verbal working memory and speech errors eleanor drake 15 th february 2008
TRANSCRIPT
Verbal working memory Verbal working memory and speech errorsand speech errors
Eleanor DrakeEleanor Drake
1515thth February 2008 February 2008
OutlineOutline
• IntroductionIntroduction• Working memoryWorking memory
– Phonological loopPhonological loop– Measuring functionMeasuring function
• Short-term memory and speech errorsShort-term memory and speech errors• Saito and Baddeley (2004) exp. 1Saito and Baddeley (2004) exp. 1• Current studyCurrent study
– AimAim– Speech taskSpeech task
• Results and interpretationResults and interpretation– Correlatory findingsCorrelatory findings
• Results and interpretationResults and interpretation
IntroductionIntroduction
• Saito and Baddeley (2004) report Saito and Baddeley (2004) report correlation between vSTM correlation between vSTM performance and speech error rate.performance and speech error rate.
• Current study aim = determine if Current study aim = determine if correlation replicable using different correlation replicable using different error-elicitation paradigm. error-elicitation paradigm.
Working MemoryWorking Memory
• Responsible for temporary storage and Responsible for temporary storage and processing of information.processing of information.
• Baddeley and Hitch (1974) model has been Baddeley and Hitch (1974) model has been highly influential.highly influential.
• Proposed a system involving a central Proposed a system involving a central executive + subservient modality specific executive + subservient modality specific visuo-spatial sketch pad and phonological loop visuo-spatial sketch pad and phonological loop (and an episodic buffer, in later versions).(and an episodic buffer, in later versions).
• Phonolgical loop function focus of current Phonolgical loop function focus of current study.study.
Working memory modelWorking memory model
Phonological loopPhonological loop
• Responsible for short-term storage, and checking Responsible for short-term storage, and checking of verbal materialof verbal material
• Bi-componential:Bi-componential:– Phonological store:Phonological store:
• storage of phonologically encoded informationstorage of phonologically encoded information• information subject to decay (c. 1.5-2s)information subject to decay (c. 1.5-2s)
– Subvocal articulatory rehearsalSubvocal articulatory rehearsal• Cycles phonologically encoded information to refresh Cycles phonologically encoded information to refresh
storestore• Recoding non-phonological (i.e., printed word) into Recoding non-phonological (i.e., printed word) into
phonological formphonological form• Evidence- “word-length effect”, “articulatory Evidence- “word-length effect”, “articulatory
suppression”suppression”
Measures of phonological loop Measures of phonological loop function function
•Forward serial recallForward serial recall– (e.g., Wechsler 2003 digit span)(e.g., Wechsler 2003 digit span)– Taken to relate to rate of articulatory Taken to relate to rate of articulatory
rehearsal (i.e., individuals with a higher rehearsal (i.e., individuals with a higher score rehearse more rapidly)score rehearse more rapidly)
– Repetition or matchingRepetition or matching– Measures Measures verbalverbal short-term memory short-term memory
Measures of phonological loop Measures of phonological loop functionfunction
• Nonword repetitionNonword repetition– Repetition of phonotactically acceptable but Repetition of phonotactically acceptable but
lexically empty syllable stringslexically empty syllable strings– Minimize contribution of long-term Minimize contribution of long-term
phonological, lexical, and semantic knowledge phonological, lexical, and semantic knowledge (Gathercole & Baddeley, 2003)(Gathercole & Baddeley, 2003)
– Greater reliance on auditory discrimination, Greater reliance on auditory discrimination, motor planning and execution (Edwards and motor planning and execution (Edwards and Lahey, 1998)Lahey, 1998)
– Measures Measures phonologicalphonological short-term memory short-term memory
STM and speech production STM and speech production errorserrors
• Why might there be a correlation?Why might there be a correlation?– Qualitatively similar errors = common locus in Qualitatively similar errors = common locus in
information processing? information processing? • E.g., Ellis, 1980; Page et al, 2007E.g., Ellis, 1980; Page et al, 2007• Similarity can be simulated via computational Similarity can be simulated via computational
modelling (Page and Norris, 1998a) modelling (Page and Norris, 1998a)
– ““Serial-ordering mechanism” malfunction as Serial-ordering mechanism” malfunction as origin of normal speech errors origin of normal speech errors • E.g, Nespoulos et al. 1984E.g, Nespoulos et al. 1984
– vSTM lower in PWS vSTM lower in PWS • (Bosshardt 1990; Bosshardt 1993)(Bosshardt 1990; Bosshardt 1993)
Saito and Baddeley (2004)Saito and Baddeley (2004)
• Experiment 1Experiment 1– Does elicited speech error rate relate to Does elicited speech error rate relate to
phonological loop function?phonological loop function?•Measured PL function via serial digit recall Measured PL function via serial digit recall
– Auditory distractor technique as error-Auditory distractor technique as error-elicitation paradigmelicitation paradigm•Target word (e.g., “shizuoka”) produced in Target word (e.g., “shizuoka”) produced in
response to tone.response to tone.•Cue tone replaced 1/10 by similar distractor Cue tone replaced 1/10 by similar distractor
word (e.g., “shiozuke”), 1/10 by dissimilar word (e.g., “shiozuke”), 1/10 by dissimilar distractor distractor
Saito and Baddeley (2004)Saito and Baddeley (2004)
Saito and Baddeley (2004)Saito and Baddeley (2004)
• FindingsFindings– Error rate = .031/ .291 (dissimilar/ similar condition)Error rate = .031/ .291 (dissimilar/ similar condition)
– Speech error rate correlated with digit span task Speech error rate correlated with digit span task performance; r = - 0.33, p < .05.performance; r = - 0.33, p < .05.
• InterpretationInterpretation– Both vSTM and speech task performance share Both vSTM and speech task performance share
predication on a “phonological planning factor”predication on a “phonological planning factor”
Current studyCurrent study
• Does the correlation hold for: Does the correlation hold for:
– another error-elicitation paradigm?another error-elicitation paradigm?•Uses Wilshire’s (1999) tongue-twister taskUses Wilshire’s (1999) tongue-twister task
– different tests of auditory short-term different tests of auditory short-term memory?memory?•Uses serial digit recall, serial digit matching, Uses serial digit recall, serial digit matching,
and NWRand NWR
Speech errorsSpeech errors
• Task involves reading monosyllabic Task involves reading monosyllabic wordswords
• Errors (i.e., deviations from the speech Errors (i.e., deviations from the speech plan) assumed to occur at level ofplan) assumed to occur at level of– FormulationFormulation
• Phonological encodingPhonological encoding
– ArticulationArticulation•Execution fault in control of muscular processes Execution fault in control of muscular processes
– Not conceptualisationNot conceptualisation
Speech error elicitationSpeech error elicitation
• Wilshire (1999)Wilshire (1999)• 64 “tongue-twisters”64 “tongue-twisters”
– 4 monosyllabic words on screen, sequence 4 monosyllabic words on screen, sequence to be repeated 4 times, at rate of 100 wpmto be repeated 4 times, at rate of 100 wpm
– 32 control, 16 ABBA onset, 16 ABAB onset32 control, 16 ABBA onset, 16 ABAB onset– Of 32 non-control items 16 alliterating Of 32 non-control items 16 alliterating
similar, 16 alliterating dissimilar (near even similar, 16 alliterating dissimilar (near even distribution)distribution)
– Egs.- Egs.-
Quadruple examplesQuadruple examples
• ControlControl- CUB TIME DATE SIN - CUB TIME DATE SIN • ABBA dissimilarABBA dissimilar - BED COUGH CARD - BED COUGH CARD
BEEF BEEF • ABBA similarABBA similar - DIRT BUS BOOT - DIRT BUS BOOT
DOSEDOSE• ABAB dissimilarABAB dissimilar - TIN MAP TYPE - TIN MAP TYPE
MOON MOON • ABAB similarABAB similar - SAP TIFF SURF TOP - SAP TIFF SURF TOP
Other parametersOther parameters
• Serial digit recall repetitionSerial digit recall repetition
• Serial digit recall matchingSerial digit recall matching
• Nonword repetitionNonword repetition
• Digit articulation timeDigit articulation time
• N =15N =15
Tongue-twister resultsTongue-twister results
• Error-rate:- 16.7% quadruples errorfulError-rate:- 16.7% quadruples errorful
• Error-rate by individual words within Error-rate by individual words within quad:-quad:-– Overall = 1.5% Overall = 1.5% – Control = .71%Control = .71%– ABBA dissimilar = .63%ABBA dissimilar = .63%– ABAB dissimilar = 1.77%ABAB dissimilar = 1.77%– ABBA similar = 1.77%ABBA similar = 1.77%– ABAB similar = 4.13%ABAB similar = 4.13%
Tongue-twister resultsTongue-twister results
• Analysis of varianceAnalysis of variance– A/S significantly more error prone thanA/S significantly more error prone than
•Control (t = 3.66, p < .005)Control (t = 3.66, p < .005)•A/D (t = 3.51, p < .005)A/D (t = 3.51, p < .005)•(control v A/D; no significant difference)(control v A/D; no significant difference)
– ABAB significantly more error prone thanABAB significantly more error prone than•Control (t = 4.138, p < .005)Control (t = 4.138, p < .005)•ABBA (t = 3.032, p < .05)ABBA (t = 3.032, p < .05)•(control v ABBA; no significant difference) (control v ABBA; no significant difference)
Tongue-twister resultsTongue-twister results
• Word position:Word position:– Word 3 (e.g., sap tiff Word 3 (e.g., sap tiff surf surf top) significantly more error top) significantly more error
prone (for ABAB sub-set)prone (for ABAB sub-set)Percentage of syllables produced incorrectly by quad type and position within quad
0
1
2
3
4
5
6
7
8
9
10
control ABBA dissimilar ABAB dissimilar ABBA similar ABAB similar
Quad type
per
cen
tag
e o
f sy
llab
les
pro
du
ced
in
corr
ectl
y
words1
Word 2
Word 3
Word 4
Tongue-twister resultsTongue-twister results
• ReiterationReiteration
– 11stst recitation less error prone recitation less error prone•2 v 1: t = 2.68 2 v 1: t = 2.68 •3 v 1: t = 2.48 3 v 1: t = 2.48 pp < .05 all cases < .05 all cases•4 v 1: t = 2.634 v 1: t = 2.63
– No significant differences amongst other No significant differences amongst other recitation numbers.recitation numbers.
Tongue-twister results Tongue-twister results
• SummarySummary
– Error rate higher whenError rate higher when•ABAB … and/ orABAB … and/ or•A/SA/S•Word position 3Word position 3
– Error rate significantly lowerError rate significantly lower•11stst recitation recitation
Discussion speech taskDiscussion speech task
• Speech error rate (1.5%) Speech error rate (1.5%) – Wilshire (4.5%)Wilshire (4.5%)
•Heterogeneity of participants, experimental Heterogeneity of participants, experimental conditions, error-sensitivity of scorerconditions, error-sensitivity of scorer
– Saito and Baddeley (29.6%)Saito and Baddeley (29.6%)• In error-eliciting conditionsIn error-eliciting conditions
– Spontaneous speech (0.1-0.2%) Spontaneous speech (0.1-0.2%) (Garnham et al. (Garnham et al.
1981)1981) •Suggests paradigm does elicit errorsSuggests paradigm does elicit errors
Discussion speech taskDiscussion speech task
• A/S > A/D:- phonological similarity A/S > A/D:- phonological similarity effecteffect– Conforms with Wilshire (1999)Conforms with Wilshire (1999)– Cf. PSE in serial recallCf. PSE in serial recall
• ABAB > ABBA:-ABAB > ABBA:-– Contrary to Wilshire (1999)Contrary to Wilshire (1999)– Contrary to Sevald and Dell (1994)Contrary to Sevald and Dell (1994)– Design flaw? Design flaw? – Requires control experimentRequires control experiment
Discussion speech taskDiscussion speech task
• Word 1 > words 2/ 3/ 4Word 1 > words 2/ 3/ 4– In ABAB subsetIn ABAB subset– Point at which A and B onsets are Point at which A and B onsets are
equiprobableequiprobable– Experimental design -> preference for Experimental design -> preference for
ABBA?ABBA?
Discussion speech tasksDiscussion speech tasks
• Recitation 1 < recitation 2/ 3/ 4Recitation 1 < recitation 2/ 3/ 4– Consistent with Wilshire (1999)Consistent with Wilshire (1999)
• She proposes explanationsShe proposes explanations
• Phonological planning fatigue?Phonological planning fatigue?
• Strategies by which plan re-used?Strategies by which plan re-used?
• Interference/ decay (cf. WM model)Interference/ decay (cf. WM model)
– Another suggestionAnother suggestion• Overt articulation having occurred impacts on Overt articulation having occurred impacts on
psychological/ physiological representations of the psychological/ physiological representations of the phrasephrase
Discussion speech tasksDiscussion speech tasks
• How?How?– Feedback from production creating Feedback from production creating
interefernceinterefernce– Overt articulation (opposed to Overt articulation (opposed to
articulatory rehearsal) ->articulatory rehearsal) ->•Proprioceptive or sensorimotor feedbackProprioceptive or sensorimotor feedback
•Auditory feedbackAuditory feedback
Discussion speech tasksDiscussion speech tasks
• Proprioceptive/ sensorimotorProprioceptive/ sensorimotor– Suggested in stuttering literature (e.g., Suggested in stuttering literature (e.g.,
Max et al. 2004)Max et al. 2004)– A/S more error-prone because of gestural A/S more error-prone because of gestural
similarity (cf. Articulatory Phonology)similarity (cf. Articulatory Phonology)– PSE becomes apparent with overt PSE becomes apparent with overt
articulation (Oppenheim and Dell, 2007)articulation (Oppenheim and Dell, 2007)– Non-canonical errors Non-canonical errors
Discussion speech tasksDiscussion speech tasks
• Auditory feedbackAuditory feedback– Gestural adjustment to correct perceived Gestural adjustment to correct perceived
speech errorspeech error– Conventional PSE from participant’s self-Conventional PSE from participant’s self-
generated auditory inputgenerated auditory input– Cf. effect of similar auditory distractor in Cf. effect of similar auditory distractor in
Saito and Baddeley studySaito and Baddeley study– Cf. stuttering literature on error-inducing Cf. stuttering literature on error-inducing
properties of own speech (Stuart et al., properties of own speech (Stuart et al., 1996) 1996)
Correlatory resultsCorrelatory results
• Descriptive stats and correlation matrix Descriptive stats and correlation matrix attachedattached
• Tongue-twister results correlate withTongue-twister results correlate with– Serial repetition (r = -.527)Serial repetition (r = -.527)– Serial matching (r = -.673)Serial matching (r = -.673)
• Serial recall results also correlate withSerial recall results also correlate with– Digit articulation time (r = -.671)Digit articulation time (r = -.671)– Age (r = -.570)Age (r = -.570)
• NWR no significant correlationsNWR no significant correlations
Correlations discussionCorrelations discussion
• Serial digit repetition and matchingSerial digit repetition and matching– Correlation suggests shared underlying factor(s)Correlation suggests shared underlying factor(s)– Phonological planning ?Phonological planning ?
• Saito and Baddeley (2004) factor analysisSaito and Baddeley (2004) factor analysis
• Matching doesn’t require digit articulation motor Matching doesn’t require digit articulation motor executionexecution
– Correlation with digit articulation timeCorrelation with digit articulation time• Serial digit recall relates to speed of subvocal Serial digit recall relates to speed of subvocal
articulatory rehearsal articulatory rehearsal
• S & B DAT relates to motor execution because more S & B DAT relates to motor execution because more repetitions?repetitions?
Correlations discussionCorrelations discussion
• NWR no correlationNWR no correlation– No inter-rater reliability assessmentNo inter-rater reliability assessment– Individual variation in auditory discriminationIndividual variation in auditory discrimination
•Novel stimulus -> greater reliance on auditory Novel stimulus -> greater reliance on auditory perception and processing abilitiesperception and processing abilities
– Doesn’t tap same processes as vSTM Doesn’t tap same processes as vSTM measuresmeasures•Sounds require novel phonological representationSounds require novel phonological representation
Correlations discussionCorrelations discussion
• NWR no correlationNWR no correlation– Contrary to findings of Hulme et al. Contrary to findings of Hulme et al.
(1991)(1991)•Hulme et al. familiarised participants with Hulme et al. familiarised participants with
nonwords before testingnonwords before testing
•So nonwords already had long-term So nonwords already had long-term phonological representation and phonological-phonological representation and phonological-articulatory speech planarticulatory speech plan
•Long-term memory contribution to serial Long-term memory contribution to serial recall tasks (& tongue-twister task?) recall tasks (& tongue-twister task?)
Correlations discussionCorrelations discussion
• vWM and speech errorsvWM and speech errors– Higher digit span score associated with fewer Higher digit span score associated with fewer
speech errorsspeech errors– Replicates findings of Saito and Baddeley Replicates findings of Saito and Baddeley
• Different explicit task demands of 2 error-elicitation Different explicit task demands of 2 error-elicitation paradigmsparadigms
– Correlation only exists for tasks involving real Correlation only exists for tasks involving real words words • Items which are already represented within the lexiconItems which are already represented within the lexicon
• PSE in both error-elicitation tasks implicates store?PSE in both error-elicitation tasks implicates store?
Correlations discussionCorrelations discussion
• vWM and speech errorsvWM and speech errors– Speech error-rate doesn’t correlate with Speech error-rate doesn’t correlate with
DATDAT•Although DAT involves articulation under Although DAT involves articulation under
pressurepressure•Although DAT indicator of articulatory rehearsal Although DAT indicator of articulatory rehearsal
rate (and hence capacity)rate (and hence capacity)
– Does correlation of speech-error rate and Does correlation of speech-error rate and vWM occur because performance on both vWM occur because performance on both tasks in predicated on robustness of verbal tasks in predicated on robustness of verbal representations ?representations ?
Discussion correlationsDiscussion correlations
• How?How?– Phonological representations vulnerable to Phonological representations vulnerable to
interference at a phonological level (PSE)interference at a phonological level (PSE)– Phonological representations subject to Phonological representations subject to
(limited) interaction from other levels of (limited) interaction from other levels of speech production system (e.g., Rapp & speech production system (e.g., Rapp & Goldrick, 2000)Goldrick, 2000)•Activation can be influenced by lexical Activation can be influenced by lexical
informationinformation•Feedback from an articulatory level Feedback from an articulatory level
Discussion correlationsDiscussion correlations
• EvidenceEvidence– Wilshire (1998) Wilshire (1998)
•Onset-effect for word but not nonword Onset-effect for word but not nonword tongue-twisters indicates lexical-level tongue-twisters indicates lexical-level contribution contribution
– Tehan and Lalor (2000)Tehan and Lalor (2000)•Serial recall tasks involve lexical accessSerial recall tasks involve lexical access
– Hulme et al. (1991)Hulme et al. (1991)•Serial recall better for newly-learnt words if Serial recall better for newly-learnt words if
participant learns sound + meaningparticipant learns sound + meaning
Discussion correlationsDiscussion correlations
• Explanation of correlatory pattern?Explanation of correlatory pattern?
– Serial recall correlates with DAT because Serial recall correlates with DAT because both predicated on speed of articulatory both predicated on speed of articulatory rehearsalrehearsal
– Serial recall correlates with error-rate Serial recall correlates with error-rate
because both predicated on strength of because both predicated on strength of phonological representationsphonological representations