IDyOT: A Computational Theory of Creativity as Everyday Reasoning from Learned Information - Computational Creativity Research: Towards Creative Machines - page 124

For the current work, we will be focusing on a level of abstraction very distant
from the neural substrate, looking at combinations of high-level abstract mathemat-
ical operations to build an explanatory model at that level. The intention, subse-
quently, is to proceed downwards, viewing the components of the model and their
interactions 1 as descriptive and developing explanatory models below, so as to lead
down, eventually reaching the neural level. This long-term view will not, however,
prevent us from proposing links direct to the neurophysiology where appropriate.
7.3 The Theoretical Lineage of IDyOT
The theory presented here draws on multiple backgrounds, in cognitive science (par-
ticularly the cognitive science of music), in information theory, and in consciousness
research. The methodology is drawn from computer science, and is outlined explic-
itly above to avoid misunderstandings between disciplines about the purpose and
approach of computational modelling research.
7.3.1 Expectation in Music and Other Cognitive Phenomena
Pearce andWiggins [ 39 ] argue for the importance of expectation inmusic, continuing
a tradition that began with Meyer [ 31 ] and includes the important work of Huron
[ 23 ]. Pearce's doctoral work [ 33 , 37 ] concerns the construction and validation of
a computational model of sequence processing, based on Markov modelling, and
building on ideas by Conklin [ 7 , 8 ] regarding multidimensional models, that admit
prediction of human musical expectation, based on observed likelihood. Musical
melody is an interesting subject of study in this context: it is readily separable from
other musical constructs (such as harmony) and thus subject with minimal damage
to reductionist science, and it is clearly present in the vast majority of the world's
musics, in excitingly varied forms. This variation encourages us to decompose the
phenomenon into its temporal component [ 16 ], which is universally observed, and
its musical content, which is (therefore) culture-specific. Pearce's model, known as
IDyOM for “Information Dynamics of Music”, when exposed to a representative
selection of Western tonal melody, predicts the musical expectations of Western
listeners very well, explaining 81% of the variance in empirical validation studies.
It remains the most successful such model to date [ 36 , 37 ].
Importantly for the current work, the IDyOMmodel, combined with a descriptive
rule, also predicts musical phrase segmentation, achieving an F 1 score of 0.61. This
result is favourably comparable with other computational models, all of which are
programmed by humans, not learning-based [ 36 ]. The same model, when presented
1 The interactions are particularly important: by considering them, we avoid the trap of naïvely
assuming Fodorian modularity [ 15 ].