The Role of Absolute Pitch Memory in Oral Transmission of Folksongs

(1)

The Role of Absolute Pitch Memory

in Oral Transmission of Folksongs

An Explorative Study

Abstract:

Evidence has suggested that unlike the ability to label pitch in an absolute fashion, memory for absolute pitch in a melody is widespread. In this experiment, we examined whether there is tonic pitch consistency in recordings of folksongs. At the time of recording, there were no standardized versions of these folksongs available, suggesting the singers reproduced the

melodies from (auditory) memory only. To detect pitches, we used Yin, a pitch detection algorithm, and then manually determined the tonic pitch of the first verse of every recording. Two datasets were analysed, one consisting of five melodies (20 recordings each) and one of

two melodies (respectively, 53 and 67 recordings). In the first dataset, only one melody showed significant pitch consistency. For the second dataset, when controlled for possible

factors of variance, both melodies selected showed significant pitch consistency. Several possible factors of pitch variance are discussed, such as gender, geographical origin, and

lyrics. Together, the results are taken as evidence of absolute pitch memory in oral transmission of folksongs.

Information:

Name: Merwin Eward Olthof Supervisor: Berit Janssen

Student ID: 5974097 Research Institute: Meertens Institute

Amount of EC: 20 Research group: Music Cognition Group

Research Period: 14-02-2013 until 01-07-2013 UvA Representative: Henkjan Honing

Master: Brain & Cognitive Sciences Co-assessor: Henkjan Honing

(2)

Introduction

While the ability to instantly identify and label an isolated tone as being a particular note in the tonal system is very rare (Takeuchi & Hulse, 1993), evidence suggests that memory for absolute pitch information is in fact widespread. (e.g. Levitin, 1994; Schellenberg & Trehub, 2003). Expanding on these earlier studies, in this experiment, pitches of Dutch folksong recordings (available via the Dutch Song Database which can be found in the Liederenbank1) were analysed. The goal of the experiment was to determine whether there is consistency in sung tonic pitch height across when individuals sing the same song independently of each other. The results of our study show that there is indeed pitch consistency to be found in recordings of folksongs, and it is the first study to indicate that absolute pitch memory has a role to play in oral transmission of folksongs.

In the rest of the introduction, I will provide some background information on the subject of absolute pitch. After elaborating on the difference between two types of absolute pitch, I will discuss how singers recall melodies in general and how absolute pitch memory could play a role in this recall process. After that, I will elaborate on some evidence on widespread absolute pitch memory that leads up to our own experiment. Then, the current experiment, its goals and expectations will be discussed. Lastly, some possible contributing factors to results, such as lyrics, gender and geographical origin are mentioned.

Absolute pitch

Traditional absolute pitch – the ability to instantly identify and produce a certain tone – is extremely rare. Only 1 in 10000 individuals have this ability (Takeuchi & Hulse, 1993), and as a result it has been termed as being a gift by some researchers (e.g. Athos et al., 2007; Bachem, 1940, Gregersen, Kowalsky, Kohn & Marvin, 2001). Others insist that anyone has the potential to acquire this traditional sense of AP, but that training in a critical period is needed to fully acquire the skill (e.g. Vitouch, 2003). Interestingly, all infants process pitch information in an absolute fashion, and some researchers have suggested that humans gradually shift to relative pitch processing when they get older and as a result will lose their absolute pitch processing abilities (Saffran & Griepentrog, 2001). More recent evidence suggests that infants are in fact capable of relative pitch processing as well, if a slightly

different task is used (Plantinga & Trainor, 2005). If infants possess both relative and absolute

1

(3)

pitch processing capabilities, and use them both depending on task requirements, why would adults lose either of the two?

Evidence suggests that adults do not lose either of the processing capabilities mentioned above. The past 25 years or so, researchers have partly shifted their attention to what some experts call “absolute tonality” (Vitouch, 2003) or “residual AP” (Deutsch, 2002). This has resulted in an increasing body of evidence suggesting that the capability to store absolute pitch information is in fact retained for melodies, even among adult non-musicians (e.g. Terhardt & Ward, 1982; Schellenberg & Trehub, 2003; Levitin, 1994; Vitouch & Gaugusch, 2000). It has been argued that real AP possessors in the traditional sense – as opposed to the non-possessors – are able to label the pitches verbally (Levitin, 1994), whereas anyone in fact has absolute memory for pitch (Levitin, 1994). This is also supported by cross-cultural evidence suggesting that the prevalence for traditional AP is higher in cultures where a tone language is spoken (Deutsch, 2002; Pfordresher & Brown, 2009). This higher

prevalence is attributed to the fact that the people in these populations are better trained to derive meaning from pitch height, as they do this in their language as well.

The studies mentioned above, some of which will be discussed in more detail later, suggest that absolute pitch memory for melodies is indeed widespread. Therefore, even though singers in the recordings used in our experiment most likely had bad memory for isolated pitches, as they did not possess traditional AP, they still might have had absolute pitch memory. If absolute pitch information is retained, this suggests that other aspects of a melody are part of the memory’s representation and influence the retention as well (Levitin, 1994; Deutsch, 1999). Therefore, absolute pitch memory can be regarded as piece AP rather than tone AP (Bergeson & Trehub, 2002).

An intriguing question is whether such “piece AP” or “residual AP” does also exist for folksongs. Folksongs are especially interesting because they are often performed without instrumental accompaniment, and are sung from long term memory. This is also the case for the recorded folksongs that were analysed in this experiment. These recorded Dutch folksongs are part of a collection called “Onder de Groene Linde” which is available in the

Liederenbank. The recordings in this collection were made from the 1950s until the 1980s by Will Scheepers and later on by Ate Doornbosch. Most of the singers in the recordings were older adults, but some of the performers were younger adults or even children at the time.

(4)

Especially the older adults had often not performed these folksongs for a long time, suggesting they sang it from long term memory at the time of recording.

Memory of songs

One question that arises is how singers recall melodies from long term memory. Some scientists have proposed that attributes such as pitch (among tempo, timbre and others) are stored in a multiple-trace memory system (Levitin & Rogers, 2005) that is connected to a perceptual system (Dalla Bella, 2003). Levitin and Rogers suggest that memory for absolute pitch is a low-level feature and co-exists with an abstract memory for other features such as the sequence of relative pitches in the melody and most likely the emotion associated with the piece as well (Eschrich et al., 2008).

It has been suggested that imagery has a role to play in the retrieval of a memory. This means that, before and while singing a melody, a person is actively imagining the melody in his or her head. This follows neurological evidence from Zatorre and Halpern (1993),

demonstrating that once damage to auditory cortical areas is there (for example due to a focal auditory cortex lesion), perceptual and imagery deficits coexist, resulting in worse retrieval of memories that include auditory features.

When a person is trying to recall and (re)produce a melody, they are accordingly combining all features (such as absolute and relative pitch, tempo, timbre, etc.) into the end product, the melody itself, while getting immediate perceptual feedback from their own voice and accordingly “imagining” the melody as well. The more often a melody has been

rehearsed, the more accurate the representations are (Keller et al. 1995). This in turn results in better recall and thus reproduction of the melody. This could be a reason why musicians often perform better in musical, and specifically in song production tasks.

One interesting notion is that there is some kind of muscle memory for the vocal tract (Brown et al. 2007). This muscle memory might have an important role to play in the retrieval of a melody. Located in the motor cortex, this specific area could potentially be connected to the several representations of the melodies stored in memory as well as the vocal tract itself. In this scenario, it is as though the vocal tract “remembers” the exact tension associated with each pitch in the representation. However, this possibility thus far remains mostly theoretical, as little evidence of such muscle memory has been presented, especially in the vocal domain (see e.g. Brown & Palmer, 2012).

(5)

Evidence for widespread absolute pitch memory

Next, some evidence for a widespread memory for absolute pitch will be discussed. This evidence can roughly be divided in two types of research. One line of research focusses on identification of the correct version of musical pieces, and is thus of perceptual nature, whereas the other investigates to what degree are able to (re)produce a melody in the correct key. The first type of experiment mentioned (identification experiments) often use an

experimental design in which participants are presented two versions of an excerpt of a musical piece. One of these two versions played to them in the correct key, while the other is transposed (‘shifted’) a certain amount of semitones. The participants then have to judge which version is correct. It has been shown that people need remarkably short excerpts (as small as 100 msec) to be able to identify a familiar melody (Schellenberg, 1999). The question is whether they also have information about the correct key of the piece stored in their representation of the melody.

The first study on more widespread absolute pitch memory was of this perceptual nature and dates back to 1981. In an exploratory study, Terhardt and Ward (1982) found that 18 out of 22 participants (both musicians and non-musicians) were able to judge the correct version of a Bach excerpt above chance, even if the shifted version was only shifted up or down one semitone. However, in this experiment, the music was also presented in written form, and the participants thus had at least some (non-auditory) reference. Nonetheless, these results led to renewed curiosity for absolute pitch.

A more controlled version of this experiment was done by Vitouch and Gaugusch (2000). Even when a shifted version of a familiar tune was only one semitone higher or lower, non AP possessors were able to identify the correct version of J.S. Bach’s Wohltemperiertes Klavier with a hit rate of 59%, which is significantly above chance. Schellenberg & Trehub (2003) also used an identification task in their experiment. Under ecologically valid

conditions, even non-musicians were able to identify the correct version of familiar TV program tunes highly above chance. When the difference in tonic pitch between the correct and the incorrect version was one semitone, the participants chose the correct version 58% of times. For a two semitone difference, the participants even identified the correct version 70% of times. The (significantly) above chance results of the two studies above again suggest that at least some absolute pitch memory is retained into adulthood.

(6)

As opposed to the studies presented above, which are of perceptual nature, the current study focusses on the production of a melody. It is however not the first study on production of absolute pitch. For example, Levitin (1994) showed above chance pitch accuracy in a production task. Non-musicians were asked to select a familiar (rock) song from CD’s that Levitin provided and then sing (part of) this song on the correct pitch height in two trials. 40% of participants sang familiar rock tunes on the correct pitch in at least one of two trials, again suggesting that absolute pitch memory is widespread and also that even non-musicians are to some extent able to (re)produce these pitches from memory.

More evidence for absolute pitch in the production domain comes from within subject experiments done by Bergeson and Trehub (2002) and Halpern (1989). In the first study, mothers were asked to sing a song to their infant that they would also sing to their infant in a non-experimental setting on two occasions. They were instructed to sing the same song on both trials. Tonic pitch height (along with other musical components such as tempo) was then measured. The mean tonic pitch deviation of the second, as compared to the first performance was less than a semitone. In Halperns (1989) experiment, participants had to sing the opening tones of holiday and children songs in two trials. The opening tones of their second

performance only deviated two semitones on average, compared to their first performance. These last two studies are especially interesting, because the songs that the participants had to sing closely resembles the type of songs used in the current study in that there was no “correct” version available to the participants. There is no standardized version for these songs, so the participants did not have an external reference, such as sheet music or a recording, suggesting one particular tonic pitch to be correct. Instead, the absolute pitch information was retrieved from memory only, and therefore this memory of earlier performances (among which the participants’ performance in the first trial) served as an internal reference for their second performance.

Absolute pitch in folksong recordings

In the current study, recordings of folksongs were analysed to show pitch consistency over a large population. Oral transmission of folksongs is highly influenced by memory (as well as perception, performance, creativity (van Kranenburg, 2012)). It is thus important to show that, even for orally transmitted folksongs, there is some pitch consistency on a between-subject scale, because we can then identify absolute pitch memory as an important existing factor in

(7)

auditory memory for folksongs. As van Kranenburg (2012) has noted, once one is able to define components of auditory memory that play a role in oral transmission, one can study these and use the resulting knowledge for a theory of oral transmission of folksongs.

Next, the current experiment, its goals and our expectations are being discussed. To examine whether there is indeed pitch consistency when subjects are singing the same

melody, two datasets were analysed. One consisted of 5 tune families with ~20 recordings and was used in experiment 1. The other consisted of 2 tune families, both with over 50

recordings and was used in experiment 2. Whereas the first experiment examined only the pitch consistency in the tune families, the second also tested for possible factors of within-tune-family variance. The concept of tune family was developed by Samuel Bayard (1950). He defined it as:

““a group of melodies showing basic interrelation by means of constant melodic

correspondence, and presumably owing their mutual likeness to descent from a single air that has assumed multiple forms through processes of variation, imitation, and assimilation.” (Bayard, 1950, p. 33).

The Liederenbank groups the recordings (based on expert’s judgement) by tune family and these recordings can therefore be used as a source to explore the potential role of AP in the memory of songs transmitted in oral traditions.

Even though this study is of explorative nature, there are several possible outcomes. One possible outcome of our experiment is that all of the tune families show tonic pitch consistency. In that case, the results support a role for absolute pitch memory in oral transmission of folksongs, as perhaps can be expected from the literature mentioned above (e.g. Halpern, 1989; Levitin, 1994). However, an alternative explanation for such results is that singers sang on the most convenient pitch for their voice. Importantly however, if this is the reason for pitch consistency, the mean tonic pitch for all songs would be around the same pitch for every single song.

A more likely outcome is that some or none of the investigated tune families show inter-recording tonic pitch consistency. Such results would be more likely because of various possible reasons. One explanation for (partly) negative results could be that the singers were not able to produce the tone they wanted to produce (Levitin, 1994). In that case, the

(8)

representation of the melody might have been correct, but the singers unsuccessful in reproducing the (absolute) tone they had in mind;. This could be due to physical tone

production problems, but also due to a lack of musical training. In general, pitch memory and the perception of pitch relations are a function of musical experience (Krumhansl, 2000; Schellenberg & Trehub, 2003). The singers in the recordings of the “Onder de Groene Linde collection” were mostly not musically trained, and therefore it might indeed be the case that these people had less accurate pitch memory.

If the subjects do not have accurate absolute pitch information for every single

folksong (regardless of musical training) or perhaps even for none of the folksongs at all, this could indicate that absolute pitch information is not stored in memory for these types of songs, in turn suggesting that absolute pitch has no role to play in oral transmission of

folksongs. In that case, this shows the need of a “correct” standardized version of the song as a reference to establish absolute pitch information in memory, as Levitin (1994) already suggested. This can be the original version, played by musical instruments, or a written standard version of the song. If there is no such standard version, subjects might sing the tune on a different tonic pitch every time they sing the folksong, and therefore no absolute pitch information was stored in the representation of the song. Usually, a standard version does not exist for folksongs, and especially not for those in the Onder de Groene Linde collection, because Ate Doornbosch and Will Scheepers specifically recorded folksongs of which there was no such correct version available.

There are several other reasons for a possible lack of pitch consistency in the

recordings. In most of their recording sessions, Doornbosch and Scheepers asked the singers to sing more than one folksong. If there is no pitch consistency to be found, this might be due to influences of (earlier) performances of different songs during the same recording session. According to Levitin (1994), the tonal center of the first song that was sung during the session might simply determine the tonal center of the second song as well.

Another possible reason why there might be a lack of pitch consistency is that perhaps the subjects might not have rehearsed some of these songs often enough (Keller et al., 1995). In this case, as a result, the subjects would simply not know the song well enough to have accurate absolute pitch information for that particular song.

(9)

Last but not least, in the second part of our experiment, we controlled for three factors that might influence the absolute pitch information of a song. These can also explain any negative results. These factors are:

 Lyrics:

One of the factors could be the lyrics of the song. Contextual cues and affect (Bergeson & Trehub, 2002) have been proposed to be facilitate retrieval of performance details of musical melodies. One of these contextual cues could be the lyrics accompanying the melody. Singers most likely use ad hoc codes tied to the lyrics of the melody (Levitin & Rogers, 2005).

 Gender:

Secondly, as mentioned earlier, gender may influence the pitch on which a folksong is usually song, especially because there is no “standardized” notated or recorded version. Perhaps there is a difference in tonic pitch for women and men for the same song, due to the different fundamental frequencies men and women sing (and talk) on. These differences can in turn be attributed to physiological differences, for example the size of the larynx (Titze, 1989).

 Geographical Origin:

If there is a role for absolute pitch memory in oral transmission of folksongs, these folksongs might change over time in several due to geographical factors. In one region of the country, the population might sing a melody slightly differently than in another region. Various attributes of the melody, such as tempo and lyrics, but perhaps also absolute tonic pitch height, might differ regionally. These regional differences could potentially affect our results as well.

As has been said before, we have tested for these factors in our second experiment, but the results are tentative. To truly control for these variables, one would require a larger dataset than the one available in this study. If one would control for the factors mentioned above (gender, geography, contextual factors), one would most likely find tonic pitch consistency for every song (according to the “AP memory is widespread” hypothesis).

(10)

Experiment 1

Methods

In experiment 1, we analysed recordings of five tune families that were chosen from the Meertens Institute’s Liederenbank2

database. The tune families selected for experiment 1 were: Daar was laatst een meisje loos 1 (N=20), Er reed er eens een ruiter (N=20), Het was

laatst op een zomerdag (N=20), Al is ons prinsje nog zo klein 1 (N=19), Het vrouwtje van Stavoren 1 (N=20). All recordings were part of the “Onder de Groene Linde” collection as

recorded by Will Scheepers and Ate Doornbosch from the 1950s and onwards, and are publicly available on the internet. These recordings feature non-musicians singing Dutch folk songs. The particular tune families used were chosen based on the criterion that there were at least 20 recordings of these tunes available. One recording of “Al was ons prinsje nog zo klein

1” was removed from the data set, because it contained an instrument playing the tune rather

than someone singing it, leaving 19 recordings of this tune family for data analysis.

The recordings in the OGL databank that were used for analysis are of various lengths. In fact, one recording may have a duration of 10 minutes whereas another might only consist of the first verse. This caused several problems for the analysis. First, the tonic pitch tended to fluctuate a lot in longer recordings. This made it harder to identify the tonic fundamental frequency of the overall piece. Second, it does not seem appropriate to compare a 10 minute sample to one of 15 seconds. It was therefore decided to only analyse the fundamental frequency of the first verse of each of the recordings. The decision to only analyse the first verse was made based on both practical and theoretical concerns. This way, we always measured same “stage” of recall. Later verses might reflect a different stage of recall than the first verse, and later verses were not available for all recordings. Also it might be that the first verse best reflects the representation of the tonic fundamental frequency that the subject retrieved from memory (Halpern 1989). A last reason to use the first verse for analysis was that the representation of the first verse might be the most accurate. Children have been found to often have the most accurate representation of the first verse (Klinger et al., 1998). Most singers in the recording were elderly people however. Elderly people have been shown to better remember the first and last verse of a song, which has been attributed to the primacy / recency effect by Smith (1991).

2

(11)

According to Deutsch (1969), the brain has octave equivalent pitch categories.

Therefore, we used octave generalization. Practically this meant that an A’ was not considered any different from an A’’. Tones are often measured in frequency (Hertz), which uses a logarithmic scale. This means that, for example, 440 Hz (which is considered an orchestral A) and 220 Hz as well as 110 Hz were all judged to be in the same pitch category, A. As a

consequence, our data was of circular nature. This is because an “A” is a minor third or a major sixth away from “C”, depending on whether you are judging the relative distance of “A” to a higher or a lower “C”. This is not a problem when using circular statistics, because this type of statistics treats 0 degrees as being the same as 360 degrees. Because we wanted octaves to be equivalent to one another, we had to find a way to make an octave equivalent to 360 degrees to make sure that a lower A would fall into the same “bin” as a higher A. Thus, for example, if 220 Hertz (equivalent to A) is 0 degrees, we wanted 440 Hertz (equivalent to A’) to be 360 degrees, so that they were judged as being the same in the statistical analysis. To make the data suitable for circular statistics however, the data had to be somehow converted to degrees. The conversion of the data from fundamental frequency in Hertz to degrees is explained in the upcoming “procedure” section.

Procedure.

The author, who is musically trained, manually determined fundamental pitch of the first verse of the 5 tune families in experiment 1. A tune generator3 was used to determine the frequency (in Hertz) of these fundamental frequencies.

Yin, a pitch detection algorithm (de Cheveigné & Kawahara, 2002), was then used to detect the energy levels of various pitches in the recordings. From the output of this

algorithm, a density plot and a density scale (see figure 1 for an example of such a plot) can be derived using an implementation by van Kranenburg. These overviews can be used to identify certain energy peaks in the recordings (see figure 1). One of the goals for experiment 1 was to examine whether the algorithm was accurate enough to determine fundamental frequency of the tonic pitch. However, pitch energy varied much over recordings. In one recording, the pitch energy may have been highest for the tonic, whereas in another recording of the same tune family it was the major third that showed the highest pitch energy.

Therefore, the results still had to be checked manually by the experimenter to determine which peak represented the fundamental frequency of the tonic pitch. Nonetheless, the

3

(12)

algorithm was able to give an indication of the tonic fundamental frequency, speeding up the process of identification for experiment 2.

A: B:

The fundamental frequencies in Hertz were then converted to MIDI tones (a (in this case, as can be seen in figure 1B) continuous scaling system for pitches) to be able to do the statistical analysis with a linear scale rather than a logarithmic one. The calculation used for this was

where n = MIDI tones and f = frequency in Hertz. Because an octave can be divided in twelve semitones, every semitone can be thought of as 30 degrees. The MIDI tones were then in turn

Figure 1A: Example of the density plot used to determine candidates for fundamental frequencies of tonic pitch of recording

NLB070560 of tune family “Daar was laatst een meisje loos (1)”. The Y-axis shows the energy density of the recording, whereas the X-axis reflects the pitch height (in MIDI tones). In this particular example, the highest density peak was the fundamental frequency (around 59,5 MIDI tones).

Figure 1B: Example of a density scale reflecting the peaks of figure 1A in MIDI tones, frequency (Hz) and density of recording

(13)

converted to angles (in degrees) by multiplying them by 30. This was done to make the data suitable for circular statistical analysis (see results section). Statistical analysis was done using Oriana for windows (version 4.01, Kovach Computing Services, Pentraeth, Anglesey, Wales, U.K.)a tool specially developed for this type of statistics.

Results

Because of the circular nature of octave normalized pitch (Levitin, 1994) a circular test such as the Rayleigh test was needed to determine or reject uniformity of the data (Fisher, 1993). If Rayleigh’s test is significant, the frequency of occurrence of the sung pitches significantly deviates from a uniform distribution, in favour of a unimodal distribution. If the singers would simply sing the melody on a random pitch, logically, every pitch would be sung 8,3% of times, as there are 12 tones in one octave in the Western tuning system, which is the tuning system of the folksongs.

Figure 2A: Frequency of occurrence of pitch categories measured in degrees of tonic fundamental frequency in recordings of

Daar was laatst een meisje loos.

Figure 2B: Legend for converting measures in degrees to actual pitch categories. This figure applies for every result graph in

(14)

Manually determined fundamental frequencies of the tonic pitch for the 5 tune families in dataset A were analysed. A Rayleigh test was conducted for all of them to determine

uniformity. The null hypothesis of uniformity was rejected for the tune family Daar was

laatst een meisje loos, r =.49, p <.01 (see figure 2A). It was nearly rejected for Er reed er eens een ruiter, r =.35, p =.08. The null hypothesis of uniformity could not be rejected for Het was laatst op een zomerdag, r =.16, p =.61, Het vrouwtje van Stavoren, r =.19, p =.51 and Al was ons prinsje nog zo klein, r =.24, p =.34.

The (near) significant results of Daar was laatst een meisje loos and Er reed er eens

een ruiter indicate that there may be some absolute pitch memory involved in the recall and

reproduction of the folksongs.

The remaining tune families may have not yielded significant results for various possible reasons. Among others, one could be that the dataset was simply not large enough to generate significant results, especially if there are possible factors of variance such as lyrics, gender and geographical origin involved. To test these assumptions, two other tune families of which there were more recordings available in the Liederenbank were selected to be used in experiment 2.

Experiment 2

Method

The dataset used for experiment 2 consisted of two tune families. The tune families used were “Wat hoor ik hier in het midden van de nacht” (N=53) and “Mijn vader zei laatst tegen mij” (N=67). The procedure used to determine the fundamental frequencies of the first verse of these recordings was similar to the one used in experiment 1, except for the fact that this time, the algorithm was first used to determine possible tonic fundamental frequencies. These were then checked by the author. The conversion from frequency (Hertz) to degrees was then done in the same manner as the conversion of the frequencies in experiment 1. Also, gender of the singer(s), geographical origin, and difference in lyrics (if any) were noted for additional analysis.

(15)

Vader

F re q u e n c y Degrees 12 11 10 9 8 7 6 5 4 3 2 1 0 360° 270° 180° 90° 0° Results

The fundamental frequencies of the tonic pitch for the 2 tune families were analysed

algorithmically, and if necessary corrected by the author. A Rayleigh test was conducted for both tune families to reject uniformity. The null hypothesis of uniformity was rejected for

Mijn vader zei laatst tegen mij, r =.26, p <.01. It was nearly rejected for Wat hoor ik hier in het midden van de nacht, r =.35, p =.06. (For the graphs of these two tune families, see figure

3).

Next, additional analyses were done for “Mijn vader zei laatst tegen mij” to determine whether there was any effect of the lyrics that were sung. As it turned out, there were two major textual versions among the recordings. One textual version will from now on be referred to as “VaderVader” (N=33) and the other will be referred to as “VaderBoerenzoons” (N=32). A Rayleigh test was conducted to determine uniformity. The null hypothesis of uniformity was rejected for both VaderVader, r =.36, p <.02 and VaderBoerenzoons, r =.34, p <.02. Interestingly, the means (respectively 17 and 111 degrees, see figure 4) for both textual versions were as much as 3 semitones apart, suggesting that different lyrics, but highly similar relative melodies may have two separate representations in memory. However, yet a larger dataset is needed to be able to test for such interpretations.

Nacht

F re q u e n c y Degrees 12 11 10 9 8 7 6 5 4 3 2 1 0 360° 270° 180° 90° 0°

Figure 3: Frequency of occurrence of pitch categories measured in degrees of tonic fundamental frequency in recordings of

(16)

Gender might also to be a possible factor of variance in the sung pitches, as discussed in the introduction. Therefore, a Rayleigh test was conducted to determine uniformity when gender was included as a factor. The null hypothesis of uniformity was rejected for recordings of Vader that were sung by female singers, r = .29, p <.02 (N=49) but not for recordings of Vader that were sung by male singers, r = .24, p = .49 (N=12). Similarly, the null hypothesis of uniformity was

rejected for recordings of Nacht, that were sung by female singers, r =.39, p < .01 (N=33) but not for recordings of Nacht that were sung by male singers, r = .20, p = .56 (N=14). Intriguingly, the pitch consistency in female recordings is remarkably higher than the consistency in male recordings. However, there were relatively few male singers, which could be a reason for the non-significant results.

Text: Boerenzoons

Text: Vader

Figure 4: Frequency of occurrence of pitch categories measured in degrees of tonic fundamental frequency in recordings of

Mijn vader zei laatst tegen mij for two textual versions, Vader and Boerenzoons.

Figure 5: Geographical origins of the recordings of tune family

(17)

One last factor that we included in our analysis was geographical origin. In the Liederenbank, the origin of the recordings is also included in the description. For example, figure 5 shows the map of the Netherlands (and Flanders, the northern part of Belgium) with all the geographical origins for the recordings of the tune family Prinsje. The soft grey lines indicate the provinces. There are 12 such provinces in the Netherlands, most of which have their very own identity and culture. Most provinces did not have enough data points however to actually do a statistical analysis based on geographical origin, for both the tune families analysed in experiment 1 and those analysed in experiment 2. Two provinces however did have enough recordings for particular tune families in the second dataset. These were

Groningen (the province in the right upper corner of the map) for Vader (N=22) and Drenthe (the province directly south of Groningen) for Nacht (N=16). A Rayleigh test was conducted to determine uniformity when geographical origin was included as a factor. The null

hypothesis of uniformity was rejected for both recordings of Vader originating from Groningen, r =.38, p < .05 and Vader that were sung by male singers, r = .24, p < .03. This suggests that, perhaps, the same songs may be sung on different tonic pitches in different geographical areas. However, again, a bigger sample is needed to draw more definite conclusions about this matter.

Discussion

A wrap-up of the main results of the experiments will be presented. Then, the methods used in our study will be briefly evaluated. After that, possible explanations for the acquired results are discussed. Last, implications of the results and future research suggestions are noted.

In this experiment, we tested whether there was any tonic pitch consistency in

recordings of folksongs available via the Liederenbank. For the first dataset, one tune family (Daar was laatst een meisje loos) showed significant tonic pitch consistency over 20

recordings, and another one showed near significant tonic pitch consistency. This was enough reason to analyse the tonic pitch consistency in another dataset, consisting of two tune

families, both containing over 50 recordings. One tune family, Mijn vader zei laatst tegen mij, showed tonic pitch consistency over the whole dataset, but also when grouped based on lyrics, suggesting a possible role for contextual information such as the lyrics. The second tune family in this dataset, Wat hoor ik hier in het midden van de nacht, showed near significant tonic pitch consistency. When controlled for gender, both tune families showed tonic pitch consistency for female, but not for male singers. Also, for both these tune families, it looked

(18)

like there was significant consistency for tonic pitch when grouped on geographical origin (of the recordings). To really identify the influence of these factors however, a larger dataset is needed.

The Liederenbank was an excellent source for folksong recordings. The recordings in the database are well documented and therefore easily retrievable. They can be grouped on tune family, which is important when looking at phenomena related to oral transmission of folksongs. Yin (de Cheveigné & Kawahara, 2002), the algorithm used to detect the pitches, was useful but was not sufficient to confirm the tonic pitch itself. Therefore, the output of the algorithm had to be reviewed by the (musically trained) author to determine which peak represented the tonic pitch and sometimes, if necessary, corrected. The algorithm is thus useful, as it saves some time, but it does not automate the tonic pitch identification process.

The results indicate that absolute pitch memory possibly has a role to play in oral transmission of folksongs, even though there is no standardized version of these folksongs available to the singer. The singers thus depended solely on auditory memory while

performing. However, not all tune families showed between recording tonic pitch consistency. This could be explained by a lack of musical experience of the singers (Krumhansl, 2000; Schellenberg & Trehub, 2003) or by the fact that some of the songs were possibly not rehearsed often enough to encode absolute pitch information of these songs (Keller et al., 1995). Also, singers may have not been able to produce the tone they had in mind and were unable to correct themselves, at least in the first verse on which we based our results (Levitin, 1994). Contextual cues and affect (Bergeson & Trehub, 2002), such as lyrics, have been proposed to potentially facilitate recall and performance details of musical melodies, indicating that the various lyrics versions of the songs might also led to tonic pitch variance (Levitin & Rogers, 2005).

Another explanation could be that a standardized version is in fact needed to have absolute pitch memory for a song, as Levitin (1994) has suggested. One song, Daar was laatst

een meisje loos, has appeared in a standardized fashion in songbooks in the early 20th century. Interestingly, this was the only song showing significant tonic pitch consistency in the

experiment 1. However, this does not explain why the two larger tune families, which both did not have a standardized version, showed tonic pitch consistency after controlling for certain variables such as gender. There was a remarkably higher tonic pitch consistency among females for both the songs in the second dataset. This may have been due to puberty

(19)

voice changes for male subjects (Harries et al., 1997). As these songs are often learned at young age, male singers might have learned these songs before puberty. Because their voice changed after that, this might have distorted their reproduction of the melody as they had learned it. It was thus harder to reproduce the initial learned pitches. However, this remains mostly speculative.

Regional variance may also explain the lack of pitch consistency in the smaller datasets. If geographical origin is indeed a factor of pitch variance, as has been proposed in the introduction, this might lead to non-significant results, especially for the smaller tune families, because the song might have been sung on a different tonic pitch in different parts of the Netherlands. For example, for tune family Al is ons prinsje nog zo klein, the recordings originate from all over the Netherlands (See figure 5),

Unfortunately, if one would now gather recordings, these would not be eligible for research of oral transmission of folksongs. The recordings in the Onder de Groene Linde collection were gathered before the rise of the new (social) media. These days, nearly every (Dutch) (folk)song seems to be available somewhere on the internet, making it impossible to draw any conclusions about oral transmission. However, further research could be done on recordings from databases such as the Liederenbank and use similar methods as were used in our research.

Our research has implicated absolute pitch memory in oral transmission of folksongs for the first time, and has provided several possible factors of variance which can be studied in future research. Understanding the contribution of attributes of melodies such as absolute pitch in turn leads to better understanding of oral transmission of folksongs (v. Kranenburg, 2012). Also, in this experiment a new method to determine absolute pitch was explored. The pitch detection algorithm used has proven to be beneficial, but not sufficient for determining tonic pitch. Future research could also focus on improving this method, which could in turn lead to better insight in the role of absolute pitch memory in oral transmission of folksongs, but also to a better indication of the exact prevalence of absolute pitch in the population for other types of songs.

(20)

References:

Athos, E. A., Levison, B., Kistler, A., Zemansky, J., Bostrom, A., Freimer, N., et al. (2007). Dichotomy and perceptual distortions in absolute pitch ability. Proceedings of the National

Academy of Sciences, USA, 104, 14795-14800.

Bachem, A. (1940). The genesis of absolute pitch. Journal of the Acoustical Society of

America, 11, 434-439.

Bayard, S. (1950). Prologomena to a study of the Principal Melodic Families of British-American Folk Song. Journal of British-American Folklore, 63 (247), 1-44.

Bergeson, T. R., & Trehub, S. E. (2002). Absolute pitch and tempo in mother’s songs to infants. Psychological Science, 13, 72-75.

Brown, R. M., & Palmer, C. (2012). Auditory- motor learning influences auditory memory for music. Memory & Cognition, 40, 567-578.

Brown, S., Ngan, E., & Liotti, L. (2007). A larynx area in the human motor cortex. Cerebral

Cortex, 18, 837-845.

Dalla Bella, S., Peretz, I., & Aronoff, N. (2003). Time course of melody recognition: a gating paradigm study. Perception & Psychophysics, 65, 1019-1028.

De Cheveigné, A., & Kawahara, H. (2002). Yin, a fundamental frequency estimator for speech and music, The Journal of the Acoustical Society of America, 111(4), 1917-1930. Deutsch, D. (1969). Music Recognition. Psychological Review, 76, 300-307.

Deutsch, D. (1999). The Psychology of Music. San Diego: Academic Press.

Deutsch, D. (2002). The puzzle of absolute pitch. Current Directions in Psychological

Science, 11, 200-204.

Eschrich, S., Münte, T. F., & Altenmüller, E. O. (2008). Unforgettable film music: the role of emotion of episodic long-term memory in music. BMC Neuroscience, 48 (9).

Fisher, N. I. (1993). Statistical analysis of circular data. Cambridge: Cambridge University Press.

(21)

Gregersen, P. K., Kowalsky, E.,Kohn, N., & Marvin, E. W. (2001). Early childhood music education and predisposition to absolute pitch: teasing apart genes and environment.

American Journal of Medical Genetics, 98, 280-282.

Halpern, A. R. (1989). Memory for the bsolute pitch of familiar songs. Memory and

Cognition, 17, 572-581.

Harries, M. L. L., Walker, J. M., Williams, D. M., Hawkins, S., & Hughes, I. A. (1997). Changes in the male voice at puberty. Archives of Disease in Childhood, 77, 445-447. Jammalamadaka Rao S., & Sengupta, A. (2001). Topics in circular statistics. Singapore: World Scientific Publishing.

Keller, T. A., Cowann, N., & Saults, J. S. (1995). Can auditory memory for tone pitch be rehearsed? Journal of Experimental Psychology: Learning, Memory & Cognition, 21(3), 635-645.

Klinger, R., Campbell, P. S., & Goolsby, T. (1998). Approaches to children’s song

acquisition: immersion and phrase-by-phrase. Journal of Research in Music Education, 46 (1), 24-34.

Kovach, W. L. 1994. Oriana. Kovach computing services. Anglesey, Wales, United Kingdom. Krumshansl, C. L. (2000). Rhythm and pitch in music cognition. Psychological Bulletin, 126, 159-179.

Levitin, D. J. (1994). Absolute memory for musical pitch: evidence for the production of learned melodies. Perception & Psychophysics, 56, 414-423.

Levitin, D. J., &Rogers, S. E. (2005). Absolute pitch: perception, coding and controversies.

Trends in Cognitive Science, 9, 26-33.

Pfordresher, P. Q., & Brown, S. (2009). Enhanced production and perception of musical pitch in tone language speakers. Attention, Perception & Psychophysics, 71(6), 1385-1398.

Plantinga, J., & Trainor, L. J. (2005). Memory for melody: infants use a relative pitch code.

(22)

Saffran, J. R., & Griepentrog, G. J. (2001). Absolute pitch in infant auditory learning: evidence for developmental reorganization. Developmental Psychology,

Schellenberg, E. G., Iverson, P., & McKinnon, M. C. (1999). Name that tune: identifying popular recordings from brief excerpts. Psychonomic Bulletin & Review, 6(4), 641-646. Schellenberg, E. G., & Trehub, S. E. (2003). Good pitch memory is widespread.

Psychological Science, 14, 262-266.

Smith, D. S. (1991). A comparison of group performance and song familiary on cued recall tasks with older adults. The Journal of Music Therapy, ,28(1), 2-13.

Takeuchi, A. H., & Hulse, S. H. (1993). Absolute pitch. Psychological Bulletin, 113, 345-361. Terhardt, E, & Ward, W. D. (1982). Recognition of musical key: exploratory study. The

Journal of the Acoustical Society of America, 72(1), 26-33.

Titze, I. R. (1989). Physiologic and acoustic differences between male and female voices. The

Journal of the Acoustical Society of America, 85(4), 1699-1707.

van Kranenburg, P. (2012). A Computational Approach to Content-Based Retrieval of Folk

Song Melodies. Ph.D.. Utrecht: Utrecht University.

Vitouch, O. (2003). Absolutist models of absolute pitch are absolutely misleading. Music

Perception, 21, 111-117.

Vitouch, O., & Gaugusch, A. (2000). Absolute recognition of musical keys in non-absolute-pitch-possessors. In C. Woods, G. Luck, R. Brochard, F. Seddon, J. A. Sloboda (Eds.),

Proceedings of the 6th International Conference on Music Perception and Cognition. Keele,

UK: Department of Psychology, Keele University.

Zatorre, R. J., & Halpern, A. R. (1993). Effects of unilateral temporal-lobe excision on perception and imagery of songs. Neuropsychologia, 31(3), 221-232.

The Role of Absolute Pitch Memory in Oral Transmission of Folksongs