High-level feature descriptors and corpus-based musicology: Techniques for modelling music cognition

In recent years large electronic collections of music in a symbolically-encoded form have been made available. They have enabled music researchers to develop and test precise empirical theories of music on large data sets. Both the availability of music data and the development of new empirical theories creates a new perspective for Systematic Musicology, which, as a discipline, often sets out to explain or describe music through the induction of empirical laws, regularities or statistical correlations in relation to music objects or music related behaviour (see e.g. Karbusicky, 1979; Karbusicky & Schneider, 1980; Schneider, 1993; Huron, 1999; Parncutt, 2007). We present two methodological frameworks, feature-extraction and corpus-based musicology, which are the core approaches of a particular research project, M4S, whose aim is to discover mechanisms of music cognition. These two frameworks are also very useful for many other empirical tasks in Systematic Musicology.