An Information-Theoretic Account of Musical Expectation and Memory Kat Agres (kathleen.agres@eecs.qmul.ac.uk) Centre for Digital Music and Cognitive Science Research Group School of Electronic Engineering & Computer Science Queen Mary, University of London London E1 4NS, United Kingdom Samer Abdallah (samer.abdallah@eecs.qmul.ac.uk) Centre for Digital Music School of Electronic Engineering & Computer Science Queen Mary, University of London London E1 4NS, United Kingdom Marcus Pearce (marcus.pearce@eecs.qmul.ac.uk) Cognitive Science Research Group, Centre for Digital Music and Centre for Research in Psychology School of Electronic Engineering & Computer Science Queen Mary, University of London London E1 4NS, United Kingdom This paper examines the process of learning novel music over time, with a focus on mental anticipatory processing and musical structure. By using carefully constructed tone sequences, we are able to test how the statistical structure of music, as measured using information theory, affects the expectedness of tones, as well as memory for specific exemplars, over a period of increasing exposure. Abstract When listening to music, we form implicit expectations about the forthcoming temporal sequence. Listeners acquire knowledge of music through processes such as statistical learning, but how do different types of statistical information affect listeners’ learning and memory? To investigate this, we conducted a behavioral study in which participants repeatedly heard tone sequences varying within a range of information- theoretic measures. Expectedness ratings of tones were collected during three listening sessions, and a recognition memory test was given after each session. This enabled us to examine how statistical information affects expectation and memory for tone sequences over a period of increasing exposure. We found significant correlations between listeners’ expectedness ratings and measures of information theory (IT), and although listeners demonstrated poor overall memory performance, the IT properties significantly impacted on musical memory. Generally, simple sequences yielded increasingly better memory performance. High-information sequences, for which making accurate predictions is difficult, resulted in consistently poor recognition memory. Keywords: Music cognition; information computational approach; predictive models. Information Theory and Music Information theory has contributed to fields as diverse as engineering and linguistics by describing and quantifying the information contained in a signal. This is especially useful for clarifying how the brain processes temporal signals; and indeed, information-theoretic measures such as entropy, a measure of uncertainty, have successfully described and predicted how the human brain anticipates forthcoming sensory input, such as music and language (e.g., Manning & Schutze, 1999; Abdallah & Plumbley, 2009). Within the domain of music, there has been a long- standing interest in anticipation and prediction, and statistical and probabilistic approaches to learning have been influential for decades (consider Krumhansl & Kessler, 1982; and Saffran, Johnson, Aslin, & Newport, 1999). Computational models such as IDyOM (Pearce, 2005) derive information-theoretic properties of music that accurately reflect and predict listeners’ expectations during music listening (Pearce & Wiggins, 2006; Pearce et al., While statistical and computational approaches have modeled human performance on a variety of music perception tasks, these approaches have not yet been extended to modeling the learning trajectory of listeners: we do not yet know how information-theoretic measures capture musical learning over increasing exposure to musical exemplars, and how much exposure is necessary to theory; Introduction Music is a fruitful domain for exploring the mechanisms responsible for learning structured temporal sequences, a type of learning that subserves a wide range of human behaviors. Research by Krumhansl (1990), Pearce & Wiggins (2006), Huron (2006), and others shows that listeners implicitly acquire knowledge about the statistical structure of music. But is this implicit learning influenced by the information contained in the musical signal and, if so, how? Using computational methods, the pitch structure of music can be manipulated systematically to help reveal the ways in which various information-theoretic properties of melody interact and influence human learning and memory.
[1]
Geraint A. Wiggins,et al.
EXPECTATION IN MELODY: THE INFLUENCE OF CONTEXT AND LEARNING
,
2006
.
[2]
Mark D. Plumbley,et al.
Information dynamics: patterns of expectation and surprise in the perception of music
,
2009,
Connect. Sci..
[3]
Carla L. Hudson Kam,et al.
Humans Rapidly Learn Grammatical Structure in a New Musical Scale.
,
2010,
Music perception.
[4]
Marcus T. Pearce,et al.
The construction and evaluation of statistical models of melodic structure in music perception and composition
,
2005
.
[5]
María Herrojo Ruiz,et al.
Unsupervised statistical learning underpins computational, behavioural, and neural manifestations of musical expectation
,
2010,
NeuroImage.
[6]
C. Krumhansl,et al.
Tracing the dynamic changes in perceived tonal organization in a spatial representation of musical keys.
,
1982,
Psychological review.
[7]
F. Bartlett,et al.
Remembering: A Study in Experimental and Social Psychology
,
1932
.
[8]
Elizabeth K. Johnson,et al.
Statistical learning of tone sequences by human infants and adults
,
1999,
Cognition.
[9]
Hinrich Schütze,et al.
Book Reviews: Foundations of Statistical Natural Language Processing
,
1999,
CL.
[10]
J. Bartlett,et al.
Memory for Melodies
,
2010
.
[11]
C. Stevens,et al.
Sweet Anticipation: Music and the Psychology of Expectation, by David Huron . Cambridge, Massachusetts: MIT Press, 2006
,
2007
.