Ginsborg & sloboda

http://pom.sagepub.comPsychology of Music

DOI: 10.1177/0305735607072654 2007; 35; 421 Psychology of Music

Jane Ginsborg and John A. Sloboda Singers' recall for the words and melody of a new, unaccompanied song

http://pom.sagepub.com/cgi/content/abstract/35/3/421 The online version of this article can be found at:

Published by:

http://www.sagepublications.com

On behalf of:

Society for Education, Music and Psychology Research

can be found at:Psychology of Music Additional services and information for

http://pom.sagepub.com/cgi/alerts Email Alerts:

http://pom.sagepub.com/subscriptions Subscriptions:

http://www.sagepub.com/journalsReprints.navReprints:

http://www.sagepub.com/journalsPermissions.navPermissions:

© 2007 Society for Education, Music, and Psychology Research. All rights reserved. Not for commercial use or unauthorized distribution. at Royal Northern College of Music on August 30, 2007 http://pom.sagepub.comDownloaded from

http://www.sempre.org.uk

http://pom.sagepub.com/cgi/alerts

http://pom.sagepub.com/subscriptions

http://www.sagepub.com/journalsReprints.nav

http://www.sagepub.com/journalsPermissions.nav

http://pom.sagepub.com

Singers’ recall for the wordsand melody of a new,unaccompanied song

421A RT I C L E

Psychology of Music

Psychology of MusicCopyright © 2007

Society for Education, Musicand Psychology Research

vol 35(3): 421‒440 [0305-7356

(200707) 35:3; 421‒440]10.1177⁄0305735607072654


J A N E G I N S B O R G RO YA L N O RT H E R N C O L L E G E O F M U S I C , M A N C H E S T E R , U K

J O H N A . S L O B O DA K E E L E U N I V E R S I T Y, U K

A B S T R A C T The nature of the relationship between words and music in memoryhas been studied in a variety of ways, from investigations of listeners’ recall forthe words of songs stored in long-term memory to recall for novel information setto unfamiliar melodies. We asked singers to perform an unaccompanied songfrom memory following deliberate learning and memorization of the words andmelody separately and together. Participants with high levels of musical expertisegave more accurate and fluent performances than those with lower levels ofmusical expertise, but only when they had memorized the words and melodytogether. While some errors were conjoint, such that erroneous recall of onecomponent – words or melody – affected recall for the other, a higher proportionwere separate, such that participants were able to preserve one component whenthey recalled the other erroneously. Words and melody are thus recalled inassociation with one another, so that retrieving one enables retrieval of the other,but are not integrated to the extent that failure to recall one accurately invariablyresults in failure to recall the other. Finally, more hesitations were made at theends of phrases than at the start or mid-phrase, suggesting that the formalstructure of a song provides a framework for recall. In conclusion, memorizingwords and melody together is an effective strategy, but perhaps only for singerswith high levels of musical expertise.

K E Y W O R D S : expertise, learning, melody, memorization, retrieval, strategy

In this article, we ask two questions. One is theoretical, the other practical.The answers to both are of interest to singers and teachers of singing:

1. What is the relationship between the words and melodies of songs inmemory? Are they stored and retrieved as separate components, inassociation with one another, or as a single, integrated unit?

sempre :



2. What is the best way for a singer to learn and memorize a song,assuming that both the words and melody are novel? Should the wordsand melody be memorized separately or together and, if separately,should the words or melody be memorized first?

We carried out our investigation of memory for songs using two methodsthat have not, so far as we know, been used before. First, we asked singers torecall – i.e. to produce – both the words and the melody of the song they hadmemorized, rather than to produce one or the other, or to make a judgementof familiarity on hearing the song. Second, we made a detailed time-basedanalysis of the location of errors in production and their nature: involvingone component only (words or melody) or conjoint (involving both words andmelody).

The theoretical question – what is the relationship between the words andmelodies of songs in memory? – has been asked by several groups of resear-chers, using a variety of methods. These differ from ours in several ways.

First, we are concerned with expert singers’ memory. Only two studies, bySloboda and Parker (1985) and Kilgour et al. (2000), have investigated theeffect of expertise (defined in terms of levels of musical training) on recall.The remainder involved the participation of non-musically trained adults(Rubin, 1977; Calvert and Tart, 1993; Rainey and Larsen, 2002), musicallytrained adults included in groups of non-musically trained adults (Wallace,1994), children (Calvert and Billingsley, 1998), or compared adults’ andchildren’s recall (Chazin and Neuschatz, 1990).

Second, we operationalize memory in terms of production: recall thatinvolves singing both the words and melody of the song. Sloboda and Parker(1985) investigated recall for melody only. The other studies referred to aboverequired participants to write down the words only (Rubin, 1977; Chazin andNeuschatz, 1990; Calvert and Tart, 1993, Experiment 2; Wallace, 1994;McElhinney and Annett, 1996; Kilgour et al., 2000) or to recite them(Calvert and Billingsley, 1998). Only Calvert and Tart (1993, Experiment 1)gave participants the choice of singing or reciting. Rainey and Larsen (2002)presented lists of names either spoken or sung and measured how longparticipants took to re-learn the lists a week later. Meanwhile, Serafine andher colleagues (1984, 1986; Crowder et al., 1990) investigated recognitionmemory, rather than recall, for words and melodies of songs.

Third, our focus is on recall for the words and melodies of songs that havebeen learned, rehearsed and memorized by singers using a notated musicalscore, but which are still ‘fresh’ in memory. Earlier studies used auditorypresentation only of unfamiliar words that were or were not paired withunfamiliar melodies, either once or several times (Serafine et al., 1984, 1986;Crowder et al., 1990; Calvert and Tart, 1993, Experiment 2; Wallace, 1994;McElhinney and Annett, 1996; Kilgour et al., 2000). Others presentedunfamiliar words or information with melodies stored in long-term memory,

422 Psychology of Music 35(3)



e.g. the characteristics of minerals sung to the tune of ‘Mary Had a LittleLamb’ (Chazin and Neuschatz, 1990); the names of baseball players set to thetune of ‘Pop Goes the Weasel’ and characters from Tolkien’s The Hobbit set tothe tune of ‘Yankee Doodle’ (Rainey and Larsen, 2002). A final group ofstudies investigated recall from very long-term memory, when both the wordsand melody presented would have been familiar, e.g. ‘The Star-SpangledBanner’ (Rubin, 1977) and the Preamble to the Constitution (set to music)(Calvert and Tart, 1993).

Taken together, the findings of the existing research indicate that adultparticipants have better recall for the words of songs when they are heardwith their ‘proper’ melodies than with different melodies or none (Rubin,1977), suggesting reliance on an existing association. Recall is likely to bebetter when participants rehearse (Calvert and Tart, 1993; Rainey andLarsen, 2002) and thereby strengthen an existing association between wordsand a familiar melody. Where a novel melody accompanies the words of asong, recall is likely to be better when the melody is repeated several times(Wallace, 1994; McElhinney and Annett, 1996; Kilgour et al., 2000), enablinga new association between them to be formed. While it may be harder forchildren to form new associations between words and melodies, particularlywhen the words are difficult to understand (Chazin and Neuschatz, 1990)and/or they are set to an unfamiliar melody (Calvert and Billingsley, 1998), itseems to be easier for expert musicians: Kilgour et al. (2000) found thatmusically trained adults had better recall for sung words than did their non-musically trained counterparts. This may well result from their superiorrecall for melodies, as shown in Sloboda and Parker’s (1985) study. Aspredicted by research on expertise generally (e.g. Ericsson, 1996), the moreone has experience of performing music, and the more one knows aboutmusic, the better one is able to encode and represent music and therefore torecall and perform it accurately from memory. For example, Gruson’s (1988)group study of pianists practising and learning unfamiliar pieces of musicshowed that experts were quicker than novices to learn new pieces moreeffectively, not because they used more efficient learning strategies for theparticular new pieces they practised, but because they had developed moreefficient learning strategies over many years of practising many pieces. Thesestrategies include drawing on an understanding of the formal structure ofthe music being learned. Ginsborg (1999) showed that four-fifths of segmentssung by expert singers while memorizing a song began at the start of aphrase (particularly the beginning of the song, half-way through the firstverse, the beginning of the second verse and the difficult final phrase); morethan half finished at the end of the same, or a subsequent phrase. Williamonand Valentine (2002) also showed that expert pianists were more likely thannovices to start and stop playing from memory at phrase boundariesrepresenting structurally significant locations in the music, suggesting thatthey had memorized the music in segments made up of phrase units.

Ginsborg and Sloboda: Singers’ recall of a new, unaccompanied song 423



We have argued that the presence of melody enhances recall for wordswhen there is a learned association between them. In a series of influentialstudies, Serafine and her colleagues investigated the nature of thisassociation. First, they asked non-musically trained listeners to judge thefamiliarity of songs heard once and then presented a second time, with one,both or neither component changed. Participants were more likely to judgewords as familiar if they were paired with their original melody, and viceversa. Serafine et al. (1984) refer to this as the ‘integration effect’, robusteven when melodies are paired with nonsense texts (Serafine et al., 1986).Crowder et al. (1990) argue that this is either the result of association ‘bycontiguity’ (if one component is altered it no longer provides the sameretrieval cues for the other component) or ‘physical interaction’ (when thewords of a song are altered, the melody undergoes subtle changes, forexample to the rhythm as well). ‘Integration’ has strong connotations thatsuggest that recall or recognition for two associated elements is all-or-nothing: either they are both remembered, or neither is. We believe that thereis rather little evidence to support such a strong concept. Rather, there is aprobabilistic relationship that increases the likelihood of one element beingretrieved if the other one is. So we will use the broader concept of ‘asso-ciation’ to frame the phenomena we are studying. Research investigating therelationship between the words and melodies of songs in memory, then,using both recall-for-words and recognition paradigms, suggests that theyare less likely to be stored and retrieved as separate components thantogether: if not as an integrated, single unit, then at least in association withone another. The present study aimed to determine whether this is the casefor expert singers who are required to perform songs from memory.

Little research has been published on singers’ actual memorizingstrategies (i.e. as observed in normal rehearsal situations). Ginsborg (2000),in one of a series of studies exploring this issue, interviewed experiencedprofessional singers, asking them to report on their strategies and theirreasons for choosing them. The majority of respondents reported studyingand memorizing the words and melodies separately in the initial stages ofpreparation for performance from memory. Their memorizing strategies wereparticularly for the words, since they were more anxious that they wouldforget the words than the melody; they also wanted to be confident that theyunderstood the words and would be able to convey their semantic meaning.However, in an observational study comparing the memorization strategies ofsingers of varying levels of expertise (Ginsborg, 2002), participants’ actualbehaviours did not fully reflect the reported strategies of the interviewees.Initial attempts on the melody separately from the words (playing,accompanying, vocalizing and counting beats aloud) gave way over thecourse of a series of practice sessions to attempts on the words and melodytogether. Although very few attempts were made on the words separatelyfrom the melody, expert singers were more likely than less expert singers to




recite the words of the song aloud. Yet memorization of the words andmelody separately was found to be associated with slower memorization ofthe song and less accurate performance of the song from memory. This wouldbe consistent with the hypothesis that recall for songs is more reliable whenwords and melody are encoded, from the start of deliberate memorization, inassociation with each other rather than as two independent components.

The present experiment, then, was designed to compare more and lessmusically expert singers’ recall for a newly memorized song. Participantswere asked to learn, memorize and perform the whole song, not just torecognize it or recall its words. They either learned the words and melodytogether, or – for a large proportion of the memorizing phase – separately. Theprincipal measures used were accuracy and fluency. We used fluency –performing without hesitations – as a measure because accuracy, on its own,is a rather crude measure for highly skilled time-sensitive performance.Fluency is a more fine-grained (and task-relevant) means of probing memoryprocesses, since even fully accurate performances can reveal uncertainties inmemory. As Sloboda and Lehmann (2001) demonstrated in their study ofexpert pianists’ performances of a Chopin prelude, listeners’ perception ofperformance is affected by the smallest differences in timing: the mainte-nance of ‘flow’ in music is crucial.

The following hypotheses were made: (1) recall for the song would be moreaccurate when (a) singers were more musically expert, and (b) when theyhad memorized the words and melody together; (2) recall for the song wouldbe more fluent when (a) singers were more musically expert, and (b) whenthey had memorized the words and melody together. In the light of evidencethat words and melodies of songs are stored and retrieved in association witheach other, (3) more ‘conjoint’ errors and omissions would be made,involving both the words and the melody, than ‘separate’ errors and omis-sions, involving one component only. Finally, in the light of evidence thatexpert musicians memorize and recall in segments made up of one or morephrases, it was hypothesized (4) that more hesitations would be made at theends of phrases than at the starts or mid-phrase.

Method

PARTICIPANTS

Sixty singers, 35 female and 25 male, took part in the study. Their agesranged from 18 to 61, with the majority below the age of 25. Most wereuniversity-based music students or sang in university choirs; six wereexperienced professional or semi-professional singers. Other than two singingteachers who were paid their usual fee for an hour’s consultation, allparticipants were unpaid.

Musical expertise was defined by the extent of musical training undergoneby participants and determined in a brief interview in the course of the




experimental procedure. Thirty-five participants had passed AssociatedBoard of the Royal Schools of Music (ABRSM) Grade 8 (or equivalent) insinging or on an instrument, a higher qualification such as a diploma, or hadgraduated from university with a degree in music and were therefore deemedto have a high level of musical expertise. Twenty-five had not taken anyexaminations in singing, music theory or on a musical instrument, or hadonly taken them up to and including Grade 7 (usually taken around twoyears before Grade 8); they were deemed to have – by comparison with the‘high-expertise’ group – ‘low’ musical expertise. All took part in a shortscreening task (singing along with a recording of the song, with the score infront of them) prior to taking part in the experiment proper. The 60participants from whom data were gathered were all able to sing the melodyaccurately at the second attempt.

MATERIALS

A novel, unaccompanied song was constructed by setting the second verse ofthe poem ‘The Moon and a Cloud’, by W.H. Davies (Davies, 1929), to themelody of an English folk song, ‘The Lowlands of Holland’ (Karpeles, 1974).This melody was chosen because it is typical of folk song melodies, not toodifficult for the singers with low levels of musical expertise, but not too easyor repetitive for the highly musically expert singers, for whom the song had topresent a meaningful challenge. Slight alterations were made to the text andmelody to remove repetitions of words and music.

The verse, headed ‘Cloud’, was printed as a poem on a single sheet ofpaper. The melody alone, with commas to indicate phrase endings, wasprinted in large type, two phrases per line, in two keys to suit the differentvoice types of the participants: F major, in the treble clef (for sopranos andtenors), and D major, in the treble (for altos) and bass clefs (for basses). Thewhole song, conventionally notated with the words of the poem below thenote-heads of the melody, was printed using the same format as the melodyalone, in both keys and clefs (see Appendix).

Cassette recordings of the melody and the whole song were provided forthe participants (stimulus tapes). The melody was played on a YamahaClavinova using the sonority of a conventional piano in the keys of F majorand D major; the whole song, words and melody, was sung by the researcher,a classically trained soprano, in both keys.

APPARATUS

A clock was provided for the duration of each experimental session, clearlyvisible to participants. One Sony TCM-919 cassette-recorder was used to playthe stimulus tapes and record the participants’ performances on test tapes;another was used to record the whole experimental procedure for eachparticipant on back-up tapes. A Sanyo TRC-8080 transcriber was used totranscribe the test tapes and the interview sections of three back-up tapes.




DESIGN

The study was of mixed design. There were two independent, between-subject,variables: expertise, with two levels, high and low – see the ‘Participants’section earlier – and condition, with three levels, to which participants wererandomly assigned. Condition 1 (WM) involved memorizing the words of thesong first, then the melody and finally the words and melody together.Condition 2 (MW) involved memorizing the melody of the song first, then thewords and finally the words and melody together. Condition 3 (B) involvedmemorizing the words and melody together throughout the whole memo-rizing phase. Table 1 shows how the participants were distributed betweenthe two levels of expertise and three experimental conditions.

The dependent, within-group variables were measures of accuracy andfluency, and errors (see the ‘Analysis’ section later).

PROCEDURE

The memorizing task took 20 minutes in all, for each participant. InConditions 1 (WM) and 2 (MW) it was subdivided into three parts:memorizing the words (five minutes), memorizing the melody (sevenminutes) and memorizing the words and melody together (eight minutes).Each performance from memory took one to two minutes. The twoperformances were separated by a 10-minute period, during which a briefinterview was carried out. This was used to establish participants’ levels ofmusical expertise, prevent the possibility of rehearsal and ensure that thesecond performance tested long-term memory. The 10-minute period wasalso long enough to provide an experimental analogue for the real-lifesituation in which singers often refresh their memory for songs shortly (butrarely immediately) before performing them on the concert platform.

Conditions 1 (WM) and 2 (MW)In Condition 1, the participant was given the poem ‘Cloud’ and asked tomemorize it over the course of the next five minutes. The participant wasthen left alone. After five minutes, the researcher returned, and the screeningprocedure took place. In three cases, volunteers recruited to take part inthe study were unable to sing the melody accurately at this stage, with the


TA B L E 1 Distribution of participants

Condition

Expertise 1 (WM) 2 (MW) 3 (B) TotalHigh 13 (8f *, 5m*) 13 (8f, 5m) 9 (4f, 5m) 35 (20f, 15m)Low 7 (4f, 3m) 7 (4f, 3m) 11 (7f, 4m) 25 (15f, 10m)

Total 20 20 20 60

* f = female, m = male



stimulus tape playing and the score in front of them. In these cases the tasks,tests and interviews proceeded as usual, so that the volunteers did notperceive themselves as having ‘failed’. However, the tests were neithertranscribed nor scored.

The participant was then told that s/he had seven minutes to memorizethe melody and invited to use the stimulus tape, piano or keyboard, asrequired. Again, the participant was then left alone. The procedure up to thispoint was the same in Condition 2 (MW), except that screening took place atthe start of the session, since the melody was presented first and the wordssecond.

On her return, the researcher removed the musical score of the melody(or, in Condition 2, the text of the poem) and replaced it with the musicalscore of the complete song, saying: ‘Here, at last, is the whole thing.’ She alsoreplaced the recording of the melody with a recording of the complete song,in case the participant wished to listen to it and explained that s/he wouldbe asked to sing the whole song from memory after eight minutes. Theparticipant was once again left alone.

Condition 3 (B)Participants were told that the purpose of the study was to investigatewhether performance from memory was more likely to be accurate if thewords and melody of the song had been memorized together, or separately.They would be required to memorize the words and melody together, andalthough it might be more usual for individual participants to memorizewords and melody separately, perhaps writing out the words as a poem orsinging the melody to vowel sounds or nonsense syllables, they were askednot to do so on this occasion. Indeed they were to sing the words of the songat all times, or to speak the words if they played the melody or any part of themelody on the piano or keyboard at any time. Before they started thememorizing task, however, they carried out the screening task as in the otherconditions, listening to the stimulus tape of the whole song before singingalong with it twice. From this point on, the preparation for memorizing thesong was the same as for the preparation for memorizing the melody. Theparticipant was told s/he had 20 minutes to memorize the song, and wouldbe expected to sing it from memory on the researcher’s return. Theparticipant was then left alone.

First performance from memoryFrom the point at which the researcher returned, the procedure was the samein all three conditions. The musical score was removed from the participant.The researcher asked the participant to sing the song from memory. Sheexplained that the participant should sing the whole song without stopping, ifpossible, but if s/he was unable to do so, then it was acceptable to ‘doanything to get started again – including going back to the beginning of the




song and starting again, though only once’. The researcher then gave thesame cue to each participant (the first two notes and words of the song in theappropriate key) and the participant attempted to sing the whole song frommemory. When s/he had finished, the researcher invited the participant tocontinue with the next stage of the procedure: the interview.

Interview: second performance from memoryThe interview lasted 10 minutes, and had two purposes: to determine theparticipant’s level of musical education and early memories of listening tomusic, and to provide a short delay in which the participant could notmentally rehearse the song. At the end of the interview, the participant wasasked to sing the song again from memory. After the second performance,participants were thanked, assured that they would not be required to singthe song again and debriefed.

TranscriptionWhen all the experimental sessions were completed, the cassette tapes onwhich the performances from memory had been recorded were re-numberedby a colleague so that they could be transcribed ‘blind’ by the researcher. Akey was made so that the transcripts of the tapes could be matched up againwith the corresponding back-up tapes and interview notes once thetranscriptions were complete. The researcher listened to the recordings of thetwo performances from memory given by each participant, and annotated aphotocopy of the musical score of the song so as to make a visual record ofwhat was actually sung by the participant. The musical score was amendedto show the locations of hesitations and where the words and/or melody wererepeated, omitted, substituted or performed in error. If there were very manycorrections, repetitions or performances of words and/or music apparentlyunrelated to the score as printed, they were notated by hand on separatesheets of manuscript paper.

ScoringThe numbers of word substitutions and omissions, melody substitutions andomissions, and hesitations made by each participant were then counted andentered on a scoring sheet.

ReliabilitySix of the 60 test tapes, selected at random, were transcribed and scored byan independent scorer following the researcher’s instructions. Each tape heldtwo performances from memory given by the same participant, immediatelyafter memorizing, and following the 10-minute delay. Both performanceswere transcribed. Each transcription yielded five scores: word errors, wordomissions, melody errors, melody omissions and hesitations at each time oftesting, making a total of 10 scores in all for each participant. Each scorer




therefore produced 60 scores. There was virtually no disagreement betweenthe researcher and independent scorer: the Pearson correlation obtainedbetween the 60 scores calculated by each of the scorers was statisticallysignificant (r [59] = 0.99, p < 0.0001).

For the purposes of this paper, only second performances were analysed.This is because we wished to avoid ceiling effects: first-time recalls could havebeen ‘contaminated’ by just-memorized information still retained in short-term memory.

Results

HYPOTHESIS 1It was hypothesized that singers with high levels of musical expertise wouldhave more accurate recall for the song (a) than those who were less musicallyexpert, and (b) when the song had been memorized in Condition 3 (B) ratherthan in Conditions 1 (WM) and 2 (MW). In order to assess recall for both thewords and the melody, the unit of the song chosen was the half-bar, i.e. twocrotchet beats. The more accurate the performance was, the more half-barswere sung accurately out of a possible total of 32.

Both hypotheses were upheld: participants with high musical expertisegave significantly more accurate performances from memory than those withlow musical expertise (F [1,54] = 10.64, p = 0.002), as did participants inCondition 3 (B) (F [2,54] = 4.58, p = 0.015). There was also a significantinteraction between expertise and condition (F [2,54] = 3.95, p = 0.025),illustrated in Figure 1 (see Table 2 for means and SDs). The differencesbetween the mean numbers of half-bars sung accurately by highly musicallyexpert participants in Condition 3 (B) were significantly different from thosesung by high- and low-expertise participants in Condition 1 (WM) (t [20] =2.76, p = 0.012 and t [14] = 4.79, p < 0.0001), high- and low-expertise


0

5

10

15

20

25

30

Condition 1 (WM) Condition 2 (MW) Condition 3 (B)Mean

nu

mb

er

of

half

-bars

su

ng

accu

rate

ly

(po

ssib

le t

ota

l =

32)

High expertise

Low expertise

F I G U R E 1 Accuracy by expertise and condition.



Ginsborg and Sloboda: Singers’ recall of a new, unaccompanied song 431T

AB

LE

2A

ccur

acy

and

fluen

cy

Nu

mbe

r of

Mea

n n

um

ber

ofh

alf-

bars

su

ng

Sepa

rate

err

ors

Con

join

t err

ors

Hes

itat

ion

s pe

r Ex

pert

ise

and

con

diti

onpa

rtic

ipan

tsac

cura

tely

(pos

sibl

e to

tal =

32

)pe

r h

alf-

bar

(%)

per

hal

f-ba

r (%

)be

ats

sun

g (%

)

Hig

h e

xper

tise

Con

diti

on 1

(WM

)1

31

3.7

7(1

0.4

7)

30

.76

(17

.74

)1

8.6

2(2

5.3

3)

4.9

2(3

.31

)Lo

w e

xper

tise

Con

diti

on 1

(WM

)7

11

.14

(5.1

8)

40

.1(1

6.5

4)

18

.55

(11

.62

)9

.79

(6.9

8)

Hig

h e

xper

tise

Con

diti

on 2

(MW

)1

31

1.7

7(7

.44

)2

8.4

7(1

8.0

4)

29

.23

(18

.84

)4

.46

(4.7

8)

Low

exp

erti

se C

ondi

tion

2 (M

W)

79

.13

(5.4

5)

38

.01

(20

.53

)1

0.6

(14

.7)

8.7

(6.6

7)

Hig

h e

xper

tise

Con

diti

on 3

(B)

92

4.8

9(7

.22

)1

4.5

9(1

2.5

)7

.64

(16

.17

)1

.58

(2.5

9)

Low

exp

erti

se C

ondi

tion

3 (B

)1

11

0.3

6(8

.81

) 4

7.2

1(1

7.3

6)

13

.05

(14

.9)

5.2

5(3

.98

)

All

Hig

h3

51

5.8

9(1

0.0

1)

25

.75

(17

.56

)1

9.7

4(2

2.0

7)

3.9

(3.9

3)

All

Low

25

10

.24

(5.4

6)

42

.64

(17

.81

)1

3.9

1(1

3.7

9)

7.4

9(5

.85

)

All

Con

diti

on 1

20

12

.8(8

.9)

34

.03

(17

.5)

18

.6(2

1.1

6)

6.6

3(5

.71

)A

ll C

ondi

tion

22

01

0.8

5(6

.78

)3

1.8

1(1

8.9

9)

22

.71

(19

.38

)5

.97

(5.7

1)

All

Con

diti

on 3

20

16

.9(9

.8)

32

.79

(22

.4)

10

.62

(15

.31

)3

.6(3

.84

)

All

part

icip

ants

60

13

.53

(8.8

1)

32

.79

(19

.42

)1

7.3

1(1

9.1

4)

5.4

(5.1

)



participants in Condition 2 (MW) (t [20] = 4.12, p = 0.001 and t [14] = 4.25,p < 0.0001) and low-expertise participants in Condition 3 (B) (t [18] = 4.92,p < 0.0001). All other differences were non-significant. That is to say,participants’ performance was equivalent irrespective of condition unlessthey were high-expertise participants who memorized words and melodytogether: these singers had significantly more accurate recall.

HYPOTHESIS 2It was hypothesized that singers with high levels of musical expertise wouldhave more fluent recall for the song (a) than those who were less musicallyexpert, and (b) when the song had been memorized in Condition 3 (B).Fluency was measured by counting hesitations between beats, out of a totalof 64. Two types of hesitation were noted. The first type was when singerspaused and started again, often having made comments such as ‘I’veforgotten’, or ‘Hang on a moment’. The second type was when singersstopped to correct themselves. Significantly more hesitations of the first type(M = 3.57, SD = 3.14) were made than the second (M = 2.11, SD = 2.1,t [59] = 3.44, p = 0.001). Hesitations of the second type were not analysedfurther. Preliminary investigations revealed that 24 participants stoppedsinging altogether before the end of the song, so hesitations were calculatedas a percentage of beats sung.

Both hypotheses were upheld: participants with high musical expertisegave significantly more fluent performances from memory than those withlow musical expertise (F [1,54] = 11.63, p = 0.001), as did participants inCondition 3 (B) (F [2,54] = 3.83, p = 0.028). There was no significantinteraction between expertise and condition (F [2, 54] = 0.076, NS). Meansand standard deviations are shown in Table 2.


0

2

4

6

8

10

12

Condition 1 (WM) Condition 2 (MW) Condition 3 (B)

He

sit

ati

on

s p

er

be

ats

su

ng

High expertise

Low expertise

F I G U R E 2 Fluency by expertise and condition.



HYPOTHESIS 3In the course of determining the number of half-bars sung accurately inorder to test Hypothesis 1, the errors initially noted on the transcriptionswere recoded as follows: (1) separate errors: substitutions or omissions of thewords while preserving the melody; substitutions or omissions of the melodywhile preserving the words; (2) conjoint errors: substitutions of bothcomponents, substitutions of one and omissions of the other, and omissionsof both, prior to resumption of the performance. These were calculated as apercentage of half-bars sung. Although it had been hypothesized that moreconjoint than separate errors would be made, reflecting conjoint processingof the words and melody, this was not the case when errors were calculatedas a proportion of half-bars actually sung.

A 2 × 2 × 3 mixed ANOVA with error type (separate vs. conjoint) as thewithin-subjects factor and expertise and condition as between-subjectsfactors revealed a main effect of error type (F [2,54] = 22.13, p < 0.0001).Nearly twice as many separate errors were made (M = 32.70, SD = 19.4) asconjoint errors (M = 17.31, SD = 19.14), indicating that participants wereable to preserve one element when they made errors in the other component.Yet there was a significant interaction between error type and expertise (F [2,54] = 9.03, p = 0.004) illustrated in Figure 2 (see Table 2 for means and SDs).The difference between the proportions of separate and conjoint errors madeby highly musically expert participants was much less pronounced than thatbetween the proportions of separate and conjoint errors made by lessmusically expert participants.

HYPOTHESIS 4Because singers are more likely to recall songs in phrase units, it washypothesized that more pauses would be made at the ends of phrases than atthe starts or mid-phrase (irrespective of expertise or condition). This was the


F I G U R E 3 Error type by expertise.

0

5

10

15

20

25

30

35

40

45

Separate errors Conjoint errors

Err

ors

per

half

-bar

su

ng

(%

)

High expertise

Low expertise



case. Hesitations after beats 1 or 2 (‘I see’), 9 or 10 (‘the moon’), 17 or 18(‘she turns’), 25 or 26 (‘a love-’), 31 or 32 (‘it shines’), 39 or 40 (‘that I’) or47 and 48 (‘she’s turned’) were deemed hesitations at starts of phrases.Hesitations after beats 7 or 8 (‘-by cloud’), 15 or 16 (‘its back’), 23 or 24 (‘itbright’), 31 or 32 (‘-ly black’), 39 or 40 (‘-dark cloud’), 47 or 48 (‘-riouslight’), 55 or 56 (‘my heart’) and 31 or 32 (‘of white’) were deemedhesitations at ends of phrases. Hesitations after any other beat were deemedmid-phrase hesitations. Again, the numbers of hesitations made in each ofthe three categories were calculated as percentages of the beats actuallysung. Because there were twice as many opportunities for mid-phrasehesitations as start- and end-phrase hesitations, the number of mid-phrasehesitations made by each participant was halved prior to analysis. As shownin Figure 4, there were significantly more end-phrase (M = 2.55, SD = 2.65)than start-phrase (M = 0.82, SD = 1.69; t = 4.87, d.f. = 59, p < 0.0001) ormid-phrase hesitations (M = 0.99, SD= 1.2; t = 4.92, d.f. = 59, p < 0.0001).The difference between the number of pauses made at starts and mid-phrasewas not significant.

Discussion

HYPOTHESES 1 AND 2: ACCURACY AND FLUENCY OF RECALL

As predicted, singers with high levels of musical expertise who memorizedthe words and melody of the song together for the whole of their practicetime had more accurate, and more fluent recall for the song than those whospent the bulk of their practice time memorizing the words and melodyseparately, and those who were less musically expert.

Kilgour et al. (2000) found that listeners with musical training had betterrecall for words set to music, probably because – as Wallace (1994) suggestedin her study of recall for the words of ballads – the melody provides a


0

0.5

1

1.5

2

2.5

3

Start-phrase hesitations Mid-phrase hesitations (/2) End-phrase hesitations

Hesit

ati

on

s p

er

beats

su

ng

F I G U R E 4 Hesitations by type.



framework for retrieving the words that is more meaningful and thereforemore memorable for those with musical expertise. In the present study, wefound that singers with high levels of musical expertise, given the oppor-tunity to sing rather than to recite, as in Kilgour et al.’s (2000) study, hadsuperior recall not only for the words, but also for the melody of the song.This demonstrates superior strategies for memorizing and recall, not justreading music, since all the participants had passed the initial screening test:singing the song (or at least the melody) accurately at sight.

The strategy of memorizing words and melody together was predicted tobe more effective than memorizing them separately, regardless of level ofexpertise. First, it reflects the desired outcome: the ability to recall and singboth components of the song. Second, it utilizes the association between thecomponents shown not only to enhance recognition when they are paired(Serafine et al., 1984, 1986; Crowder et al., 1990), but also to enhance recallfor words set to music (Rubin, 1977; Chazin and Neuschatz, 1990), partic-ularly when participants have the opportunity to hear the same melodyseveral times (Wallace, 1994; McElhinney and Annett, 1996) or, better still,rehearse them (Calvert and Tart, 1993; Kilgour et al., 2000; Rainey andLarsen, 2002). Rehearsal involves motor or kinaesthetic memory: it could beargued that preparation to sing involves an even more advanced form ofmotor memory than preparation for recitation of the words only.

What was the nature of the advantage to singers with high levels ofmusical expertise of using the strategy of memorizing words and melodytogether? It may simply be that they had learned, or had been taught tomemorize the words and melodies of songs as composites, in such a way thatmemory for one component enhances memory for the other, so that whenthey used this strategy under experimental conditions they performed betterfrom memory than when deprived of the opportunity to use a well-learnedstrategy. Nevertheless, some experienced professional singers do memorizewords and melodies separately, according to those who were interviewed byGinsborg (2000) and some of those who took part in her observational study(Ginsborg, 2002). This enables them to focus on the semantic meaning of thewords, which may reinforce recall for, or help to resolve confusions in, thewords once they have been memorized with the melody. However, separatememorization took longer and was associated with less accurate performancefrom memory, thus supporting the findings of the present study.

HYPOTHESIS 3: THE EXTENT TO WHICH WORDS AND MELODY ARE

RECALLED IN ASSOCIATION WITH EACH OTHER OR SEPARATELY

Given the evidence outlined above that the presence of melody enhancesrecall for the words of songs, and indeed the evidence of the present studyindicating superior recall for both components when rehearsed together (atleast for highly trained musicians), we predicted that words and melodywould be recalled together and that this would be reflected by a greater




proportion of conjoint than separate errors. Because so many participants –two-fifths – failed to sing the whole song from memory, errors were calculatedas a proportion of half-bars actually sung, and our prediction was notupheld. Taken together, more than twice as many separate errors were madeas conjoint errors. Yet high-expertise participants were only marginally morelikely to make separate errors than conjoint errors. Low-expertise partici-pants made nearly twice as many separate errors as the high-expertiseparticipants, and fewer conjoint errors.

For example, errors made by highly musically expert participants weremost likely to involve the substitution of musically acceptable pitches andrhythms. In the second half-bar of Phrase 8, for example, a rising sequence ofpitches including a minor third interval was often sung incorrectly as thestart of a major scale. Meanwhile, less musically expert participants foundthe rhythms in Phrases 4 and 6 problematic; the frequent occurrence oferrors at the beginning of Phrase 5 seem to be attributable to the setting ofthe word ‘lovely’ for a second time, but to a different melody.

Two locations where there were high proportions of errors were in Phrase3 and the beginning of Phrase 4. Here separate errors involved word substi-tutions, perhaps because participants had failed to consider the semanticmeaning of the words while they were memorizing the song. For example, thewords ‘make’ and ‘turns’ were often substituted in Phrases 3 and 4 (‘I look tosee her make it bright / She turns it to a lovely black’) and the word ‘light’, atthe end of Phrase 7, was often substituted or omitted following ‘glorious’despite the participant’s having just sung the verb ‘shines’.

To reiterate: when singers make separate errors they preserve onecomponent while forgetting or recalling the other erroneously; when theymake conjoint errors we must assume that erroneous recall for onecomponent has affected recall for the other. If Crowder et al.’s (1990)‘physical interaction’ hypothesis were to be supported by our findings, allerrors would have had to be conjoint. This was not so: words and melody arenot integrated to such an extent that if one is forgotten the other is too. Theoccurrence of some conjoint errors, however, supports their weaker‘association-by-contiguity’ theory. This theory is extended by Stevens et al.(1998), who propose that words and melody may be representedindependently and simultaneously associated by contiguity (‘composite’representation). Thus it is not just the words and melody of the song that arememorized, but also the associations between them, in such a way thatsingers are more likely to be able to retrieve one component accurately whenrecall for the other fails, as was the more frequent case in the present study.As Rubin (1977) and Wallace (1994) suggest, music can structure recall forwords so that when words are forgotten memory for the music provides cuesenabling the words to be ‘picked up’ again. Another possibility is thatmemory for the words structures recall for the melody, as shown by Gilboa(2002). However, as Calvert and Billingsley (1998) showed, the melody has




to be sufficiently simple for it to be useful as a framework for recalling words,and/or – as we have seen – the memorizer must be sufficiently musicallyexpert to learn and recall it adequately, if not perfectly.

HYPOTHESIS 4: HESITATIONS AT PHRASE BOUNDARIES

Participants were more likely to hesitate at the ends of phrases than at thestart or mid-phrase. This supports the existing evidence that musicians’ recallis in the form of segments beginning and ending at phrase boundaries(Ginsborg, 1999; Williamon and Valentine, 2002).

CONCLUSION

The relationship between the words and melodies of songs in memory, forsingers who have deliberately memorized songs, is such that they are neitherrecalled entirely separately nor integrated to such an extent that if onecomponent is recalled erroneously, recall for the other is inevitably affected.Rather, they are stored and retrieved in association with one another. Thus,the best way for an expert singer to learn and memorize a song, assumingthat both the words and melody are novel, is to memorize the words andmelody together.

A C K N O W L E D G E M E N T S

This study was carried out in partial fulfilment of the PhD degree awarded to the firstauthor, and funded by a Research Studentship, R00429534122, from the Economicand Social Research Council. We are grateful to Andreas Lehmann and ananonymous reviewer for helpful comments on an earlier draft.

R E F E R E N C E S

Calvert, S.L. and Billingsley, R.L. (1998) ‘Young Children’s Recitation andComprehension of Information Presented by Songs’, Journal of AppliedDevelopmental Psychology 19: 97–108.

Calvert, S.L. and Tart, M. (1993) ‘Sung versus Verbal Forms for Very-long-term, Long-term, and Short-term Verbatim Recall’, Journal of Applied Developmental Psychology14: 245–60.

Chazin, S. and Neuschatz, J.S. (1990) ‘Using a Mnemonic to Aid in the Recall ofUnfamiliar Information’, Perceptual and Motor Skills 71: 1067–71.

Davies, W.H. (1929) ‘The Moon and a Cloud’, in The Complete Poems of W.H. Davies,p. 248. New York: J. Cape & H. Smith.

Crowder, R.G., Serafine, M.L. and Repp, B.H. (1990) ‘Physical Interaction andAssociation by Contiguity in Memory for the Words and Melodies of Songs’,Memory and Cognition 18(5): 469–76.

Ericsson, K.A. (1996) The Road to Excellence. Hillsdale, NJ: Erlbaum.Gilboa, A. (2002) ‘Do Words Affect Memory for Melodies?’, unpublished PhD thesis,

Bar-Ilan University, Ramat-Gan, Israel.Ginsborg, J. (1999) ‘Expert Singers’ Memorization and Recall of Songs: The

Interaction of Words and Music in Memory’, unpublished PhD thesis, KeeleUniversity, UK.




Ginsborg, J. (2000) ‘Off by Heart: Expert Singers’ Memorization Strategies and Recallfor the Words and Music of Songs’, paper presented at the Sixth InternationalConference on Music Perception and Cognition and Fourth Triennial Conferenceof the European Society for the Cognitive Sciences of Music, August, KeeleUniversity, UK.

Ginsborg, J. (2002) ‘Classical Singers Memorizing a New Song: An ObservationalStudy’, Psychology of Music 30(1): 56–99.

Gruson, L.M. (1988) ‘Rehearsal Skill and Musical Competence: Does Practice MakePerfect?’, in J.A. Sloboda (ed.) Generative Processes in Music: The Psychology ofPerformance, Improvisation and Composition, pp. 91–112. London: Oxford UniversityPress.

Karpeles, M. (ed.) (1974) ‘The Lowlands of Holland’, in Cecil Sharp’s Collection ofEnglish Folk Songs, Vol. 1, pp. 22–3. London: Oxford University Press.

Kilgour, A.R., Jakobson, L.S., and Cuddy, L.L. (2000) ‘Music Training and Rate ofPresentation as Mediators of Text and Song Recall’, Memory & Cognition 28(5):700–10.

McElhinney, M. and Annett, J.M. (1996) ‘Pattern of Efficacy of a Musical Mnemonicon Recall of Familiar Words over Several Presentations’, Perceptual and Motor Skills82: 395–400.

Rainey, D.W. and Larsen, J. (2002) ‘The Effect of Familiar Melodies on Initial Learningand Long-term Memory for Unconnected Text’, Music Perception 20(2): 173–86.

Rubin, D.C. (1977) ‘Very Long-term Memory for Prose and Verse’, Journal of VerbalLearning and Verbal Behavior 16: 611–21.

Serafine, M.L., Crowder, R.G. and Repp, B.H. (1984) ‘Integration of Melody and Textin Memory for Songs’, Cognition 16: 285–303.

Serafine, M.L., Davidson J., Crowder, R.G. and Repp, B.H. (1986) ‘On the Nature ofMelody–Text Integration in Memory for Songs’, Journal of Memory and Language25: 123–35.

Sloboda, J.A. and Lehmann, A.C. (2001) ‘Tracking Performance Correlates ofChanges in Perceived Intensity of Emotion during Different Interpretations of aChopin Piano Prelude’, Music Perception 19: 87–120.

Sloboda, J.A. and Parker, D.H.H. (1985) ‘Immediate Recall of Melodies’, in P. Howell,I. Cross and R. West (eds) Musical Structure and Cognition, pp. 143–67. London:Academic Press.

Stevens, K.J., McAuley, J.D. and Humphreys, M.S. (1998) ‘Relational Information inMemory for Music: The Interaction of Melody, Rhythm, Text and Instrument’,Noetica: Open Forum 3(8), URL (accessed April 2006):http://www2.psy.uq.edu.au/CogPsych/Noetica/OpenForum.html

Wallace, W.T. (1994) ‘Memory for Music: Effect of Melody on Recall of Text’, Journal ofExperimental Psychology: Learning, Memory and Cognition 20 (6): 1471–85.

Williamon, A. and Valentine, E. (2002) ‘The Role of Retrieval Structures inMemorizing Music’, Cognitive Psychology 44: 1–32.





Appendix: Music and words of ‘Cloud’



JA N E G I N S B O RG did her PhD, on memory for words and music, with John Sloboda atKeele University. She has been a lecturer in psychology at the University ofManchester, a research associate at the Unversity of Sheffield, senior lecturer inpsychology at Leeds Metropolitan University and is now a Research Fellow in theResearch Centre for the Vocational Training of Musicians, part of the Centre forExcellence in Teaching and Learning at the Royal Northern College of Music,Manchester, UK.Address: Royal Northern College of Music, 124 Oxford Road, Manchester, M13 9RD,UK. [email: [email protected]]

J O H N A . S L O B O DA has been a member of the School of Psychology at KeeleUniversity since 1974 and Director of the Unit for the Study of Musical Skill andDevelopment since 1991. He is a Fellow of the British Psychological Society, and hasbeen President of both the Psychology and General Sections of the British Associationfor the Advancement of Science, as well as President of the European Society for theCognitive Sciences of Music, where he serves on the editorial board of its journalMusicae Scientiae. He was recipient of the 1998 British Psychological Society’sPresidents’ Award for Distinguished Contributions to Psychological Knowledge, andin 2004 was elected to Fellowship of the British Academy.Address: School of Psychology, Dorothy Hodgkin Building, Keele University, Keele,Staffordshire, ST5 2BG, UK. [email: [email protected]]




Ginsborg & sloboda

Documents