 3.2 4.7

sound classes, with repeated measures on thelast two factors, was run on the children'sper cent error scores. Significant effectswere found for age group (F (2, 37) = 5.06,p < .05), sound class (F (3, 111) = 4.99, p <.01), mode of elicitation (F 148) = 12.31,P < .01), and the mode by sound class inter-action (F (12,444 = 5.84, p < .01). (Repeatedmeasures effects were also significant, p <.05, when the Geisser-Greenhouse correctionto degrees of freedom was applied.) No otherinteractions were significant.

The significant age group effect arosefrom a decrease in per cent error with in-creasing age. Average per cent error for asound class (excluding FCL) in a mode was6.7 for preschoolers, 4.1 for kindergartners,and 2.5 for first graders.

The average per cent error in a soundclass is shown for each mode of elicitationcondition in Table 3. The error rate in thenonsense condition was approximately twicethose of the four real word conditions; threetimes greater, when the FCL sound class isincluded. A post-hoc Scheffe comparisonshowed the nonsense condition error rate tobe significantly greater than the average ofthe real word conditions (p < .05, df = 4, 148).

The main effect of sound class wouldappear to arise from the overall greater errorrate on IC phonemes.4 The post-hoc Scheffetest of IC versus the average of the other

tor the 15 words used in this study, priornormative data from 162 kindergartners and148 first graders also showed higher error ratesfor initial consonants than other sound classes.In studies where the same sounds or clustersare tested both initially and finally, however,error rates typically increased from initial tofinal position (Templin, 1957).

t.)

sound classes, however, did not reveal a sig-nificant difference.

The significant mode by sound class inter-action arose from an error pattern differencebetween the nonsense condition and all of theother conditions. When the nonsense mode isremoved from the analysis, the interaction dis-appears. The FCL sound class was omittedfrom these analyses but, if included, wouldhave contributed strongly to the interaction;FCL error rates were zero everywhere but thenonsense condition. The comp'ete interactionis displayed in Fig. 1; each bar in that graphrepresents the difference between error ratefor a single sound class and the average errcrrate, for a particular mode. In the real wordmodes, most errors are made on initial conso-nants, fewer on vowels and final consonants,and fewest on final consonant clusters. Inthe nonsense word condition, fewest errorsare made on initial consonants and clustersand most on vowels and final consonant clusters.That is, the error pattern in the nonsense modewas the reverse of that found in the real wordmodes. In the college student data, few er-rors were observed except in the nonsensecondition; there, as in the children's data,error rates increased in final position.

FAILURES TO ELICIT TEST WORDS

The test word was elicited from Ss 98.2%of the time overall. In Table 4 is shown thepercentage of times test words were not givenout of the total opportunities to give them foreach mode and group, including college Stu -dents. The picture conditions account for 91%of the failures to elicit test words from Ss;most failures resulted from the S giving aninappropriate label for a picture. Even whenthe appropria.,..; labels were given by E

12

+7 -a)4-rorx

0

N

N

IC

Nonsense Mode

Picture 1 Mode

Picture 2 Mode

Word Mode

Sentence Mode

ICL

1111111-

V

Sound Class

FCL FC

Fig. 1. Difference Between Error Rate for a Sound Class and Av2rage Error Rate for Each Mode.

Table 4

Per Cent Failure to Give Test Word, by Group and Mode of Elicitation(Preschool, n = 22; Kindergarten, n = 8; First Grade, n = 10; College, n = 24)

Mode of ElicitationG ,up Picture 1 Picture 2 Word Sentence Nonsense All modes

Preschool 7.9 4.8 2.1 0.3 0.0 3.0Kindergarten 9.2 5.8 0.0 0.0 0.0 3.0First Grade 5.3 0.7 0.0 0.0 0.0 1.2College 1.7 0.6 0.0 0.0 0.0 0.4

All Ss 5.3 2.7 0.7 0.1 0.0 1.8

13

immediately prick to the list (second picturecondition), 5 to 6% of the test words were notappropriately elicited in preschool and kinder-garten Ss. The word repetition mode is clearlypreferable to picture elicitation for normalchildren, since error rates are comparable andmore complete data are assured.

NOISENSE WORD ERRORS

The high error rate on nonsense wordsmay arise from failure to perceive some pho-nemes correctly. To account for the presentdata, it must be further assumed that misper-ception of sounL:s in nonsense words is morelikely for souncti in final position than thosein initial position.

An indirect check of this explanationwas carried out by asking an independentsample of 10 adults to listen to the nonsenseword tapes and write down what they heard inconventional spelling. Consonant spellingswere examined for evidence of misperception.Ss spelled /8/ as f 8 out of 10 times as bothIC and FC. Only one other spelling suggestedIC misperception, and four others suggested FCmisperception.5 These data would predict an

5r:he spellings were ch for initial and,it firal position, z for /s/, ng twice for /n/,and celetion for /k/.

8

18% error rate on IC items for college students(.8% obtained), an 8.5% error rate on FC (4.4%obtained) and 0 error rates on ICL and FCL(.2% and 8.3% obtained). The fit of predictedand obtained data is far from good; although/0/ in particular may have presented identi-fication problems in the absence of meaningcues, the supposed errors of misperceptiondo not show the same increase in final posi-tion that the nonsense articulation errorsexhibit.

An alternate, and simpler, explanation isthat short-term memory for the nonsense pho-neme string decays rapidly during production,leading to an increase in error rate for thelatter part of the string. One finding in par-ticular bears out this explanation: deletionerrors of final consonants show an unusualincrease in the nonsense condition, account-ing for 75% of the deletion errors occurringfor the FC error type. (Nonsense FC substitu-tion and insertion errors account for approxi-mately 50% of such errors in FC position.]Further, substitution errors for the single FCLitem occur only in the nonsense condition.The deletion errors for the ICL items, in con-trast, occur with equal frequency in all elici-tation modes.

A direct test of the memory-decay hypoth-esis for the increasing error effect could becarried out by examining the probability ofmispronunciation of a final consonant as afunction of the number of preceding phonemes.

14

IV

SUMMARY AND CONCLUSIONS

The most striking finding of the study isthat nonsense word stimuli yield an error ratethat is not only higher than the error rate forphonologically similar real word stimuli, butthat also increases sharply from initial tofinal position.

Misperception of the nonsense stimuliwas rejected as an explanation of the non-sense data, since indirect evick'nce indicatedit could account only for a general error rateincrease rather than the pattern of increasingerrors observed. The hypothesis of short-term memory decay was advanced as a simplerexplanation of the increasing error effect.On this explanation, word meaning wouldeliminate short-term decay of phonologicalinformation for real words, since the word isstored as a single unit which is used to re-trieve phonological (or production) informationfrom long-term memory. Although the alterna-tive possibility of better phonological chunk-ing of familiar phoneme sequences cannot beruled out, it arnaara implausible in the facecf evidence that listeners sharply differen-tiate nonsense English words from non-Englishwords but not from real English words.

Further research is required to establishthe extent and probable cause of the. increas-ing error effect in nonsense word repetition.It is clear from the present data, however,that articulation test makers should take ex-ceptional care in avoiding test words likelyto be unknown to the children being tested.When two groups differing in size and struc-

GPO SIIINI17-1

ture of vocabulary (e.g., upper and lowersocio-economic status children) are comparedon the same articulation test lists, differen-tial articulation scores may be found solelybecause more test words constitute nonsenseitems for one group. Reported correlations ofarticulation scores with socio-economicstatus and I.Q. must be viewed with someskepticism until it can be established thatthe test words were equally familiar to allchildren tested. Theoretical conclusions aL,to the relative difficulty of initial and finalsounds are, of course, also biased by the in-clusion of nonsense test items.

The failure to find different error ratesor error patterns among the three real wordelicitation modes leaves the choice of modeopen on these, counts. The repetition modes,however, allow a wider choice of test wordand a higher probability of elicitation thanthe picture elicitation technique. Testingwith single words has the advantage of shorten-ing test time; testing with sentences confersmore semantic context and the possibility ofvarying a test word' phonetic environment.

Finally, it shoulc' be noted that the datalay to rest again the speculation that childrenare significantly more likely to mispronouncespontaneously produced words than to mis-pronounce imitated words. Also unsupportedby the data is the extension of that argumentto predict higher error rates for repetition ofwords in sentences than for repetition ofwords in isolation.

159

REFERENCES

Burkhard, M. J. The effect of phonetic en-vironment of misarticulations of the /s/sound. Unpublished M.S. thesis, Uni-versity of Wisconsin, 1965.

Fitschen, R. A. Investigations of the mis-articulation patterns of the /8/ sound infirst grade children. Unpublished M.A.thesis, University of Wisconsin, 1965.

Morrison, C. E. Speech defects in young chil-dren, The Psychological Clinic, 1914,8, 138-142.

Scott, D. A., and R. Milisen. The effect ofvisual, auditory and combined 1,;.sual-auditory stimulation upon the speechresponses of defective speaking children.Journal of Speech and Hcaring Dis-orders, Monograph Supplement, 1954,4, 37-43.

Siegel, G. M., H. Winitz, and H. Conkey. Theinfluence of testing instrument on articu-latory responses of children. Journal

GPO 022-417-2

of Speech and Hearing Disorders, 1963,28, 67-76.

Snow, K., and R. Milisen. The influence oforal versus pictorial presentation uponarticulation testing results. Journal ofSpeech and Hearing Disorders, Mono-graph Supplement, 1954, 4, 29-36.

Templin, M. C. Spontaneous versus imitatedverbalization in testing articulation inpreschoo, children. Journal of SpeechDisorders, 1947, 12, 293-300.

Templin, M. C. Certain language skills inchildren. Institute of Child WelfareMonograph Series No. XXVI. Minnea-polis: University of Minnesota, 1957.

Venezky, R. L., R. S. Chapman, and R. C.Calfee. Articulation in young children.Technical report from the WisconsinResearch and Development Cemer forCognitive Learning, the University ofWisconsin (in preparation).

16