Acknowledgements:
I would like to acknowledge my appreciation to my advisors Berit Janssen, for her insightful knowledge and inspiring advice on the subjects about which I am passionate, and Makiko Sadakata for her extensive knowledge and continued enthusiasm and support for my studies.
Also, my appreciation goes to the city of Amsterdam from which I learned so much, and the institution UvA that provided me with such an opportunity.
I would also like to thank my family for their unwavering support and continual source of entertainment.
Abstract: This study focuses on the transmission of musical traditions. It comprises a melodic recall experiment, and then data analysis of such mental representations of folk-melody fragments. This analysis will aid us in exploring theories of stability and organization of melodies.
Keywords: Transmission, Melody, Recall
1. Background
Oral traditions are kept stable purely through cultural memory, and this phenomenon of transmission of cultures is quite remarkable to look back upon. Richard Dawkins coined the term ‘memes’ as a theory of mental content in his work The Selfish Gene (Dawkins, 1976) as a useful equivalent to the biological term genes in order to refer to units of cultural development that are transmitted, in the same way as genetics refers to units of biological development. An example of a stable musical meme within an oral tradition can be anything from a line of a folk melody to a line from Beethoven’s 5th to the melody to ‘Happy Birthday’. David C. Rubin has proposed in his work Memory in Oral Traditions that ‘oral traditions must [sic] have developed forms of organization (i.e., rules, redundancies, constraints) and strategies to decrease the changes that human memory imposes on the more casual transmission of verbal material’ (Rubin, 1995). We can say that such memes have these rules as ‘survival strategies’ to encode themselves effectively in order to be transmitted easily from person to person.
These stable ‘memes’ that exist currently in oral traditions ought to tell us something about human memory, a focus that caused Rubin to write Memory in Oral Traditions in the first place. Dawkins would term prominent aspects (in terms of memory) of culture as being memetically successful. Memes are mentioned here because the concept of units of cultural development, by whichever name one calls them, are by this way of thinking privy to the same evolutionary tendencies as genetics (Dawkins, 1976) throughout our various cultural developments. They naturally tend towards more stable states, by using such rules as mentioned by Rubin, that cause them to be
encoded efficiently upon human memory, a concept that is to be a central focus of the present study. This paper entails, among other things; a study of transmission and mental representations of such units of culture. We will be observing a series of transmissions and recalls of fragments of meme-sized western tonal folk melodies by a collection of individuals. For musicological purposes, we wanted to use a group of existing melodies, and we also know that they are all stable in some sense to have been so memetically successful in the first place. This would inform us about the cognition and mental representation/memory of these aspects of culture, but also about the music-memes themselves, what ever the focus - studying the transmissions of the folk melodies would inevitably tell us about mental representations.
A certain level of predictability would tend to mean that the human brain does not have to work very hard in order to understand and remember information. Our memories and predicting capabilities are formed by our personal experiences and are affected by our cultural memories. The ‘folkish’ collective memories that we share, coupled with the individual life experiences; tend to inform our expectations, all for the purpose of efficiency. Throughout time, all functioning animal minds use their memories in order to become more adept at predicting future events. In other words, they anticipate future events for aid of survival - for ultimately, failure to do so in such harsh environments is detrimental not only to the individual, but potentially to the offspring (Huron, 2006). Therefore those genes of phenotypes incapable of predicting tend to have been lost through the machinations of time at this stage. As David Huron states in his book Sweet Anticipation: Music and the Psychology of Expectation, ‘natural selection has favored the development of perceptual and cognitive systems that help organisms to anticipate future events’ (Huron, 2006). Predicting something accurately even something small such as a plottwist in a film -generates a sense of relief or satisfaction (Huron, 2006). Expectation is one of the tendencies of the mind that helps individuals make sense of, and remember patterns for future reference. In the present case, we will be addressing aural patterns.
In a paper by Marcus Pearce and Geraint Wiggins on Auditory Expectation, there is presented a probabilistic model of the phenomenon of expectation in music, and they
assert that in general, in order to deal appropriately with changes in the environment, expectations must be grounded in processes of learning and memory (Pearce and Wiggins, 1991), and the central idea of their paper is around information transmission via musical structure during the listening experience; in context of prior knowledge built up in life experiences. In other words they emphasize, as we would like to emphasize in the present study; that our experiences shape our predicting capabilities (not least in music; although music will be the focus of the present study also). Simply put, we think to past events and discern such patterns in order to anticipate future events, in every aspect of our existence, for our brains are evolutionarily predisposed to do so, for evolution has favored cognitive systems to function in such a way - and the case holds true with the inherent satisfaction of music.
Artistic output in general exploits our memories and the ability to predict future events that formed from our memories – both the memories acquired through our perception in relation to our own experiences, and those stories told to us of past lives for emotional impact. A story or a film might play on our memories of formats we have seen before and therefore we will build an expectation based on what stories we’ve heard before, or experiences we’ve had in order for the events to yield to and play with our expectations for emotional impact. We can say that one of the properties of art is that this emotional impact or significance is communicated through abstract means, and music is of course no different in this case. Music is unusual even when compared to other art-forms for it is, to use Georgina Born’s apt term, an ‘utterly self-referential aural abstraction’ (Born, 1991) and not as easy to verbalize as images are in terms other than its own. However, like other art forms it acquires its significance through memory. With a painting, one can observe the image and relate it to an experience or an idea that holds significance in terms of their life. For example: the Rorschah inkblot test has been in use in psychology in order to assess a person’s emotional functioning by finding out their verbal interpretations of ambiguous designs, so we can say that by such an example, artistic design does serve a function in telling us about individuals’ minds, memories, and emotions. Music however in a sense could be considered as being more comparable to cuisine than image or denotative language; with its own (self-referential) motifs however, it still holds similar emotional impacts as do other forms of art, and is therefore able to be
remembered and passed along from person to person. According to David Huron, emotions are ‘motivational amplifiers’ (Huron, 2006) as emotions encourage
organisms to pursue behaviors that are normally adaptive, and to avoid behaviors that are normally maladaptive, this of course being an adequate evolutionary purpose or explanation to having emotions, and to having art. Huron then also goes on to explicate how performing arts have featured significantly in our attempts to understand the dynamics of human emotion (Huron, 2006). In particular, music exploits such tendencies of the human mind, as well as in particular our tendency to remember significant experiences and derive satisfaction from anticipating future developments from patterns wrought from past events.
Music tends to increase our self-awareness and awareness of others, which is why music in particular has served such a prominent function in a variety of social
situations, we can agree that it communicates, it is just harder to say by which means. As with our various forms of artistic output in general, cultural norms tend to dictate what is significant, what will be remembered and therefore what forms our
expectations. Like with story telling, playing with learned/ingrained expectations is at the heart of the satisfaction of music (Krumhansl, 2002). Our expectations of music are informed by what music we’ve heard before (Pearce and Wiggins, 1991) Our musical cultures and individual experiences would affect and inform our musical expectations, preference and consequential emotional impact and memory retention. Continuing from Huron’s work (2006) and through to Pearce and Wiggins (2011) we can assert that the enjoyment of music comes from it establishing predictive patterns, similar to the satisfaction and emotional impact of significant events in daily life, which build meaning with repetition. Part of our evolutionary make-up as well is that if we predict something accurately, it stimulates the reward-related neural circuits (Salimpoor et al., 2013). Meyer also proposed that the rise and fall in tension expressed in the pitches is relating to the inherent satisfaction of music in his 1984 lecture to Stanford University named Music and Ideology. With music, exploitations of pitch and rhythm in this sense are means of allowing time to pass in a more meaningful fashion, and every known culture has a form of communication such as this, and so significance in this case would be relative to experience. Remembering something at the right time is satisfying, for instance: taking a gamble and reaping
rewards is satisfying, as is running for and managing to catch a bus. The general process in accurately predicting an event and the risk being satisfied with the correct outcome is a tendency of an evolutionarily viable species. These tendencies are at the crux of the satisfaction of music, and therefore emotional significance is at the heart of remembering as well.
Existing folk melodies are meaningful to many and memetically successful, and therefore in those terms, they are successful if they are still heard today for they have been kept ‘alive’ by the memories of a number of minds through time. It is also logical to say that repetition entails that we are more likely to remember something, and music tends to contain a lot of repetition in its structure in order for it to be in any way pleasant or coherent, most of which we are unconscious of. For instance, pitch and rhythm might work together to form a structure, and as a result; rules (or as Rubin would also put it – constraints) of what is to be most efficiently remembered and expected.
For one thing, the ability to anticipate a beat is something we developed rather
recently in terms of evolution of man (Sacks, 2003). Apes cannot predict a measure to the same degree we can – while there are examples of budgies moving to music, and of course passerine songbirds learning melodies from one another (the most
complicated of which are for sexual selection) the ability to predict ‘beats’ so far seems to be a distinctly human ability, and so in this way this aspect of our
expectation is very telling of our biological and cultural evolutions. We are also the only species with watches, who think about and trivialize time so consciously. Our tendency to remember and form predictive patterns out of past events in mind of potential future events is entirely applicable as a survival mechanism. A line of melody that speaks in the ‘tonal idiom’ (Lerdahl and Jackendoff, 1983) that we are familiar with would be easier to be remembered and recalled than a series of
arbitrarily tones; because in arbitrary tones there would be no inherent rise and fall in tension, and no emotional impact or meaning. Sometimes musical meaning can even imprint itself even without our permission in our minds in the form of earworms (Serra, 2012), or the tendency to loop small portions of tunes over and over again in the human mind.
Music is something that forces an interpretation to any person who hears it, and so if we are imagining a world without recordings and written notation, the most inherently memetically powerful melodies would be able to transmit themselves from mind to mind throughout time using the evolutionary faculties of memory to do so, having certain traits or rules or constraints in order to encode themselves (if we were to think about it in a deterministic sense). The listener in turn makes use of whatever musical knowledge they have acquired in their lifetimes with which to form an analysis, whether a musician or not. Let us say that we consider a hypothetical situation of an individual listener of tonal music being presented with a series of arbitrary notes that do not fit into any key, and doesn’t have discernible tempo or predictable rhythm. The individual listener therefore presumably cannot decipher any sort of discernible pattern and so the listener would likely be harder to remember than something that had followed rules they were familiar with such as stayed in the key of G, kept an even, predictable tempo, and had a rhythm that had a discernible pattern. Meyer has stated that ‘in axial melodies, the prime process generating tonal tension is that of departure from a central pitch, followed by the satisfaction of return to that pitch.’ (Meyer, 1984) – in other words, a melody that satisfies expectations must follow some sort of recognisable structure or set of rules with which to build tension and to satisfy the listener. Music with such forms of organization would therefore be more easily remembered than one that didn’t build any in the first place.
Lerdahl and Jackendoff provide an allegory of music to denotative language, which is that ‘In order to appreciate the poetic or dramatic structure of French, one must first understand the tenents of the French language… Similarly in order to appreciate Beethoven quartet as art one must understand the idiom of tonal music’ (Lerdahl and Jackendoff, 1983). There are features that musical cultures tend to have in common, such as there being a coherent mathematical relationship between the frequencies of pitches such as octave equivalence, whose frequencies will have a mathematical relationship of 2:1. While the specific rules for the way tonal material is arranged tends to be at least partially culturally relative (such as there being multiple modes and microtone use in some cultures), the pentatonic scale tends to be used all around the world (Tan, 2010) for instance. Aniruddh Patel pointed out that ‘In every culture there is some form of music with a regular beat, a periodic pulse that affords temporal coordination between performers, and elicits synchronized motor response from
listeners’ (Sacks, 2003). From this point of view we can conceive of music as a patterned and predictable code – regular pulses being something that is seemingly universal among humans. We can also guess that the earth might take 24 hours to complete a cycle of rotation, or 365 and-a-quarter days to orbit the sun. Whether this expectation is partly biological or entirely culturally relative goes back to a
philosophical nature/nurture debate far beyond the scope of present study. However, suffice to say that this process of expectation and memory that is exemplified through music is at the crux of the study of humanities.
With a melody, from the onset of a note or two, a listener of any form of music is presumably already given a clue of what to expect if it has followed some familiar rules/constraints already and for it to follow a format of sorts, for they already have a memetic framework from which to build expectations. This particular predictive aspect was explored in the paper Expectancies generated by Melodic Intervals (Thompson et al., 1997), in which the researchers asked highly trained and non-trained musicians asked them to complete a melody based on an interval. This study revealed that all results were similar between musicians and non-musicians. The results of the Expectancies paper seemed to suggest that listeners of tonal music, no matter what the musical expertise, might typically form similar expectations, whether unconscious or conscious. One might be that a new tune will remain ‘in key’ for example, and perhaps they can expect the intervals to be small and for the rhythm to be repetitive and easy to move along to. To a listener of a particular musical culture there are certain expectations built by that cultural memory - intervals (Thompson et al., 1997), progressions, modulations and cadences. We think that these effects of the expectations will play into the memory and recall element of the present study. This thesis is also focused around European tonal music (specifically Dutch folk music), and therefore what Lerdahl and Jackendoff describe as the ‘idiom of tonal music’ (Lerdahl and Jackendoff, 1983) from a European cultural experience and background. We consider Rubin’s means of oral traditions organizing themselves in terms of rules/constraints in the present study, in particular; to form our experiment design and hypotheses. In this case the rhythm and the pitch would be two constraints that might facilitate memory of music as Rubin describes, Meyer also having used the term whilst discussing tonality, asserting that: ‘the constraints of tonality were
indispensable for the organization of music’ (Meyer, 1984) – and that which fits our processes of organization is easier to remember. As Lerdahl and Jackendoff explicate in A Generative Theory, there can be coherent rules of musical grammar applied to tonal music. In the present study, we seek to examine the encoding/memorability of structurally stable and structurally unstable tonal Dutch folk melodies on participants. Given that the experiment is performed in Amsterdam and utilizing Dutch folk melodies we can assume that the effects of being immersed in a musical culture will hold true in this case.
In direct relation to a Sloboda and Parker paradigm presented in the paper Immediate Recall of Melodies (Sloboda and Parker, 1985), we will present a similar recall experiment in order to examine the mental representations of folk melodies. Sloboda and Parker presented eight participants with folk melodies played six times to each participant in succession, and had them recall the melodies back to them while they recorded. The recordings were thereafter transcribed by a human expert for purpose of metrical comparisons of the transcriptions of the recalls to the original stimuli. They gaged that while the recalls were never perfect – indeed, they did not improve over the six impressions of the melody - every participant showed that they understood the melodies in some sense by (while they especially never got the rhythm perfect) they would often substitute metrically logical equivalencies to the original.
In this case Sloboda and Parker manually compared the melodic contours of the recalls to the original stimuli, whereas we will be using computational analysis in order to assess the faithfulness of the transcribed recalls to the original. They
presented a new melodic sequence from a folk melody to each participant to hear six times successively, and the participant was to give an immediate sung recall after each hearing of the melody, which was recorded. Unlike previous studies on memory recall, the Sloboda and Parker paradigm offered an examination of how mental representations of a melody build up in memory over time, and how memory decays – which has, as Mullensiefen and Wiggins (2010) point out, definitive ‘ecological validity’1 – and yet as Mullensiefen and Wiggins also pointed out, we have not seen many of papers following this model of recall since. In the Sloboda and Parker recall 1
paradigm, the main findings were, firstly, that in recalling new melodies, structural knowledge (how the ideas within each melody were related and organized) was seemingly integral to musical memory; and, secondly, that metre is primary structural frame for melodic comprehension and recall. They also found that all subjects tended to show a basic understanding of the melodies by breathing more often in between phrases than otherwise, which also shows some understanding of the idiom of the music.
This was subjective human judgment analysis that will be ultimately hard to move away from, even thirty years later. In Mullesiefen and Wiggins’ perspective of the paradigm, they problematize the analysis and attribute it largely to the dependence on manual analysis, which had to deal with as they called it ‘dirty musical data’
(Mullensiefen and Wiggins, 2010). They reanalyzed the Sloboda and Parker's data computationally, and attested for some of the previous accuracy problems in the Sloboda Parker paradigm. They present their computational perspective so as to be able to analyze such data using MIR methods for ease of efficiency.
Like in the Sloboda Parker analysis, we will be making structural pattern comparisons with these transcriptions of the recalls of all conditions with the originals stimuli presented, in order to analyze the accuracy of this transmission. However, for the present study we will firstly; be collecting our own data from recalls of tonal folk melodies, and secondly; bring in an algorithmic pattern-finding system named Structural Induction Algorithm Matching as coined by David Meredith (Meredith, 2006). The similarity measures will be implemented and calculated for each recall2 by an algorithm (Janssen, 2015), one that will be looking at parallelisms between the recalls and the stimuli such as pitch and onset and generating a similarity value for each recall based on these criteria. In the present study, we hope to investigate a musicological question, by examining the transmission of existing folk melodies themselves, music cognition by examining how they are remembered and recalled with techniques from Music Information Retrieval.
Crucially, we want to examine in particular not only the mental representations of 2 https://github.com/BeritJanssen/MelodicOccurrences
folk-melodies, but also concepts of stability of oral traditions through such a sample of mental representations. Along with this, we are interested in the rules of stability that are theorized by David C. Rubin and Leonard Meyer. However, a recall paradigm is where we’d like to start, and therefore we take inspiration from Sloboda and
Parker’s experiment design, and theories of stability inspired by Dawkins’ theory of mental content, Rubin’s rules of stability within oral traditions, Huron’s, Pearce and Wiggins’ theories of expectation in relation to memory, and Structural Induction Algorithm Matching as a method of computational analysis in comparing musical scores, as explicated by David Meredith.
2. Experiment design
These observations on Sloboda and Parker’s recall are important when embarking upon a new related study, and we keep in mind the initial design, the results and the stated problems, which they themselves point out are largely down to a lack of technology (Sloboda and Parker, 1985); and some new perspectives offered by Mullensiefen and Wiggins' application of MIR methods to their existing data (Mullensiefen and Wiggins, 2010).
The design of the present study was inspired in part by the aforementioned Sloboda and Parker 1985 recall in terms of both the experiment design, and of their
observations and deductions from their data, which are all aspects that we keep in mind. Like them, we also present a selection of folk melodies to participants, but whereas their aim was to find out about the mental representation of melody over a succession of listens of the same melody, we additionally want to find out about how a participant recalls both immediately and after a delay on melodic phrases that are presented to each participant only once.
In short, the present study will also emtail a recall experiment, with a few named crucial differences to the Sloboda Parker paradigm in terms of the computational method of the transcription comparison, and with a focus of memetic stability through mental representations.
While the design of the present experiment maintains similarity to Sloboda and Parker’s experiment, the present study will ultimately be a departure from theirs nonetheless - while Sloboda and Parker present in their paper a recall experiment of
melodic sequences of folk melodies recalled by eight participants and subsequent manual transcriptions, their analysis thereafter involved manual structural
comparisons of the transcriptions of the recalls against the manuscripts of the original stimuli, and; whereas this has validity, it is ultimately a subjective form of analysis. The present study also involves manual transcriptions of the recalls, but the analysis involved in the present case utilizes Structural Induction Analysis Matching between the transcriptions of the recalls and the original stimuli. To be more specific, SIAM (Meredith, 2006) will be matching the onset and the pitch of the recall transcriptions to the original queries/stimuli, in order to give us values with which to draw
conclusions. This is an effective and definitive use of MIR methods in order to give us an objective measure of accuracy to work with for efficient quantitative analysis. Coupled with this, some manual estimations will be involved for an initial comparison to check that the computational analysis does not diverge with the musical intuition of the author.
Unfortunately, as of yet, the computational transcribing methods are not so advanced or available so as we can use them for our study to have totally objective digital methods, and therefore there will still be the initial problem of the manual transcriptions of data (Mullensiefen and Wiggins, 2010). Within the experiment design of the present study we look to collect new data and test different melodic stimuli, although as with Sloboda and Parker’s paradigm, our stimuli/datasets also consist of folk melodies. This will be in order to shed light on the ways in which participants make mental representations of them, in order to look at the process of transmission and mental representation of existing melodies that are shown to be stable, but also out of interests of the melodies themselves. This makes the assumption too that the recalls are representative of the mental representations. For the present experiment, we decided to use a dataset of six stable melodies and a dataset of six unstable melodies from a Dutch folk music database which has been catalogued and annotated by the Meerten’s Institute (Van Kranenburg, 2014). This is a project that previously asked collection experts to annotate how phrases in different variants of a tune family of folk-songs are related. This consists a database of 170,000 Dutch folk songs, 30,000 of which contain references to musical notation or
recordings. Within this folk melody database, which comprises of folk melodies annotated by collection experts, contains a selection of 360 melodies categorized into 26 tune families. The phrases that we hypothesize as being stable are deemed so as they were performed the same way for all sources of the song, and the ones we hypothesize as being unstable phrases being those phrases within a tune family that have a unique label as there was no other variant of the tune in which the tune was performed the same way - in the words of folklorist Samuel Bayard, ‘the salient features that members of a tune-family have in common are considered stable as they appear innumerable times from many sources’ (Bayard, 1950). For our selections, we consider that our stable phrases contain such salient features as they have appeared many times from various sources, and the unstable ones have appeared only once. From this dataset we selected twelve different Dutch folk melodies, one dataset of six that are considered to be stable and another dataset of six melodies that are considered to be non-stable. We expect a phrase that has appeared unchanged from several sources, we will consider it to be maximally stable for this study, and if a phrase only appears once, with few relations, and we are assuming that it is susceptible to being changed through transmission and it is therefore deemed unstable for this study. From these frequently appearing salient features, we can assume ‘stability’ in some, and from a phrase standing on its own, we will assume ‘non-stability’. We can say that there is proven stability in the Meerten dataset by merit of it comprising of existing in folk tunes in general, the Meerten’s dataset having been put there for use by studies of this type.
In mind of this, and in mind of certain rules organization of oral traditions that make musical memes more easily transmitted, we decided to design a three-tiered
experiment, one tier which focused exclusively on rhythm, one tier which focused on pitch, and one that contains both – a constrained, melodic condition, constrained meaning that they contain both elements of pitch and stability to encode themselves effectively. In the present study, each participant will complete three trials, one of each of the three tiers/conditions and so every participant will receive a hearing of three melodic phrases, one generated under each condition – one will be pitch-based (with the rhythm taken out), one will be rhythm based (with the pitch taken out), and one of them contains melody (with both elements of pitch and rhythm left in). With
the present study, we would like to elucidate something about the different constraints as a theory of the transmission of oral traditions (Rubin, 1995) – the constraints that we are referring to here are the pitch and the rhythm. Therefore with the two datasets (one stable dataset and one non-stable dataset) and our consideration of these
constraints, we present a three-tired recall experiment generated out of the twelve phrases.
Like in the Sloboda and Parker paradigm, we decided to take one immediate recall after the stimulus is played and recorded on a computer, and, additionally; we will take another delayed recall after 7-minutes. During these times will ensue three distractions that will be successive clips from the film “Cast Away”, which have been chosen as to show some sort of continuity of the storyline, which the participant is encouraged to focus on, and the clips will be the same for each of the participants. This film has been chosen in particular out of a few films which are not too distracting and without any music in them, as the presence of music might distract the participant from the musical task they are to perform. This is because for this delayed recalls it is essential that there are distractions that are the same level of intensity for each
participant, and it is preferable that the distractions are not boring for the participants, and that are reasonably absorbing enough so as to make the experiment interesting and comfortable over the span of 25 minutes. We hope that by asking for two recalls of each condition we can elucidate something of the decaying representation of the phrases – stable and non-stable, generated into the three conditions – rhythm, pitch and melody; which would be a departure from the Sloboda and Parker paradigm, whilst keeping in the essential element of recall-recording-transcribing-analysis intact. This experiment design will therefore offer a slightly different perspective of the mental representations of the melody than did the Sloboda and Parker experiment.
The present experiment is formulated in such a way that the method the participant chooses to remember the melody is up to them as well, and the researcher merely states the task to them that they are to recall the played melody and will be expected to recall it after the clip. We are aware that different people have different strategies of remembering and recalling, but this is an aspect of individual preference that is also
of interest to the present study. We presume that the recalls will be related to the original in some sense for everyone and one of the interests here is many participants will be presented with the same few melodies. In the present study, participants will of course not be aware whether they are about to sing an easy or difficult (or as we now deem it: ‘stable’ or a ‘non-stable’) melody while they are performing the experiment, or even that they are performing a folk-melody. The participants are simply given the impression that they have been presented with an arbitrary melody, one that they are to do their best to recall for the experiment. The author is not even certain at the time of testing which of the melodies are deemed stable or unstable. What the participants will be aware of is, before embarking upon the experiment; that they will be performing three tasks that are similar to one another, as they have to know whether they are singing or tapping before each condition starts, and also that the experiment will last roughly 25 minutes. The conditions mentioned are designed in order to analyze the aspects of the melodies that tend to be best remembered, and in combination of the initial and the delayed recall aspect, which aspects tend to be retained or degrade over time in the mental representation. The participants are not informed of this of either course, although we do not rule out the possibility that some of them will be able to guess.
One condition being purely pitch-based, one condition being rhythm-based (not sung but tapped) and one condition being both/melodic (pitch and rhythm), and two recalls of each interspersed with a clip of film, each of the conditions are to be run
consecutively and the order in which the conditions are presented is arranged for each according to Latin Square design as the order of the film clips are always the same, as they likely vary in intensity. The recalls are followed by a questionnaire, which will ask them about their method of remembering the stimuli over the delay time and which condition they found to be the most difficult. The questionnaire will also detail the participants’ demographics and musical experience (both of which will be taken into consideration in analysis) and we will classify it as having five years of recent musical experience within the past five years.
The recordings - which are to consist of six from each of the participants altogether, are then to be transcribed by a musician, and the transcriptions are to be compared to
the original stimuli for the structural comparative analysis. It is inevitable that before this point – after the experiments have run – that the author has a reasonable idea of how accurate the recalls are after having played them and recorded them, but tries to be as impartial as possible during the experiment, and act neutrally accordingly. However, before embarking on computational analysis, we believe it is important that a human gages an idea of accuracies just in case there is an error in the algorithmic computation, and therefore take a rough estimate of the similarities between patterns According to Lerdahl and Jackendoff, pattern-finding (which they call parallelism) is integral for coherent musical analysis, and has been central aspect of musical analysis stemming from Schenker (Schenker, 1907) and we expect to find, as did Sloboda and Parker, that there is a patterned relationship between all of the recalls and the original stimuli. This pattern-finding aspect of musical analysis is also related to the method of analysis we will be utilizing named Structural Induction Analysis Matching
(Meredith, 2006) that will compare our transcriptions to the original queries computationally. In Meredith’s words, this SIAM algorithm ‘takes as input a
multidimensional point set and discovers, for every vector, the points in this point set that are mapped onto other points in the point set by that vector.’ (Meredith, 2006). The vectors in question here take into account the pitch positioning and the onset of each note. With this SIAM method, the similarity measures are generated by
comparing the onset and the pitch; and generating a similarity measure value, based on these criteria. A value between zero and one is to be generated for each recall, for the purpose of efficient and objective comparative analysis. Each recall will receive a value that represents the accuracy that is between zero and one; zero being a totally inaccurate or unrelated score to the original query, and 1 being totally faithful to the original. This we will consist an objective measure to our hypotheses.
3. Research Hypothesis
For the hypotheses, we consider the above-mentioned musicological discourses, the theories of stability in cultural traditions inspired by Dawkins (1976) and Rubin
(1995), as well as the observations of the results from Sloboda and Parker (1985) and the relatively recent reanalysis of their data by Mullensiefen and Wiggins (2010), as well as in accordance with the other texts mentioned, in particular; Huron (2006) and Pearce and Wiggins (2010). We now present a series of hypotheses in accordance with the results of we formulate our central hypothesis on stability from a project from a project named from the Meerten’s institute named Tunes And Tales (Janssen et al., 2015); from which the melodies are deemed to be stable, by merit of appearing multiple times unchanged from various sources having been performed in the same way in several cases - whether written or sung. In the same way, the unstable phrases we use are those phrases for which exists no other variant in the associated tune family in which the phrase was performed in the same way - so within the collection the most unstable phrases phrase appears in their precise form only once. Therefore we hypothesize that the stable phrases will remain unchanged and be recalled significantly more faithfully than the unstable phrases across both recalls for all participants overall.
This first hypothesis being that we expect that the stable phrases which we have been chosen for analysis to be recalled significantly more accurately by participants in general under all three conditions (which are the pitch condition, the rhythmic
condition and the melodic condition) than the unstable phrases, both in the immediate recall and the delayed recall; in other words, we expect that the average SIAM
(Structural Induction Analysis Matching) measure will be significantly higher for the stable phrases than for the non-stable phrases. In addition to this, we expect that the most significant differences between the accuracies of the recalls of the stable and unstable melodies will be seen in the melodic (constrained) condition. This is in spite of the fact more information may be passed on with both pitch and rhythm together; in a stable melody, these will constrain one another and limit the search space for what to expect next and therefore be more easily remembered. In other words, the pitch and the rhythm will work together in order to facilitate memorization - in the words of Rubin for the ease of ‘casual transmission’ (Rubin, 1995) - in order to imprint themselves on a human’s mind and also remain stable (unchanged). In other words, Rubin states that oral traditions, which include melodies, require forms of organization or rules such as these to limit the scope of what to expect next, which in turn; will aid them in encoding themselves easily (Rubin, 1995).
In line with this notion of such constraints facilitating casual transmission, our 2nd hypothesis is that the melody-condition phrases will be remembered more accurately than the pitch-condition phrases and the rhythm-condition phrases. This is purely out of the position on constraints and therefore in the case of this hypothesis, it is outside of our theories of stable versus non-stable melodies. The unconstrained conditions (that is to say the pitch conditions and the melody conditions on their own) will presumably have less context/more options (and therefore be less predictable and less memorable) without the two elements of pitch and rhythm ‘constraining’ one another. We also think that the pitch phrases will be recalled more accurately than the rhythm phrases. The pitch may be more constrained by the rhythm and vice-versa, but we think that given the constraints this will make less of a difference. This marginal assumption of an elevated proficiency of pitches as compared to rhythm is more to do with the exposure of culture in Europe where generally the popular musics are more commonly more complicated pitch-wise than rhythm and time-wise (most times being 4/4 and most rhythms being somewhat repetitive). Also, in popular musics there is more of a focus on singers than percussionists in any case, and pitches tend to alternate more than rhythms. This assumption of pitch being remembered more effectively than rhythm here is also in accordance in part by Sloboda and Parker’s deduction that participants do not tend to reproduce the exact rhythms of the original, but that they tend to substitute metrical equivalents in about half of the cases.
However our analysis will not involve the acknowledgement of metrical equivalents, but will be comparing the onsets (and the pitches), and therefore the aspect of metrical equivalency will not be taken into account
Our 3rd hypothesis is that participants will generally perform the recalls more accurately in the immediate impression than in the second impression, as memory tends to decrease and degrade and mutate elements over time. It will ultimately be curious to observe the effects of the melodies deemed stable and those deemed unstable within this degradation process, as well as the degradation of the particular conditions to see which is best retained. As well as this, we may be able to find a significant difference between the musicians and the non-musicians within this degradation/retention process, as musicians may have developed more effective strategies for remembering music. Sloboda and Parker saw in their recall experiment (Sloboda and Parker, 1985), that the recalls did not improve even after six hearings of
the same melody – indeed some of them got longer but not better; and for the present study, we are presenting each melody only once. Therefore, from their resulted effect, we can assume that the mental representations of the phrases will degrade between the first recall and the delayed recall, under all three conditions; although this my depend on the participant's preferred method of memorizing the melodies, and therefore this one may also specifically depend quite heavily on the individuals’ reaction to the experiment – some may find it easier to sing a reproduce a second time and ‘correct’ their first recall if they have a good memory; individuality is ultimately a difficult factor to eliminate in the proceedings.
Our 4th hypothesis is that those with musical experience will tend to perform better than those without musical experience, with all phrases and on both recalls in all conditions - as they are more likely to have had to perform a task similar, more likely to be conscious listeners of music, and by sheer merit of having musical experience, as mentioned; have more effective strategies for remembering music than non-musicians out of previous requirements. We can guess that non-musicians will give more faithful recalls to the original stimuli than non-musicians in every aspect purely because they are presumably more used to consciously thinking about music, they are likely to have performed music prior to this, and it is possible that they have
performed a task like this previously, such as in a graded aural exam.
For the purpose of clarity: we hypothesize that the most accurately remembered melodies are those that are stable as opposed to those that are unstable. We think that this will be more so the case under the condition of ‘both/melody’ (both rhythm and pitch), while the least accurately remembered ones will be the unstable recalls. We hypothesize as well that the melodic condition will be performed more faithfully than the other conditions, followed by pitch and then by rhythm. We also hypothesize that the initial recall will be performed more accurately than the delayed recall. This is inclusive of all participants among all phrases (whether stable or non-stable) and all conditions (pitch, rhythm and melody); and we hypothesize that the musicians will perform better than the non-musicians, which is also inclusive of all participants among all phrases (whether stable or non-stable) and all conditions (pitch, rhythm and melody). With these four hypotheses in mind, we acknowledge that individual
participants will have different strengths within this also, outside of them being musicians or non-musicians. To assess these results efficiently and objectively, we will be measuring faithful recall by using Structure Measure Algorithm Matching (Meredith, 2006) which compares the pitch and the onset of the transcribed recalls to the original queries in order to generate a value between zero and one based on accuracy for each recall, and then finding the average recall score within all parameters, and testing our hypotheses this way.
4. Procedure
For the procedure of the experiment, each participant was tested on their own by the author, and every procedure of the experiment was performed in a quiet room booth with adequate privacy and reasonable acoustics. This entailed that it was run either in the home of the participant, in a relatively small room, or in a small piano room. Each participant had a notion that the experiment was to be on recalls sung and tapped, because while asking around for volunteers, this was what was explained first as it would have been understandable that some may be uncomfortable with this idea of being recorded singing (or even tapping). For each experiment, the melodies were loaded into Sibelius under an oboe sound font for the pitch and the melody conditions, and a timpani sound font for the rhythm condition. Before the commencing of the experiment, the participant was read a set of instructions that detailed what they were about to do, who can participate, the safety checks and information on confidentiality (see appendix).
Despite what the author told them, the melodies were of course not totally arbitrary – it was part of the initial design that the researcher ensured that the melodies get tested the same amount of times. However, the melodies were not selected with a particular participant in mind, and each test was to be slightly different.
The participants were each also given details on whom to contact if they have any queries or complaints, and were given an informed consent form to read over and sign. Following the reading of the instruction manual, and when the participant has
officially consented by signing the ethics document, there followed two sound checks. The first of which is to see if the participant will be able to hear the melodic
sequences clearly from the software Sibelius on all of the conditions. The second sound check was in order to ensure that the recording system on Praat would be able to pick up the sung and tapped recalls in the given acoustics of the room. Also, the participant is to be sitting comfortably, and the author did their best to make sure the participant was relaxed, not thirsty and comfortable.
For the two tasks that involved singing, the participant was asked to sing back the highest and lowest note that is to come up on the sequence before it is played, in order to ensure that they will be comfortable singing it back. This was to ensure that a lack of vocal range did not prevent the experiment from going forth if they were alarmed by the condition. As with psychological experiments in general it is effective and ethical that the participant is to feel entirely comfortable with the proceedings, in order for them to perform the recall openly. They had each been informed that their recordings would be documented in a publication in the academic field but would also remain anonymous, and that they can back out of the experiment at any point. Each participant was asked if they had understood the instructions and if they understood what they were to do, and whether they had any questions about the experiment. The participant was then asked if they were ready to hear the first sequence, and then would ensue the experiment.
Once the experiment had started the participant sat across from the researcher after having signed the document, and informed the author that they had understood the instructions after asking questions about the procedure. The first stimulus/query was played for the participant - as stated, the singing (melodic and pitch) conditions were played through an oboe sound font, and the rhythmic conditions were played through a timpani sound font. If there was a problem, for instance: one such as the participant being alarmed or disturbed by the hearing or distracted for whatever reason, a
different new melody is generated so that the participant doesn’t hear the melody more than once, so as to keep the number of hearings of each phrase between
participants fair. While they have been told in the instruction manual that the stimulus recalls range from easy to difficult, was indicated about the predicted difficulty or ease in regards to the expected recall accuracy, or any expectation on how well they will recalled the melodies. Participants were free to ask questions during the
experiment and informed as such in the instruction manual and ethics document that they were, but certain questions for the purpose of the experiment could not be answered until after the experiment.
The recalls were recorded thus onto the software Praat. With each trial, the first recall occurred immediately after the first listening, as soon as the author motioned with her hand. The author attempted to react as neutrally as possible to the recall, for given that it is a psychological experiment, it is important that it does not feel like a performance which judges on the aesthetics of the recall. After this immediate recording of the impression, the participant was informed once more that they were to give a second impression of the recall that they had just sung, after a time. This was something that the author checked was made clear to the participant - that they were to be performing the second recall after the clip of film for the experiment.
They were shown the first 7-minute clip from “Cast Away” after the first recall (which had been chosen so as for a uniform distraction as it does not contain music, and also it contains a vaguely interesting story). During the viewing the participants reacted to the clip, as they would have a normal viewing of a film, while keeping the task in mind. After this viewing, the participant was asked if they were ready for the second recall of the same melody, one that was also recorded after the researcher signals for them to give the second recall with her hand. Only one failed to do so and was given an SIAM score of zero (which did not affect them in any way). The second recall recording of the first stimulus concluded the first task.
The next two tasks were similar to the first task, except with a different condition, as we mixed up the order of conditions as the film clips are always played in the same order. In the second and the third trials, participants were played a later clip from the same movie. After the three tasks (six recalls) are completed, the participants
complete the questionnaire that detailed their demographics, musical experience (yes/no), if yes – instrument and years of musical experience, anything else they’d like to add about their relationship with music such as genre preference, and/or music related hobbies, and also their reaction to the experiment, as well as deciphering their method of recalling the melodies by prompting them to explain, if they were able to do so. The demographic section also attempted to detail the cultural origins of each
participant but as the experiment is conducted in English, this was one thing every participant had in common – that they all spoke English (see appendix). The
demographics revealed that they were predominantly European. After the participant completed the questionnaire, they were thanked and informed that the experiment was now completed, and they were now able to ask those questions that they wanted to ask about the experiment that the author had not been able to answer before for fear of tarnishing the psychological aspect of the experiment. In total, each experiment lasted overall around 25 minutes on average (a fact which each participant was made aware of before embarking upon the experiment) and this was also of course dependent on whether the participant asked questions, however much they wanted to know. This experiment did not offer a reward for it is purely voluntary, as the participants were aware. The recordings were saved and transcribed by the researcher, who throughout each experiment; was not entirely in the know herself indeed which melodic phrases were deemed stable and which melodic phrases were deemed unstable by the annotations given by collection experts in the Meerten’s Tune Collections.
Given that sometimes the participant required a new melody, if they felt
uneasy/uncomfortable or unprepared with the first one they had heard, we generated a different stimulus/phrase under the same condition when this was the case and merely started the trial again. We attempted to test each melody under each condition an equal number of times. Unfortunately the results some melodies had to be eschewed prior to data analysis due to an poorly distributed equality of phrases (some phrases ended up being tested more than others), leading to unbalanced data results, which we eliminated before we commenced with the analysis. This was so we could calculate even amounts of recalls for fragments under each condition, and it would ultimately have been impossible therefore, for the computational analysis to follow through with uneven datasets. It was unfortunate therefore, that for the present study, that we collected some recall data for the six stable and the six unstable phrases, and had to eliminate a few due to an uneven collection for some, and therefore we embarked upon analysis with five stable and five unstable phrases from which to draw comparisons and conclusions, leaving the two uneven melodies out. In addition to this, some of the data used for analytic comparison is problematic also, for some were exposed to a melody twice under two different conditions, normally if they asked for
another melody after hearing the first one, which may have affected the memorization process slightly. After we ran the experiments, we had collected recalls from 27 participants overall, and we had tested fragments 81 times through them, but
ultimately analysed two sets of 60 fragments for analysis, which comprised two sets (recalls per participant) of 30 (three conditions for each of the ten melodies).
From the recordings on Praat, the transcriptions were made for the sake of comparative analysis against the original query by the author onto the software Sibelius, converted to xml files and then analyzed computationally with SIAM (Meredith, 2006). As part of the analysis also, we checked that the accuracy values generated by the algorithm agreed with musical intuitions also, which they did. From the SIAM, we used the values generated in order to rate the accuracies of the recalls and compare the results with the hypotheses.
5. Results
From hereon in we will be referring to the three conditions as pitch, rhythm and melody. With the exception of one participant, every participant was able to give two recalls of each melodic stimulus, one instant recall and one delayed. While in this case, a percentage of participants gave note-for-note perfect recalls on all of the recalls both with the first recall and with the delayed recall; a few more participants gave a perfect first recall and an imperfect delayed recall, some even showed improvement with the delayed recall. An aspect that was of great interest to the research was the methods that the participants used to recall the melodies, and most participants indicated that they repeated the melodies over and over again in their heads from their first recall. Two participants who both reported to be pitch perfect (and their results suggested as such) said the same thing: that they were both musicians and they pictured the manuscripts in their heads and remembered the manuscript. They both gave recalls that were numerically high on the scale and with identical initial and delayed recalls. A few people who were also musicians also reported that they allowed the melodies to drift in and out of their heads, seeming to indicate that they did not have to try particularly hard for the task. One report that was
revealing and a potential facet for further investigation was that the two participants who were active composers said that they did this and did not have to try too hard to recall the melodies, which they did both, perfectly. They both also said that they were used to keeping melodies in their heads, and it was often involuntary. Another person said that one of the melodies sounded like Beethoven’s 9th and that he just
remembered the difference between that and the presented melody. As anticipated, although not part of the hypothesis, a huge difference in accuracy of melody recalls was up to the individual. Some people were extremely accurate with all conditions and acquired perfect scores, including 1 non-musician (a linguistics student), whilst some, including musicians, found the tasks more difficult. The individuality aspect was the largest factor in how accurate the recalls were as we randomized the
melodies. Fortunately, most participants reported that they found the experiment to be enjoyable, and only one out of twenty seven participants reported that they felt in any way uncomfortable or nervous at any part of the experiment.
As previously stated in the present study, the manual transcriptions of the recalls were compared to the original stimuli using Structure Induction Algorithm Matching (Meredith, 2006) in order to generate an efficient and fair judgment analysis of each recall, which would give us an objective measure with which to compare the results. This algorithm compared the onset and the pitch of the recalls versus the original queries, and generated a figure that took in these comparisons.
Our main question being one of stability, we present the accuracy measures of the stable dataset versus the non-stable dataset of all of the recalls inclusive of all of the conditions against the original queries first.
We found that the averages of each score were in fact in divergence with the
hypothesis, as averaging the SIAM values for the stable condition gave us 0.732 and averaging the SIAM values for the non-stable scores gave us 0.825, where 0 is inaccurate and 1 is accurate.
Therefore, we performed a one-tailed, unpaired t-test that resulted in t=-2.15 and p=0.98. This means that we could not reject the null-hypothesis in favor of our prediction that the stable dataset was more memorable than the unstable dataset.
We also performed a two-tailed t-test as well and acquired a result of p value=0.03, which was significant, but evidently in the opposite direction than hypothesized. Figure 1 (below) represents the average stable data recall values and the average non-stable recall values.
The y-axis shows the accuracy of the recall, from 0 indicating no similarity between recall and stimulus, to 1 indicating perfect agreement.
We calculated the average values of the recall accuracies of the stable versus the non-stable phrases within each condition. For the non-stable phrases under the melody
condition the average SIAM value was 0.65, and the average SIAM value for the non-stable phrases under the melodic condition was 0.775.
The result of the one-tail unpaired t-test yielded t= -1.57 and p=0.94. This also went against our initial hypothesis of stable dataset under the melodic condition would be most effectively remembered. Our two-tailed t-test in this case yielded t=-1.57 and p=0.12
The differences between the recall accuracies of the stable pitch phrases versus the non-stable pitch phrases resulted in being, as with the melodic condition, not in favor of our hypothesis with an average SIAM value of 0.649 for the average values of the unstable pitch phrases and an average SIAM value of 0.808 for the stable pitch phrases. In mind of our hypothesis that the stable pitch phrases would be more easily
encoded than the non-stable pitch phrases, the result of the one-tailed unpaired t-test being non-significant in this case also yielding t = -1.45 and p= 0.92.
The two-tailed t-test yielded t = -1.97 and p-value = 0.06, which was also in the opposite direction to hypothesized.
The average score for the SIAM values of the stable phrases under the rhythm condition was 0.912, and the average for the SIAM values for the unstable phrases under the rhythm condition was 0.889.
In mind of the hypothesis that the stable rhythmic phrases would be better encoded among participants than the non-stable rhythmic phrases, the difference transpired to be non-significant when we ran a two-tailed t-test which resulted in t = 0.578 and p-value = 0.567, but in this case, not against our hypothesis, although the null
hypothesis could not be rejected in this case.
Our one-tailed unpaired t-test we achieved a result of t=0.578 and p =0.28.
Figure 2 (below) represents the average SIAM values of all of the stable and all of the non-stable recall values of each condition.
The y-axis shows the accuracy of the recall, from 0 indicating no similarity between recall and stimulus, to 1 indicating perfect agreement. Orange = melody, green = pitch and purple = rhythm. Faded margins = the dataset that was expected to be recalled less accurately.
Investigating further, we calculated that the average of the scores for the melodic recalls (of the stable and non-stable datasets) received an SIAM value 0.7125. The average of the scores calculated for pitches (of the stable and non-stable
melodies) was 0.713, and the average of the scores calculated for rhythms was 0.901. We then compared the SIAM values of all of the recall conditions to one another outside of the stable and the non-stable datasets, testing whether the melodies as compared to the pitches reflect the hypothesis, which was that the melodies would be recalled more accurately than the pitches; and both the melodies and the pitches would be recalled better than the rhythms values.
The recalls of melodies when compared to the recalls of the pitches after performing a two-tailed unpaired t-test yielded t = -0.01 and p = 0.99.
Running a one-tailed unpaired t-test resulted in t = -0.01 and p = 0.5.
The same case was true of the recalls of melodies as compared to the recalls of the rhythms, after performing a two-tailed paired t-test resulted in a p <0.001 but in the opposite direction to hypothesized.
When testing the recall accuracies of the pitch to rhythm in mind of the hypothesis which was that the pitch would be recalled more faithfully than the rhythm, our two-tailed unpaired t-test yielded t = -3.98 and p<0.001, which was highly significant but also in the opposite direction of our hypothesis, given the averages. Therefore we performed a one-tailed t-test which yielded t = -3.98 and p=0.99
Figure 3 (below) represents a visualization of the average recall accuracy figures of the melody, pitch and the rhythm respectively.
The y-axis shows the accuracy of the recall, from 0 indicating no similarity between recall and stimulus, to 1 indicating perfect agreement. Orange = melody, green = pitch and purple = rhythm.
We tested the initial recall against the delayed recall for the 3rd hypothesis, and found that the initial recall was not significantly more accurate than the delayed recall overall. The average score for the initial recall was 0.785 and the average score for the delayed recall was 0.766.
Performing a two-tailed paired test yielded t=1.07 and p=0.29, and the one-tailed t-test yielded t=1.07 and p=0.14, which was not significantly in favor of our hypothesis, and therefore the null hypothesis could not be rejected in this case.
Figure 4 (below) represents the accuracy average values of each recall irrespective of the stability, participants and conditions.
The y-axis shows the accuracy of the recall, from 0 indicating no similarity between recall and stimulus, to 1 indicating perfect agreement. Faded margins = the dataset that was expected to be recalled less accurately.
The average recall accuracy result for the melodic condition in the first recall was 0.709, and the average for the second recall was 0.716, a result that suffice to say, was unprecedented.
Performing a one-tailed t-test in this case yielded t=-0.85, p=0.53, and the two-tailed t-test produced t=0.85, p=0.93 which was not significant.
The average recall accuracy result for the pitch condition in the first recall was 0.732, and the average for the second recall was 0.693. The two-tailed unpaired t-test
produced t=0.46, p=0.65, which was non-significant in terms of favor of our hypothesis.
Also, our one-tailed unpaired t-test produced t=0.46, p=0.32, and the null hypothesis could not be rejected in this case.
The average recall accuracy result for the rhythm condition in the first recall was 0.914, and the average for the second recall was 0.887.
The two-tailed unpaired t-test resulted in t=0.67, p=0.51, and the
one-tailed unpaired t-test for rhythm produced t=0.67,and p=0.25. The null hypothesis could not be rejected in this case.
The y-axis shows the accuracy of the recall, from 0 indicating no similarity between recall and stimulus, to 1 indicating perfect agreement. Orange = melody, green = pitch
and purple = rhythm. Faded margins = the dataset that was expected to be recalled less accurately.
For our 4th hypothesis, which was that musicians would perform better than non-musicians, we calculated the overall average SIAM values of the musicians versus the non-musicians, and we found that the musicians did not perform significantly better than the non-musicians. This was with testing all of the conditions as one and not taking the stable versus non-stable elements into account. The average of the musicians was 0.793 and the average for the non-musicians was 0.741. This was an expected result in terms of the hypothesis, but once more, was not a significant one. Our two-tailed t-test in this case resulted in t=0.5829 and p=0.57 therefore, the null hypothesis could not be rejected in this case either.
With the one-tailed t-test we acquired t=1.1833 and p=0.12.
Figure 6 (below) is a representation of the average SIAM value for the non-musicians versus the musicians.
The y-axis shows the accuracy of the recall, from 0 indicating no similarity between recall and stimulus, to 1 indicating perfect agreement. Faded margins = the dataset that was expected to be recalled less accurately.
We then calculated the results of the individual conditions performed by the
see that the musicians achieved significantly better scores on at least one of the conditions.
The average score for the non-musicians in the case of the melodic phrases was 0.646, and the average score for the musicians in this case was 0.757. Our two-tailed t-test produced a result of t=1.36, p=0.18 which was not significant and therefore the null hypothesis could not be rejected in this case.
For this our one-tailed t-test yielded t=1.18, p=0.12.
The average score for the pitch condition for the non-musicians was 0.701, and the average score for the musicians was 0.729.
In this case, the two-tailed t-test yielded a result of t=0.58 and p=0.57.
The one-tailed t-test yielded t=1.36 and p=0.09. The null hypothesis could not be rejected in this case either.
The average score for the non-musicians for the rhythm condition was 0.855 and the average score for the non-musicians under this condition was 0.911.
The results for musicians versus non-musicians on the rhythm condition for the two-tailed t-test yielded t=0.59 and p=0.56.
The one-tailed t-test yielded t=0.59 and p=0.28, and the null hypothesis could not be rejected in this case.
Figure 7 shows a graph of the averages for the scores of each condition results by the musicians and the non-musicians respectively.
The y-axis shows the accuracy of the recall, from 0 indicating no similarity between recall and stimulus, to 1 indicating perfect agreement. Orange = melody, green = pitch and purple = rhythm. Faded margins = the dataset that was expected to be recalled less accurately.
The results acquired could be interpreted as unexpected to say the least, although largely against the hypotheses whilst showing insignificant tendencies towards them such as the results of the musician compared to the non-musicians, which showed a slight tendency for musicians to perform better in such tasks; and the results of the first recalls as compared to the second, which showed that people tended to recall better in the first recall overall, although non-significantly so. However, as we also saw, this was with the marginal exception of the melody condition.
There were significant differences in the recalls of the stable versus the non-stable datasets, whereas this was completely the opposite direction of the hypothesis, it is still notable and a point for discussion and further investigation. The differences for the conditions as well when comparing the rhythm to the melody and the pitch were the opposite of hypothesized, and the pitch as compared to the melody were in our favor but non-significant. However the rhythmic measure was in our favor in terms of the stable versus non-stable hypothesis, although not significantly so. We managed to show that musical training did not make an appreciable difference in terms of
recalling folk melodies.
In the next chapter we will present speculations, which will aim to offer some adequate deductions on the acquired results.
6. Discussion
As we saw, it transpired that the accuracies under the stability hypothesis yielded that the stable phrases were recalled less accurately than the stable phrases, which was in utter divergence of our central hypothesis. This result was calculated amongst all participants in both recalls within all conditions. Curiously enough, our two-tailed t-test having resulted in in a significant p-value, therefore it was certainly significant in some sense, merely in the sense that it was opposite direction to the original
hypothesis, saying that our presupposed stable phrases were in fact less stable than our presupposed non-stable phrases. This result was of course highly unprecedented, for it appeared that in this case for one reason or another, that the non-stable melodies were to these participants significantly ‘more stable’ that is to say – they encoded themselves significantly more effectively on participants’ minds than the stable phrases. This was curious as certainly they were assumed to be stable upon reasonable grounds, by merit of appearing unchanged in a collection of melodies beforehand. One possible explanation for this unforeseen result was that perhaps, for such short phrases, the phrases in the ‘unstable’ dataset were perhaps in this case more idiomatic or unusual – that is to say, ‘catchy’, or striking and therefore more memorable for most participants. We cannot of course rule out the notion that we picked an odd selection of fragments from the melodies to deem as stable or non-stable, or that the melodies were too short given that most people expressed more than a passive interest in music in order to volunteer in the first place, although it could be the case that the phrases presented were too stable, predictable or typical, to be remembered; or that all of the phrases – that is to say both datasets - were stable, by merit of being in a folk-tunes dataset in the first place, or that the participants themselves, being largely European, had heard them before or likely heard phrases similar which aided them to predict and consequently form accurate mental representations. We can guess as well that if this is the case, the lengths of the phrases were ill-fitting of what we wanted to test, or for the group of (apparently musical) participants, or that the non-stable phrases were indeed more memorable by being unusual; or that the phrases were presented in an unusual way. Either way, it transpired that in this case; the standalone
