Memory & Cognition

2007, 35 (2), 242-253

Learning lyrics: To sing or not to sing?

Amélie Racette and Isabelle Peretz
Université de Montréal, Montréal, Québec, Canada

According to common practice and oral tradition, learning verbal materials through song should facilitate
word recall. In the present study, we provide evidence against this belief. In Experiment 1, 36 university students,
half of them musicians, learned an unfamiliar song in three conditions. In the sung–sung condition, the song to
be learned was sung, and the response was sung too. In the sung–spoken condition, the response was spoken. In
the divided–spoken condition, the presented lyrics (accompanied by music) and the response were both spoken.
Superior word recall in the sung–sung condition was predicted. However, fewer words were recalled when sing-
ing than when speaking. Furthermore, the mode of presentation, whether sung or spoken, had no influence on
lyric recall, in either short- or long-term recall. In Experiment 2, singing was assessed with and without words.
Altogether, the results indicate that the text and the melody of a song have separate representations in memory,
making singing a dual task to perform, at least in the first steps of learning. Interestingly, musical training had
little impact on performance, suggesting that vocal learning is a basic and widespread skill.

The notion that music may serve as a mnemonic tech- Thus, to properly test the idea that music may serve as a
nique for learning verbal material has a long history. Min- mnemotechnique for recalling words, one must not only
strels transmitted stories through songs (Calvert & Tart, examine oral responses, but also select material in which
1993; Rubin, 1995), and this practice is still influential the words are appropriately set to the music (Gingold &
today. Among the most familiar experiences of musical Abravanel, 1987); in short, one must use real songs. This
learning are jingles for brand names and the alphabet was done in the present study.
song children learn. Other examples that have been de- Although an adequate test of the idea that music fa-
scribed consist of learning the laws of physics through cilitates text recall requires consideration of both input
karaoke (Dickson & Grant, 2003) and learning English as and output factors, the influence of music on word recall
a second language via songs (Medina, 1993). The goal of starts at the encoding stage. Thus, all prior studies that
the present study was to contribute to the understanding used written recall but looked at input factors may shed
of this phenomenon from both empirical and theoretical light on the idea that sung words are easier to encode than
perspectives. spoken words. Support for this notion is mixed. In sev-
Indeed, it is not obvious why music should facilitate eral studies, participants recalled as many sung as spoken
word recall, since there is more to learn in a song than in a words (Gingold & Abravanel, 1987; Wolfe & Hom, 1993)
text. To our surprise, this simple notion has not been prop- or even did worse on sung material (Calvert & Billingsley,
erly assessed. Song learning is typically assessed through 1998). Yet, in many other studies, an advantage of sung
written recall (Kilgour, Jakobson, & Cuddy, 2000; McEl- over spoken presentation has been shown (Calvert & Tart,
hinney & Annett, 1996; Wallace, 1994). This change in 1993; Chazin & Neuschatz, 1990; Kilgour et al., 2000;
format between perception and performance introduces a McElhinney & Annett, 1996; Rainey & Larsen, 2002;
bias in word recall in favor of the spoken version, because Wallace, 1994; Wolfe & Hom, 1993).
extracting words from the sung version requires filter- This advantage of sung over spoken text at encoding
ing out the music component. Moreover, written recall has been attributed to speed (Kilgour et al., 2000) and to
requires participants to perform a task that is not familiar melody simplicity (Wallace, 1994). In effect, words are
to them. Lyrics are typically learned to be sung, not to be pronounced more slowly when singing than when speak-
written. Thus, a putative advantage of singing over recit- ing. When the sung version of a text is compressed to
ing words should be assessed with a vocal response. To our match its spoken duration, there is no longer a difference
knowledge, this procedure has been used only once (Jelli- in recall, suggesting that the slower rate of singing in com-
son & Miller, 1982), and the results were negative: Music parison with speaking is a key variable in song learnability
was found to interfere with digit recall and had no effect (Kilgour et al., 2000). Similarly, in order for a sung text
on word recall. However, in this experiment the words to be recalled better than one that is recited, it has to be
were unrelated and probably were not optimally aligned presented on a simple and repeated melody, as typically
to the music, hence introducing an additional difficulty. found in songs. Lyrics that are sung to a complex and

I. Peretz, [email protected]

Copyright 2007 Psychonomic Society, Inc. 242

 Learning Song Lyrics     243

changing melody can be more difficult to remember than & Arguin, 2004; Samson & Zatorre, 1991; Serafine,
their spoken version (Wallace, 1994). Crowder, & Repp, 1984; Serafine, Davidson, Crowder,
Songs also possess structural characteristics that may & Repp, 1986). In the recognition of song lines, melody
assist text recall. For instance, the metrical structure of and text appear to be highly associated, even after a single
music and the number of musical notes in a line can cue hearing, suggesting that lyrics and melody representations
word recall. Similarly, song lyrics are usually constrained are integrated in memory for songs (Serafine et al., 1984;
by both semantics (a story underlies the words, generally Serafine et al., 1986). However, there is increasing evi-
through a schema or a script) and sound patterns (e.g., dence that the music and language components of songs
rhymes, alliteration), which may again limit the possibili- maintain autonomy in both perception (Besson, Faïta,
ties. Indeed, when errors occur in song recall, the changes Peretz, Bonnel, & Requin, 1998; Bonnel, Faïta, Peretz, &
usually preserve the rhyme (Rubin, 1995) and the number Besson, 2001) and memory (Crowder et al., 1990, Experi-
of syllables in the line (Wallace, 1994). ment 3; Peretz, 1996). Very recently, we extended these
Nevertheless, as mentioned previously, texts of real conclusions to singing by studying brain-damaged pa-
songs are not systematically memorized better when sung tients who suffered from a severe speech disorder without
than when recited (see, e.g., Gingold & Abravanel, 1987, a a concomitant musical disorder (Hébert, Racette, Gagnon,
study with children; Wallace, 1994, Experiment 3; Wolfe & Peretz, 2003; Peretz, Gagnon, Hébert, & Macoir, 2004;
& Hom, 1993). This lack of consistency might be related Racette, Bard, & Peretz, 2006). The results indicate that
to the mode of response, as has been pointed out previ- verbal production, be it sung or spoken, is mediated by
ously. Writing down or reciting the words requires filtering the same (impaired) language output system and that this
them out from the music. This filtering process might be speech route is distinct from the (spared) melodic route.
difficult, especially when words are sung at high pitches These neuropsychological findings strongly suggest that
(Scotto Di Carlo & Germain, 1985). One way to control singing taps into distinct codes for melody and text. Thus,
for this perceptual disparity between sung and spoken pre- the present study should help us to shed further light on
sentations is to present the spoken lyrics accompanied by this issue by testing singing in the normal population.
music. We refer to this situation as the “divided song” con- The general population is musically untrained. How-
dition. By adding the musical background, this condition ever, we also considered a group of professional musi-
also maintains the presence of the melody at encoding. cians because these individuals might exploit musical cues
None of the prior studies that aimed at testing the effect of more effectively than nonmusicians, and therefore might
music on word recall have included such a control condi- benefit more from the presence of music on text recall.
tion. Finally, in order to promote the use of musical cues Moreover, musicians seem to have better verbal memory
as a structural aid in the retrieval process, one needs to than nonmusicians (Chan, Ho, & Cheung, 1998; Jellison &
assess sung recall. Miller, 1982; Kilgour et al., 2000), apparently from child-
Consideration of all these factors is not solely motivated hood (Ho, Cheung, & Chan, 2003). Thus, it is possible that
by experimental elegance, since the contribution of music musical training strengthens auditory temporal processing,
to verbal memory is a theoretically important question. which would mediate verbal recall (Jakobson, Cuddy, &
As alluded to previously, text and melody are aligned in Kilgour, 2003; Jellison & Miller, 1982). These results in
songs in such a way that they promote binding of speech turn suggest that music may assist in text recall, but only in
and musical sounds at multiple levels of processing. These those individuals who regularly use the two codes.
tight relations may enhance memory for relatively distinct Therefore, in the present study 36 students, half with
representations of text and melody in songs by linking musical expertise, had to learn novel songs in three dif-
elements of words and tones in rich, multiple-linked rep- ferent conditions. As illustrated in Table 1, the text to be
resentations (Peretz, Radeau, & Arguin, 2004). Alterna- learned was either sung or spoken. When spoken, its cor-
tively, the text and melody of songs might be integrated responding melody was sung on /la/ in the background.
in a unitary representation, especially when singing is Recall of text was either sung (on the melody) or spoken
required. Central to the distinction between these two (lyrics alone). We predicted that word recall would be su-
positions is a difference in the way recall is assumed to perior in the sung–sung condition, especially for musi-
operate. If integrated, a part of the song’s representation cians, simply because singing is slowed down relative to
will reinstate the whole—namely, singing the melody will normal speech. The sung–spoken condition was expected
reinstate the text. If separate, a part (the melody) may or to be the most difficult, because in this condition the text
may not connect with the other part (the text), depending needs to be extracted from the song. In the divided–­spoken
on the strength of the links. Thus, the integrated view of condition, there would be no cost of extracting the words,
song memory would predict superior text recall in singing
over speaking, whereas a separate-memory view of song
components would not. Table 1
The idea that melody and text might be represented in Modes of Presentation and Recall
a unitary memory trace has been relatively neglected in in the Three Conditions of Experiment 1
performance, but it has been studied in perception and Presentation of the Song Recall of the Lyrics
memory. The prevailing paradigm in the field involves the Sung Sung
recognition of unrelated song lines (Crowder, Serafine, & Sung Spoken
Repp, 1990; Morrongiello & Roes, 1990; Peretz, Radeau, Spoken (divided) Spoken
244        Racette and Peretz

but neither would there be an advantage from hearing Song Analysis

them sung (at a slow speed). The six songs used in the present study had a theme, most often
referring to love. The words used were thus predictable (e.g., amour
EXPERIMENT 1 “love,” fleurs “flowers,” cœur “heart”), but they were different
enough across songs to prevent confusions. Themes diverging from
Method love were related to music (one song) and to patriotism (one song).
Participants Frequency of usage of the song words was high. The 155 different
Thirty-six French-speaking university students (mean age, 25; words used in the songs had a mean frequency of 2,057 per million,
range, 20–37) who felt comfortable singing were recruited. Half of based on a French lexical database (New, Pallier, Ferrand, & Matos,
the participants (14 women, 4 men) were considered nonmusicians, 2001): 72% were highly frequent, with a frequency of usage higher
with a mean of 2 years of music training before the end of high than 50 per million. Only 13% of the words had a low frequency,
school. The other half (14 women, 4 men) were students in the music corresponding to less than 15 per million. Rhymes were present in
faculty or professional musicians. This second group had on average all of the six songs except one, which only rhymed in the second
13.6 years of musical training: 5 were singers, 5 had singing as their half (see the Appendix). In all songs, there was a one-to-one map-
second instrument, and 8 had no formal training in singing besides ping between syllables and tones, with each syllable coupled to a
solfeggio. Participants were paid for their participation. single note. There were no ties (see the Appendix). However, notes
outnumbered words [t(47) 5 13.49, SE 5 0.17, p , .001], because
Materials even if most of the words (69%) were monosyllabic, 25% were di-
Unfamiliar songs were chosen from the repertoire of a popular syllabic, and 6% were trisyllabic.
French-Canadian folksinger, author, and composer (Claude Gau­- Regarding musical structure, the six songs had stable and standard
thier). Six songs with few word or melodic line repetitions were meters. Four of the six songs had a duple meter (4/4), and the other
selected. From these, eight-line excerpts were chosen for the learn- two, a triple meter (3/4 and 6/8). All of the songs were in major
ing task (see the Appendix for an example). Each line carried, on mode and were written in the key of A♭, C, D, G, or F, and two songs
average, 6 words and 8 notes. Thus, on average, a song contained contained a single modulation. Even if melodies were chosen for
49 words (range, 45–57) and 68 notes (range, 64–74). An additional their diversity, the melody parts were highly coherent within a song
eight-line excerpt from an unfamiliar choir song by Johann Steuer- (see the Appendix). Lines respected the grouping preference rules
lein served as a training song. proposed by Lerdahl and Jackendoff (1983), which are mainly based
The six song excerpts from Claude Gauthier were considered to on gestalt principles, such as symmetry, proximity, similarities, and
be “good” songs, as assessed by 7 pilot participants who were unfa- parallelism. Thus, the melodic lines did not contain the same ab-
miliar with the singer. The judges were presented with the six song solute pitches or the same exact rhythm, but they were structurally
excerpts in their original version, and these were randomly mixed similar in a given song.
with excerpts of six hit songs by the same folksinger. Each excerpt The alignment between the prosody of the text and the rhythm of
was presented twice in a random order. For each song excerpt, the the melody conformed to the rules of French songs (Dell, 1989). The
judges rated its musicality, simplicity, and potential to be a hit on last accentuated syllable of the verse coincided with a strong beat.
three 6-point scales, in which 1 meant poor and 6 excellent. Very This constraint was present in each of the eight lines of the songs.
similar ratings were obtained for the hits and the experimental songs Note that, in French, there is no prescriptive rule regarding the al-
on each dimension (3.7 and 3.8, respectively, for musicality, 3.4 and ternation between strongly and weakly stressed syllables. Hence,
3.5 for simplicity, and 3.2 and 3.4 for hit potential, with first and whether the prosodic accents in French songs coincide with musi-
second ratings pooled together), supporting the idea that the selected cal metrical accents, as found in English songs, remains debatable
material corresponded to well-formed songs. (Palmer & Kelly, 1992).
The six songs and the training song were produced a cappella
(without instrumental accompaniment) by a female singer, who Procedure
learned the songs beforehand. The same singer also sang each song Because pilot observations suggested high variability in the num-
on /la/ and pronounced the lyrics with a natural intonation. The best ber of lines participants were able to recall, an adaptive procedure
performance of each song in each version was recorded on a DAT was used. This procedure is represented in Table 2. Participants first
Sony via a Shure 565SD microphone, and then transferred into a heard the whole song excerpt once, in order to familiarize them-
computer and edited with the Cool Edit program (Syntrillium Soft- selves with it. Afterward, the first line was presented, and they had
ware, 1996). The three versions of the same song served to create to repeat it. Lines were then repeated two by two. Once the first
two types of stimuli, the sung songs and the “divided” songs. The four lines were repeated, participants were asked to recall the four
latter were created by coupling each spoken line with its correspond- lines from the beginning without hearing them again. If more than
ing melody sung on /la/. In these divided songs, the intensity of the 80% of the words were recalled, Lines 5 and 6 were presented and
melody had to be decreased by 32%, on average, in order to make repeated before participants were again asked to recall every line
the spoken version intelligible. The intelligibility of songs’ lyrics starting at the beginning. If more than 80% of the words were again
in the sung and the divided versions was verified by measuring the recalled, Lines 7 and 8 were presented, and participants made a last
number of errors made by the participants when repeating the lines recall of the entire excerpt. Number of words correctly produced
in a spoken mode immediately after hearing them. Mean correct was calculated online by the experimenter so as to decide whether
repetitions were 94.8% and 95.3% for the sung and divided presen- the procedure was to be continued or stopped.
tations, respectively [t(35) 5 0.40, SE 5 1.46, n.s.], suggesting that Presentation of the song lines was either sung or spoken with the
understanding the words from the sung and divided versions of the melody in the background. Repetition was either sung or spoken (see
songs was equally easy. Table 1). In the sung–sung condition, participants listened to the sung
As expected, the duration of the spoken version was about 43% version of the lyrics and sung them back. In the sung–­spoken condi-
shorter than the sung version [M 5 2.51 and 4.40 sec per line, re- tion, participants listened to the sung version of the lines and were
spectively; t(47) 5 9.60, SE 5 0.20, p , .01]. Because the divided asked to recall only the lyrics and recite them in a natural way. In the
condition combined both the spoken and the “sung on /la/” versions divided–spoken condition, they listened to the divided version of the
(M 5 4.42 sec per line), divided and sung presentations had equiva- lines and recalled only the lyrics in a similarly natural manner.
lent durations (M 5 4.46 and M 5 4.40 sec per line, respectively) Each participant learned one song in each condition, for a total of
[t(47) 5 1.58, SE 5 0.04, n.s.]. In the divided condition, the shorter three different songs. The order of the conditions was counterbal-
spoken line was placed in the middle of the sung melody, so that it anced across participants according to a Latin square. Care was taken
was preceded and followed by equivalent durations of the melody. to test each song in each condition across participants in each group.
 Learning Song Lyrics     245

Table 2
Illustration of the Adaptive Learning Procedure
Lyrics Presented Lyrics Repeated Lyrics to Be Recalled
1 Dans cette petite boîte vide 1 Dans cette petite boîte vide 1 Dans cette petite boîte vide
2 Avec un ruban de velours
1 Dans cette petite boîte vide 1 Dans cette petite boîte vide
3 Il y a tout mon cœur et mes rides
2 Avec un ruban de velours 2 Avec un ruban de velours
4 Mon sourire et tout mon amour
3 Il y a tout mon cœur et mes rides 3 Il y a tout mon cœur et mes rides
4 Mon sourire et tout mon amour 4 Mon sourire et tout mon amour
If less than 80% of words recalled, stop.
If 80% or more words recalled, continue.

5 Il n’y a pas d’argent qui remplace 5 Il n’y a pas d’argent qui remplace 1 Dans cette petite boîte vide
6 Tout le temps que l’on peut donner 6 Tout le temps que l’on peut donner 2 Avec un ruban de velours
3 Il y a tout mon cœur et mes rides
4 Mon sourire et tout mon amour
5 Il n’y a pas d’argent qui remplace
6 Tout le temps que l’on peut donner

If less than 80% of words recalled, stop.

If 80% or more words recalled, continue.

7 À tous ceux que l’on aime hélas 7 À tous ceux que l’on aime hélas 1 Dans cette petite boîte vide
8 Trop souvent qu’on oublie d’aimer 8 Trop souvent qu’on oublie d’aimer 2 Avec un ruban de velours
3 Il y a tout mon cœur et mes rides
4 Mon sourire et tout mon amour
5 Il n’y a pas d’argent qui remplace
6 Tout le temps que l’on peut donner
7 À tous ceux que l’on aime hélas
8 Trop souvent qu’on oublie d’aimer

The practice song was learned before each condition in the corre- lines that participants were recalling. This ratio was multiplied by
sponding version. Participants were asked to do their best to recall the 100 to obtain a percentage.
exact words and, if they did not remember a part, to report whatever The raw number of words recalled and the number of lines (4, 6,
came to mind. The participants listened to digital recordings through or 8) attempted in immediate recall were also taken into account. Num-
speakers, and their performance was recorded on a Sony DAT. ber of hesitations, defined as a marked pause or a corrected attempt
In order to assess verbal memory independently from song mem- (the participant tried something and then changed her/his answer), was
ory, the Rey Auditory Verbal Learning Test (RAVLT; Rey, 1964) also noted. Finally, the locations of breaths were recorded.
was administered after participants had learned the three songs. In In the sung mode of recall, the musical notes in each final recall
this task, they had five trials to recall a list of 15 unrelated words. were transcribed by two independent musicians. The agreement be-
The RAVLT also served as a distraction task. Afterward, participants tween the judges was very low for rhythm, and therefore rhythm was
were asked to make a written recall1 of the three songs they had not considered in the present study. Instead, pitch intervals and di-
previously learned (excluding the practice song). The time elapsed rections were analyzed. The number of correct notes was defined as
between recalls was approximately 20 min. This delayed recall came the number of notes both judges gave a point to. When there was a
as a surprise test, because participants were not warned in advance disagreement (in 15% of the cases—i.e., for 228 out of 1,559 notes
that their memory would be assessed one more time. Since music produced), the note was discarded. Thus, the score corresponded to
may help long-term memory (Rainey & Larsen, 2002), 25 of the 36 the number of correct pitches divided by the total number of possible
participants were contacted 7 months later (2–10 months after the notes minus the notes both raters disagreed upon, multiplied by 100.
first administration) and asked again for vocal song recall, which
was recorded on tape. Results
Data Scoring Performance in the immediate recall of the song was
For text recall, words were considered correct or incorrect, irre- first examined by considering the percentages of words
spective of their pitch and duration when sung. Words were chosen that were correctly sung and spoken after presentation of
over syllables as the criterion because number of syllables some- sung and recited songs. The number of lines completed,
times differed across conditions; mute vowels are often sung but the total number of words recalled, the position of the for-
not pronounced. The words had to be produced in the correct order
to obtain a point. Omissions and substitutions received no points. A gotten lines in the song, the types of errors made by the
point was lost when words were added, and half-points were sub- participants, and pitch accuracy were also analyzed. Word
tracted when words were mispronounced but recognizable. In cases recall was also examined as a function of learning condi-
in which participants made an error because they misperceived a tion both after a delay of 20 min and after several months.
word and did not repeat it correctly when first heard, the repeated Finally, performance in lyrics learning was compared with
version of the word was considered correct in recall. Finally, a point performance in the Rey Auditory Verbal Learning Test.
was lost if the correct word was spoken instead of sung, and vice
versa. The total numbers of words correctly reproduced in the last
immediate recall and in the delayed recall were then divided by the Immediate Recall
number of words contained in the lines to be recalled by a given Correct words. An initial repeated measures ANOVA,
participant, thus taking into consideration the differing number of with both group (musician, nonmusician) and order of
246     Racette and Peretz

Table 3
Means (and Standard Errors) Obtained in Each Condition
on Immediate Recall in Experiment 1
Sung–Sung Sung–Spoken Divided–Spoken
Group Dependent Variable M SE M SE M SE
Nonmusicians % 63.9 4.8 73.0 3.5 74.0 3.2
Words 20.4 2.9 22.7 2.3 27.9 2.5
Lines   5.0 0.4   5.0 0.4   6.0 0.4
Musicians % 56.1 4.3 68.9 3.0 73.9 3.6
Words 18.7 2.3 23.7 2.3 27.1 3.0
Lines   5.2 0.4   5.6 0.4   5.8 0.4
Mean % 60.0 3.2 70.9 2.3 74.0 2.4
Words 19.6 1.8 23.2 1.6 27.5 1.9
Lines   5.1 0.4   5.3 0.4   5.9 0.4

presentation (1, 2, 3) as between-subjects variables and the participants, and compared it with recall of the last
condition (sung–sung, sung–spoken, divided–spoken) as line. The last line was not recalled by 42% of the partici-
a within-subjects variable, was performed on the percent- pants while singing, but by only 18% while reciting. This
age of words recalled. Since there was neither an effect difference was significant [Q(2) 5 8.78, p , .05, using
of order [F(2,30) 5 1.19, MSe 5 175.84, p . .05] nor an Cochran’s test]. Thus, serial position of the line appears
interaction between order and the other factors, order was to be more important in singing than in speaking. More-
not considered in the following analyses. over, forgetting an entire line was more frequent in sung
In Table 3, performance is expressed in percentage of recall (24% of the lines) than in the spoken recalls (11%)
words recalled, as well as in terms of the total number of [F(2,70) 5 8.25, MSe 5 0.02, p , .01]. When a line was
words recalled and the number of lines attempted. As can omitted in singing, the next line was omitted in 71% of
be seen, recall appears more difficult when participants the cases. In contrast, when a line was omitted in recit-
have to sing, regardless of their musical background. This ing, only 55% of the following lines were missed. This
was supported by an ANOVA performed on percentage of suggests that text recall in singing is more strictly sequen-
words recalled, with condition (sung–sung, sung–­spoken, tial, because it appears to be more dependent on the serial
divided–spoken) as the within-subjects variable and order of information than is reciting.
group (musician, nonmusician) as the between-subjects Word errors. Types of errors are useful in determining
variable. The ANOVA revealed a main effect of condition the nature of the memory code used by participants. For
[F(2,68) 5 11.78, MSe 5 165.78, p , .001] and no group example, one word can be substituted for another in order
effect (F , 1) or interaction between condition and group to preserve the song line structure, and this type of error
(F , 1). Post hoc Tukey tests revealed that recall did not would be expected to occur more often while singing than
differ in the two spoken-recall conditions ( p . .05) and while speaking. Indeed, words were often replaced by a
was significantly better than sung recall ( p , .01). word with the same number of syllables (e.g., “Je t’écris
The superiority of spoken recall was also apparent when cet’ lettr’ par amitié” for “Je t’écris ces mots par amitié”).
the other measures were considered. When the raw num-
ber of correct words was considered as the dependent vari-
able, the main effect of condition [F(2,68) 5 7.34, MSe 5 36
78.14, p , .01] also reached significance. There was no Professional singers
32
group effect nor an interaction with condition (Fs , 1). Singing as minor
No singing training
When considering the number of attempted lines (4, 6, or 28
Number of Participants

Nonmusicians
8 lines), a similar trend was observed, since in this respect 24
performance in the spoken conditions was also superior
to that in the sung condition [F(2,68) 5 2.83, MSe 5 20
2.14, p 5 .07]. Moreover, as shown in Figure 1, nonmusi- 16
cians could learn as much of the songs as musicians when
12
singing.
An aspect that is worth examining is serial recall. The 8
beginning of a song acts as an anchor point for the whole 4
song. This refers to the fact that the beginning of a se-
quence is a determinant for the recall of the sequence in 0
4 Lines 6 Lines 8 Lines
question (Peretz, Radeau, & Arguin, 2004). Therefore, re-
call of the first song lines should be best, and recall should Attempted Lines
decline as the song progresses. Because the first line was
Figure 1. Number of nonmusicians and musicians reaching
presented twice to participants, whereas the subsequent each level of song line recall in Experiment 1, as a function of
lines were presented only once, we considered the recall their singing experience. Nonmusicians are represented in white
of the second line, which was forgotten by only 17% of and musicians in gray shades.
 Learning Song Lyrics     247

Table 4
Mean Percentages of Errors (and Standard Errors) in Each Condition
As a Function of Line Structure
Sung–Sung Sung–Spoken Divided–Spoken
Type Line Structure M SE M SE M SE
Omissions Preserved 13 4 11 5  9 4
Altered 10 3 21 4 24 5
Substitutions Preserved 20 4 27 4 22 4
Altered 10 3 12 3 20 4

Similarly, in singing, when a word was omitted, partici- be seen, the amount of hesitations was equal for musicians
pants could replace it by a meaningless syllable (/na/) in in the singing and speaking conditions, but nonmusicians
order to preserve line structure. These omissions and sub- clearly made fewer hesitations when singing. The interac-
stitution errors were assessed with respect to the number of tion between condition and group was close to signifi-
syllables in the line. If a match was found, the line structure cance [F(2,68) 5 2.88, MSe 5 0.03, p 5 .06].
was considered preserved. When the number of syllables Finally, participants generally took a breath between
did not match, the line structure was considered altered. lines (75%) instead of during a line. While singing, 47%
The results of this analysis are presented in Table 4. Other of them took a breath after each line. While reciting,
types of errors, such as the addition of words (2% of total breaths were often taken after two or three lines. Indeed,
errors) or pronunciation errors (0.2%) were too rare to be more spoken than sung words can be produced in a single
examined. breath.
As can be seen in Table 4, errors tended to preserve the Notes. In the sung–sung condition, nonmusicians cor-
line structure, especially in singing. An ANOVA with con- rectly sang 36% of the notes (SE 5 7.8) and 65% of the
dition (sung–sung, sung–spoken, divided–spoken), type words (SE 5 4.8), whereas musicians correctly sang 48%
of errors (omission, substitution), and line structure (pre- of the notes (SE 5 7.2) and 56% of the words (SE 5
served, altered) as within-subjects variables and group 4.3). An ANOVA with material (word, note) as a within-
(musician, nonmusician) as a between-subjects variable ­subjects variable and group (musician, nonmusician) as a
yielded an interaction between type of error and line between-subjects variable revealed an interaction between
structure [F(1,34) 5 8.96, MSe 5 0.08, p , .01]. As ex- material and group [F(1,34) 5 4.61, MSe 5 413.43, p ,
pected, the omission of words tends to alter the line struc- .05]. Whereas nonmusicians recalled more words than
ture [18% vs. 11%; t(35) 5 1.90, SE 5 0.04, p 5 .07], notes ( p , .01 using a post hoc Tukey test), musicians did
whereas substitutions more often preserved it [23% vs. not. Interestingly, musicians also did not reproduce more
14%; t(35) 5 2.95, SE 5 0.03, p , .01]. This pattern was correct pitches than did nonmusicians (n.s.). When the
not significantly affected by singing, since the interaction total numbers of correct notes (M 5 13.1, SE 5 2.83, for
with condition was not significant (F , 1). There was no nonmusicians and M 5 19.6, SE 5 4.17, for musicians)
group effect, nor any interaction between group and any and words were examined instead of the proportions of
other variables. In addition, the substituted words were correct notes and words (see Table 3), there was no effect
semantically related to the target (67% of the words), thus of material [F(1,34) 5 1.85, MSe 5 97.63, p . .05] nor
keeping the gist of the line (e.g., “si jamais vous trouvez any group effect (F , 1), but the interaction was again
cet homme”/“if you ever find this man” instead of “si ja- close to significance [F(1,34) 5 3.03, p 5 .09].
mais vous tenez cet âme”/“if you ever hold this soul”). Furthermore, there was no significant correlation
Thus, participants tried to respect both the number of syl- between note and word recall, either in nonmusicians
lables and the meaning of words in their recall of lyrics, [r(16) 5 .38, n.s.] or in musicians [r(16) 5 .27, n.s.].2
regardless of the mode of vocal reproduction. Delayed recall and long-term retention. Recall after
Another factor that is known to enhance memory of a 20-min delay is presented in Table 6. As can be seen,
lyrics is the presence of rhymes at the end of lines. In performance dropped by half. Moreover, word recall ap-
order to assess the contribution of rhyme, we examined peared to persist longer after a divided–spoken presenta-
word errors as a function of their serial position in the tion. However, this trend was not significant, as revealed
line. The final words of each line—that is, those bearing by an ANOVA with condition (sung–sung, sung–­spoken,
the rhyme—were incorrectly reproduced in only 15% of
the lines (with 19% and 12.5% in singing and reciting,
respectively). This error rate was smaller than the one ob- Table 5
served for any prior position in the line (e.g., the error rate Mean Percentages of Hesitations per Line (and Standard
was 20% for the initial word of the line; t(595) 5 2.92, Errors) in Each Condition for Each Group
SE 5 0.02, p , .01). Moreover, when the last word was Sung–Sung Sung–Spoken Divided–Spoken
replaced by another word, it respected the rhyme in 39% Group M SE M SE M SE
of the cases (e.g., tour for jour). Nonmusicians  6 3 26 5 21 4
In order to assess fluency, the number of hesitations per Musicians 14 3 17 3 15 5
line was examined in each condition (see Table 5). As can Mean  9 3 21 4 18 5
248     Racette and Peretz

Table 6 more strictly sequential than spoken recall, since recall

Mean Percentages of Correctly Recalled Words (and Standard of a line was more dependent on the recall of the previ-
Errors) in the Three Conditions After a 20-Min Delay
and After Several Months (in Italics)
ous line in singing than in reciting. Yet there were fewer
hesitations when singing. Hence, singing can give an im-
Sung–Sung Sung–Spoken Divided–Spoken
pression of fluency, not because lyrics are better retrieved
Group M SE M SE M SE
from memory, but because the flow is more continuous.
Nonmusicians 36.7 6.0 26.0 7.2 38.1 5.5 This observation has also been reported in cases of speech
  2.  12 13
disorders (Hébert et al., 2003). This apparent fluency was
Musicians 25.8 6.7 32.3 6.6 42.9 7.9
14.   5 32 limited to the musically untrained; musicians did not hesi-
Mean 31.2 4.5 29.1 4.9 40.5 4.7 tate less while singing than while reciting.
The fact that music did not help text recall cannot be
ascribed to the fact that ordinary students are not used
divided–spoken) as the within-subjects variable and group to singing. Musicians, including singers, did not per-
(musician, nonmusician) as the between-subjects vari- form differently from, and recalled as many lyrics as,
able. The effect of condition failed to reach significance nonmusicians. Hence, the results suggest that oral recall
[F(2,68) 5 1.88, MSe 5 701.75, n.s.], and there was no of lyrics is a widespread ability.
group effect (F , 1) or interaction [F(2,68) 5 1.16, n.s.]. Music not only slightly impaired vocal production, it
As can be seen in Table 6, very few song lyrics were also had little impact at presentation. There was no differ-
remembered several months after they had been learned. ence in word recall between the sung and the divided pre-
The practice song, which was repeated three times during sentations. Furthermore, there was not the slightest indi-
the initial procedure, was the most frequently recalled (by cation that music helped in the long run, since the addition
12 of the 25 participants contacted), but it was sung by of music during song presentation or during participants’
only 3 of the 12 participants who recalled it. responses had no effect on long-term recall either.
RAVLT. In this standard auditory memory test, musi- Yet, all the characteristics that qualify memory in oral
cians recalled as many words as nonmusicians (see Fig- tradition also govern performance in the present study. Re-
ure 2; F , 1), and both groups improved across the five call of lyrics respected line structure, semantics, rhymes,
trials [F(4,136) 5 161.28, MSe 5 1.17, p , .001]. Fur- and front anchoring. This form of memory is commonly
thermore, no significant correlation was found between used for stories and poems (Rubin, 1995). Contrary to ex-
performance on the RAVLT and word recall from songs in pectations, music does not seem to add much to these con-
any of the learning conditions [all rs(34) , .29, p . .05]. straints. Because the focus of the task was on text recall, it
is possible that participants treated music as a secondary
Discussion task, and hence treated it as an additional demand rather
than as an aid to memory. The goal of Experiment 2 was
Contrary to expectations, music was found to interfere to examine this possibility by asking participants to focus
with rather than facilitate text recall. Participants recalled on the musical component.
fewer words when singing than when reciting, despite the
fact that words are articulated more slowly when sung.
EXPERIMENT 2
Moreover, music did not help the recovery of lines after a
memory blank. On the contrary, sung recall seems to be In this experiment, we compared the recall of text and
melody alone and in combination. We reasoned that, if
15 there was an advantage of text recall over music because
the task demands focused on words in Experiment 1, then
asking participants to pay attention to music should im-
Words Recalled Correctly

12 prove singing in general, particularly for musicians. In

contrast, if singing is a dual task, producing the melody
9
with the text should require more resources than produc-
ing either the text or the melody alone.

6 Method
Six musicians (3 women, 3 men) and 6 nonmusicians (5 women, 1
3
man) who had participated in Experiment 1, for a total of 12 partici-
pants (mean age, 23.3; range, 20–26), came back for an additional
Nonmusicians (n = 18)
Musicians (n = 18) session 11 months later (range, 5–13 months). This subgroup was
0 selected on the basis of their availability. No professional singers
participated in this second experiment.
1 2 3 4 5 The participants were presented with the sung–sung and divided–
Trial spoken versions of the same six songs that were used in Experi-
ment 1. However, care was taken to present each participant with
Figure 2. Mean number of recalled words (and standard er- the three songs that the participant had not learned in Experiment 1.
rors) on each trial of the Rey Auditory Verbal Learning Test in Recall was tested with an adaptive procedure, as in Experiment 1.
nonmusicians and musicians. Each participant once again learned each of the three songs in a
 Learning Song Lyrics     249

different condition: the sung–sung condition, in which the partici- Table 7

pant sang the lyrics after hearing them sung; the divided–spoken Means (and Standard Errors) Obtained for Words
condition, in which the participant recited lyrics presented spoken and Notes in Experiment 2
along with the melody sung on /la/; and the divided–sung condition, Production
in which the participant only repeated the melody sung on /la/. In Combined Alone
the latter condition, participants were asked to recall the full eight Dependent
lines of the song, because accuracy of note production could not be Material Variable M SE M SE
judged online by the experimenter. There were no other tasks, nor Words % 70.6 3.2 77.3 3.6
was there a delayed recall. Number 25.8 2.8 32.5 2.8
Lines   5.5 0.5   7.0 0.4
Data Scoring Notes % 58.8 9.5 50.9 6.4
The same scoring procedure was used as in Experiment 1. For the Number 23.0 4.1 20.8 4.3
sung production, raters agreed on 83% of the 562 notes produced. Lines   5.5 0.5   5.0 0.5
For the sung melodies on /la/, if the number of notes correctly pro-
duced in the first recall (Lines 1–4) was less than 80%, the score was
computed over these four lines. If 80% or more notes were recalled, was again worse when combined with music (sung) than
the second recall (Lines 1–6) was rated. If more than 80% of those when spoken. However, the trend did not reach signifi-
notes were recalled, scores were based on the last recall (Lines 1–8). cance [t(11) 5 1.34, SE 5 5.00, p . .05]. Participants
The musical note score was computed on the basis of the number
of correct notes both raters agreed upon—that is, 89% of the 565
also tended to learn fewer words and fewer lines in sing-
notes produced. ing than in speaking [t(11) 5 1.37, SE 5 4.91, n.s., and
t(11) 5 2.02, SE 5 0.37, p 5 .07, respectively]. In order
Results and Comments to assess the effect of familiarity with the task, the results
obtained by the same participants in Experiment 1 were
Because of the small number of participants, nonpara- compared with their results obtained here. Task repetition
metric tests were first performed on the data in order to improved singing [t(11) 5 2.81, SE 5 4.69, p , .05] but
assess whether nonmusicians’ performance differed from not reciting [t(11) 5 1.12, SE 5 6.70, n.s.]. Thus, par-
musicians’ performance. There were no differences be- ticipants seemed more comfortable with the task than in
tween the two groups on word recall (sung–sung and Experiment 1. However, this improvement was not suf-
­divided–spoken conditions) or note recall (sung–sung and ficient to bring word recall to a higher level in singing
divided–sung conditions; all ps . .05 by Mann–Whitney than in speaking. Again, music does not seem to facilitate
tests). Even singing on /la/ did not significantly differen- word recall.
tiate musicians from nonmusicians (see Figure 3; Mann– Melody recall was more variable than word recall, both
Whitney test, Z 5 .96, p . .05). These two groups did in singing with words and in singing on /la/ (see Table 7).
not differ in performance in Experiment 1, either ( p . The percentages of notes correctly recalled in singing
.05). Hence, all 12 participants were considered in a single with and without words did not differ [t(11) 5 1.01, SE 5
group and parametric analyses were applied. 7.76, n.s.]. In fact, note recall with words was not better
In Table 7, the combined condition refers to the sung than in Experiment 1 (M 5 50.2, SE 5 9.0; t(11) 5 0.63,
production of both words and notes (sung–sung condi- n.s.). Thus, melody recall generally appears to be poor,
tion), and the alone condition refers to the production of whether the focus is on text or music. Participants do not
words (divided–spoken) or of musical notes (divided– seem to have much flexibility in the quantity of resources
sung on /la/) only. As can be seen in Table 7, text recall they can allocate to the musical component.
The majority of participants (9) did not go farther than
the fourth line in singing on /la/ (see Figure 3), whereas
12 the majority (7) reached the end of the song when recit-
Singing as minor ing lyrics. The performances in word and note recall were
10 No singing training not significantly correlated, whether produced together in
Number of Participants

Nonmusicians singing [r(10) 5 .16, p . .05] or produced alone [r(10) 5

8 .30, p . .05].
As in Experiment 1, there were more words than notes
6 correctly recalled in singing. In an ANOVA with material
(word, note) and production (combined, alone) as within-
4 subjects variables and percentage as a dependent variable,
a main effect of material was obtained, with more words
2 produced than notes [F(1,11) 5 6.11, MSe 5 720.18,
p , .05]. There was no effect of the mode of production
0 [F(1,11) 5 3.89, MSe 5 284.39, p . .05], nor was there
4 Lines 6 Lines 8 Lines
an interaction [F(1,11) 5 2.79, MSe 5 227.23, p . .05].
The same effects were obtained with the total number of
Attempted Lines words and notes used as the dependent variables.
Figure 3. Number of nonmusicians and musicians reaching
In summary, the results are similar to those obtained
each level of melody line recall, as a function of their singing ex- previously, indicating a slight advantage of speaking over
perience, in Experiment 2. singing in text recall. This advantage of text over music
250     Racette and Peretz

does not seem related to a trade-off between the two com- contrary, it was observed that when a line was forgotten,
ponents. Accuracy in singing the melody was similar participants were usually unable to continue singing, but
whether it carried lyrics or not. Furthermore, there was that they could continue reciting after a break. This might
no correlation between words and notes recalled, suggest- be a drawback of the strictly sequential nature of sing-
ing that these two components are supported by separate ing, in which melodic lines are represented in connected
memory representations. strings with front anchoring.
Nevertheless, one important cue for auditory–vocal re-
GENERAL DISCUSSION membering that is common to both music and poems is
rhythm. The regular organization of stresses, mostly al-
The present findings suggest that the best strategy for ternating between strong and weak beats/syllables, is sup-
learning song lyrics is to ignore the melody. The melody posed to limit the words that are compatible with it, and
seems to interfere rather than facilitate word recall in thereby constrains word selection. At least in English, the
songs in both musically trained and untrained learners. rhythmic similarity between the prosodic accent structure
Music was found to be of little help for text recall in ei- of spoken words and the metric structure of the melody
ther encoding or response. Hearing the lyrics embedded is striking and has long been noted by linguists (see, e.g.,
in the melody (i.e., sung) or spoken with the melody in Hayes & Kaun, 1996) and music theorists (Lerdahl &
the background did not affect word recall, even after a Jackendoff, 1983). Moreover, Palmer and Kelly (1992)
time delay (Experiment 1) and after task familiarization have shown that linguistic accent structure and musical
(Experiment 2). The same conclusion applies to the mode meter are generally aligned in Western songs. Hence,
of expression: Performance at reproducing both the lyrics rhythmic structure, as determined by the number of syl-
and the melody while singing was either impaired (Exper- lables (or notes) and the location of primary stress, may
iment 1) or slightly inferior (Experiment 2) to the recall serve as a compatible format for setting words to tones.
of the text alone. Melody recall was generally less precise By this account, recalling a particular stress pattern in a
than word recall, whether it was sung with the lyrics (Ex- melody (or spoken text) activates a metrical grid that con-
periments 1 and 2) or on /la/ (Experiment 2), in both mu- strains the type of text (melody) that is compatible with
sicians and nonmusicians. Thus, the results suggest that in it. A common metrical grid is typically used throughout
the first steps of learning a new song, melody and lyrics a song. Therefore, metric structure provides a means by
are remembered separately, making singing a dual task. which lines of an entire song are organized in a common
The cost of singing was reflected by a 14% word loss hierarchical structure, thereby relating nonadjacent song
(Experiment 1), but it was associated with an 8% increase components and helping memory.
in the recall of notes (Experiment 2); the cost was reliable, A limitation of the present study is that we were unable
but the benefit was not. This cost–benefit analysis is more to assess the specific contribution of rhythm to memory.
compatible with the view that the melody and lyrics of First, the raters failed to provide consistent judgments for
songs are processed independently (Besson et al., 1998; the rhythmic aspect of the productions. Second, French is
Bonnel et al., 2001; Hébert & Peretz, 2001; Peretz, 1996) not a stress-based language, so it is possible that musical
rather than treated as an integrated unit (see, e.g., Serafine meter (and rhythm in general) is not as efficient a memory
et al., 1984). Thus, the present results extend to singing aid for French lyrics as it is for English lyrics. Yet, as men-
what has been found in the normal functioning of percep- tioned in the introduction, support for the contribution of
tion and memory (see Peretz, Radeau, & Arguin, 2004, for music to lyric recall in English is scant (Kilgour et al.,
a recent review and discussion). 2000, Experiment 1 but not 2; Wallace, 1994, Experi-
However, separate production of melody and lyrics does ments 1 and 2 but not 3). There are also many negative
not entail interference, unless attention to one component reports of this contribution, even in English (Calvert &
adversely affects the other. In the present case, it seems Billingsley, 1998; Jellison & Miller, 1982). Therefore, and
that lyric recall was either prioritized or much easier than even though the contribution of rhythm to lyric recall has
note recall. Such a discrepancy between the processing of not been established yet in French, musical constraints ap-
words and notes has repeatedly been found in the litera- pear to be of limited help for lyric recall in general.
ture pertaining to perception of songs, with words always This conclusion raises the question of why music is be-
being more salient than musical notes (Hébert & Peretz, lieved to be so important for verbal memory, not only in
2001; Peretz, Radeau, & Arguin, 2004). There are sev- oral tradition but also in everyday life. We believe this is
eral factors that can account for this advantage of lyrics due to a misunderstanding of the utility of music. Music is
over melody. First, the lyrics were organized like a poem, not at the service of language. In songs, music contributes
and hence their memorability benefited from the use of to the creation of a general mood that is shared with oth-
several language constraints that are known to help re- ers (Bowra, 1962; see also Thompson & Russo, 2004, for
membering (Rubin, 1995). Semantics, rhymes, and line empirical support). As Booth (1981) writes, a singer tells
structure were all found to affect recall, whether recited or people “nothing they need to decode or learn. He evokes
sung. In contrast, the melody had no semantics or rhymes, in them ways of seeing life that they already have” (p. 28).
but has rhythm, line structure, and pitch accents. These In fact, oral transmission of text is rarely word for word
musical characteristics were instrumental in decreasing (verbatim) in singing. Although singers believe that they
hesitations, making singing more fluent, but were not suf- sing the text exactly as heard, they never do so (see Rubin,
ficient to give additional assistance to lyric recall. On the 1995, for a review). This applies to music recall as well.
 Learning Song Lyrics     251

Singers, with and without musical training, never recall Engineering Research Council of Canada to I.P. We thank Claude Gau­-
note for note what they have been presented (Sloboda thier for giving us access to his song material, and Sylvie Hébert, Chantal
Bergeron, and Bernard Bouchard for assistance in creating the stimuli,
& Parker, 1985). Rather, singers memorize a schema in data transcription, and melody transcription, respectively. Correspon-
which the surface detail is not retained. Recall involves dence relating to this article may be sent to I. Peretz, Département de psy-
processes akin to improvisation that fill in structurally im- chologie, Université de Montréal, C.P. 6118, succ. Centre-ville, Montréal,
portant events according to general constraints. Learning Québec, H3C 3J7 Canada (e-mail: [email protected]).
a new song for faithful reproduction is thus a laborious
References
task that requires hours of practice.
It is interesting to note that when musicians with differ- Besson, M., Faïta, F., Peretz, I., Bonnel, A.-M., & Requin, J. (1998).
ent expertise spontaneously learn an opera song, words and Singing in the brain: Independence of lyrics and tunes. Psychological
melodies are practiced independently before they are prac- Science, 9, 494-498.
Bigand, E., & Poulin-Charronnat, B. (2006). Are we experienced
ticed together (Ginsborg, 2002). Moreover, expert singers listeners? A review of the musical capacities that do not depend on
do not take more time than novice singers practicing words formal musical training. Cognition, 100, 100-130.
and melodies together when learning a new song. Rather, Bonnel, A.-M., Faïta, F., Peretz, I., & Besson, M. (2001). Divided
they use more variable modes of learning. This is probably attention between lyrics and tunes of operatic songs: Evidence for in-
dependent processing. Perception & Psychophysics, 63, 1201-1213.
the best procedure to create detailed memory represen- Booth, M. W. (1981). The experience of songs. New Haven, CT: Yale
tations in which words and notes are tightly connected. University Press.
Hence, we may predict that in a follow-up study, a stronger Bowra, C. M. (1962). Primitive song. London: Weidenfeld & Nicolson.
association between words and melodies may emerge with Calvert, S. L., & Billingsley, R. L. (1998). Young children’s recita-
further training of the same songs under variable modes of tion and comprehension of information presented by songs. Journal
of Applied Developmental Psychology, 19, 97-108.
recall, as has been observed for highly familiar songs (Per- Calvert, S. L., & Tart, M. (1993). Song versus verbal forms for very-
etz, Radeau, & Arguin, 2004). Hence, prolonged practice long-term, long-term, and short-term verbatim recall. Journal of Ap-
may confer an advantage to singing over reciting, but this plied Developmental Psychology, 14, 245-260.
would require considerable time and effort, with probably Chan, A. S., Ho, Y.-C., & Cheung, M.-C. (1998). Music training im-
proves verbal memory. Nature, 396, 128.
little payoff for the nonmusician. Chazin, S., & Neuschatz, J. S. (1990). Using a mnemonic to aid in
Without much practice, however, experts and novices the recall of unfamiliar information. Perceptual & Motor Skills, 71,
perform quantitatively and qualitatively in a similar man- 1067-1071.
ner. This was a rather pleasant unexpected finding, be- Crowder, R. G., Serafine, M. L., & Repp, B. (1990). Physical inter-
cause it suggests that everyone is able to sing fairly well, action and association by contiguity in memory for the words and
melodies of songs. Memory & Cognition, 18, 469-476.
even in the laboratory, and that song learning is a basic, Dell, F. (1989). Concordances rythmiques entre la musique et les pa-
though difficult, skill. That is, singing appears to be a roles dans le chant: L’accent de l’e muet dans la chanson française.
musical ability that is shared by musicians and nonmu- In M. Dominicy (Ed.), Le souci des apparences (pp. 121-136). Brux-
sicians of the same culture. This result provides further elles: Université de Bruxelles.
Dickson, D., & Grant, L. (2003). Physics karaoke: Why not? Physics
support for the notion that everyone (unless tone deaf) is Education, 4, 320-323.
equipped to become musically proficient, although only Drake, C., & Palmer, C. (2000). Skill acquisition in music performance:
a minority will become experts, usually through exten- Relations between planning and temporal control. Cognition, 74, 1-32.
sive practice and explicit tutoring (see Bigand & Poulin- Gingold, H., & Abravanel, E. (1987). Music as a mnemonic: The
­Charronnat, 2006, and Peretz & Hyde, 2003, for recent re- effects of good- and bad-music settings on verbatim recall of short
passages by young children. Psychomusicology, 7, 25-39.
views). Perhaps our expert singers were at a disadvantage Ginsborg, J. (2002). Classical singers learning and memorising a new
here because they did not have the musical score to refer song: An observational study. Psychology of Music, 30, 58-101.
to, which, for them, is the normal procedure for learn- Hayes, N., & Kaun, A. (1996). The role of phonological phrasing in
ing. In the procedure used here, musicians had to draw sung and chanted verse. Linguistic Review, 13, 243-303.
Hébert, S., & Peretz, I. (2001). Are text and tune of familiar songs
on a common auditory–vocal code that is exercised from separable by brain damage? Brain & Cognition, 46, 169-175.
childhood for learning popular songs. This widespread Hébert, S., Racette, A., Gagnon, L., & Peretz, I. (2003). Revisiting
mode of vocal learning is a basic mechanism by which the dissociation between singing and speaking in expressive aphasia.
humans learn not only to sing, but also to speak. This ca- Brain, 126, 1838-1850.
pacity might very well be shaped by innate mechanisms. Ho, Y.-C., Cheung, M.-C., & Chan, A. S. (2003). Music training im-
proves verbal but not visual memory: Cross-sectional and longitudinal
The capacity to adjust vocal output so as to imitate an explorations in children. Neuropsychology, 17, 439-450.
auditory model is a remarkable ability that is rare in the Jakobson, L. S., Cuddy, L. L., & Kilgour, A. R. (2003). Time tag-
animal kingdom (Merker, 2004). Humans are vocal learn- ging: A key to musicians’ superior memory. Music Perception, 20,
ers, as are a few bird species, whales, and bats, whereas 307-313.
Janik, V. M., & Slater, P. J. B. (1997). Vocal learning in mammals.
our closest ancestors, the chimpanzees, are not (Janik & Advances in the Study of Behavior, 26, 56-99.
Slater, 1997). This confers to vocal learning a privileged Jellison, J. A., & Miller, N. L. (1982). Recall of digit and word se-
role in the study of the most sophisticated human-specific quences by musicians and nonmusicians as a function of spoken or
traits—namely, music and speech. sung input and task. Journal of Music Therapy, 19, 194-209.
Kilgour, A. R., Jakobson, L. S., & Cuddy, L. L. (2000). Music train-
Author Note ing and rate of presentation as mediators of text and song recall. Mem-
ory & Cognition, 28, 700-710.
This research was supported by a doctoral fellowship from the Cana- Lerdahl, F., & Jackendoff, R. (1983). A generative theory of tonal
dian Institutes of Health Research to A.R. and grants from the Interna- music. Cambridge, MA: MIT Press.
tional Human Frontier Science Program and the Natural Sciences and McElhinney, M., & Annett, J. M. (1996). Pattern of efficacy of a mu-
252     Racette and Peretz

sical mnemonic on recall of familiar words over several presentations. for dual encoding. Journal of Experimental Psychology: Learning,
Perceptual & Motor Skills, 82, 395-400. Memory, & Cognition, 17, 793-804.
Medina, S. L. (1993). The effect of music on second language vocabu- Scotto Di Carlo, N., & Germain, A. (1985). A perceptual study of
lary acquisition. National Network for Early Language Learning, 6, the influence of pitch on the intelligibility of sung vowels. Phonetica,
1-8. 42, 188-197.
Merker, B. (2004). Vocal learning for song: Key constraint on the path Serafine, M. L., Crowder, R. G., & Repp, B. H. (1984). Integration of
to language. Unpublished manuscript. melody and text in memory for songs. Cognition, 16, 285-303.
Morrongiello, B. A., & Roes, C. L. (1990). Children’s memory for Serafine, M. L., Davidson, J., Crowder, R. G., & Repp, B. H. (1986).
new songs: Integration or independent storage of words and tunes? On the nature of melody–text integration in memory for songs. Jour-
Journal of Experimental Child Psychology, 50, 25-38. nal of Memory & Language, 25, 123-135.
New, B., Pallier, C., Ferrand, L., & Matos, R. (2001). Une base de Sloboda, J. A., & Parker, D. H. H. (1985). Immediate recall of melo-
données lexicales du français contemporain sur Internet: Lexique. dies. In P. Howell, I. Cross, & R. West (Eds.), Musical structure and
L’Année Psychologique, 101, 447-462. cognition (pp. 143-168). London: Academic Press.
Palmer, C., & Kelly, M. H. (1992). Linguistic prosody and musical Thompson, W. F., & Russo, F. A. (2004). The attribution of emotion and
meter in song. Journal of Memory & Language, 31, 525-542. meaning to song lyrics. Polskie Forum Psychologiczne, 9, 51-62.
Peretz, I. (1996). Can we lose memories for music? The case of music Wallace, W. T. (1994). Memory for music: Effect of melody on recall
agnosia in a nonmusician. Journal of Cognitive Neurosciences, 8, of text. Journal of Experimental Psychology: Learning, Memory, &
481-496. Cognition, 20, 1471-1485.
Peretz, I., Gagnon, L., Hébert, S., & Macoir, J. (2004). Singing in Wolfe, D. E., & Hom, C. (1993). Use of melodies as structural prompts
the brain: Insights from cognitive neuropsychology. Music Perception, for learning and retention of sequential verbal information by pre-
21, 373-390. school students. Journal of Music Therapy, 30, 100-118.
Peretz, I., & Hyde, K. L. (2003). What is specific to music process-
ing? Insights from congenital amusia. Trends in Cognitive Sciences,
7, 362-367. notes
Peretz, I., Radeau, M., & Arguin, M. (2004). Two-way interactions
between music and language: Evidence from priming recognition of 1. Written delayed recall was used instead of vocal recall in order to
tune and lyrics in familiar songs. Memory & Cognition, 32, 142-152. avoid confounding the mode of production and its associated set of lyr-
Racette, A., Bard, C., & Peretz, I. (2006). Making non-fluent apha- ics. This procedure also provided an opportunity to assess the effect of
sics speak: Sing along! Brain, 129, 2571-2584. music on text encoding, as has been done in most prior studies.
Rainey, D. W., & Larsen, J. D. (2002). The effects of familiar melodies 2. This lack of correlation between verbal and musical recall suggests
on initial learning and long-term memory for unconnected text. Music that these two components are produced independently. To test for inde-
Perception, 20, 173-186. pendence between word and pitch errors, we would need to compute the
Rey, A. (1964). L’examen clinique en psychologie. Paris: Presses Uni- probability of joint word/pitch errors on the basis of the error rates for
versitaires de France. the separately occurring word (W) and pitch (P) errors (prob-W multi-
Rubin, D. C. (1995). Memory in oral traditions: The cognitive psychol- plied by prob-P, as applied by Drake & Palmer, 2000, to the pitch and
ogy of counting-out rhymes, ballads, and epics. New York: Oxford time errors obtained in piano performance). However, because 67% of
University Press. the participants had no errors on the words only and/or on the notes only,
Samson, S., & Zatorre, R. J. (1991). Recognition memory for text the probability of joint errors was mostly zero, so independence could
and melody of songs after unilateral temporal lobe lesion: Evidence not be assessed properly.
 Learning Song Lyrics     253

Appendix

Copyright 1993 by Éditions du Jour de l’An. Transcribed and reproduced with permission.

(Manuscript received December 10, 2004;

revision accepted for publication November 2, 2005.)