Similarity: haven't we heard this before somewhere?

When interviewed by a panel of the Arts and Humanities Research Council as part of the process of securing the funding for Transforming Musicology, we were asked to explain how the project would be unified, rather than a loose collection of three or more separate research projects. Tim gave an excellent presentation which focused on linking themes, and especially on Linked Data, and one of our claims was that, fundamentally, all of the projects would be concerned with using technology to better investigate whether and how one musical thing was like another. Indeed we made a bold claim that much of musicology is concerned with similarity: two pieces of tonal music are like each other because both are elaborations of the same Ursatz; a work of uncertain authorship is demonstrated to be like a set of others and so probably by the same composer; a passage of guitar music is shown to be, in some sense, 'like' a hummingbird.

In two recent developments, the topic of music similarity has been an explicit focus for research, both with a connection to Transforming Musicology through shared personnel. In January a week-long research workshop on music similarity was held at the Lorentz Center in Leiden, the Netherlands, organised by Christina Anagnostopoulou, Elaine Chew, Elizabeth Margulis and Anja Volk. Fifty-five researchers participated, at different stages of career ranging from postgraduate students to established professors, among them five members of the Transforming Musicology team. The format was broadly talks by experts from various different perspectives on similarity plus discussions in small groups, and a final wrapping up by a team of the younger researchers starting to compile a 'roadmap' of music similarity, and an attempted wrapping up of the week from myself. You can see copies of presentations from most of the talks on the Lorentz Center website. A running joke of the week was the lack of a definition of 'music similarity', though some definitions were proposed in some talks. A special issue of Journal of New Music Research, is likely to arise from the workshop.

The second development is another AHRC-sponsored project, also falling under the Digital Transformations banner. The project, 'ASyMMuS: An Integrated Audio-Symbolic Model of Music Similarity', is a collaboration between City University London, University College London, and Lancaster University (where I am Co-I), and aims to use the tools developed in the Digital Music Lab project, recently completed at City, UCL, Queen Mary, and the British Library. There is now a substantial body of research on musical similarity based on audio data, and on 'symbolic', score-like, data, but the two do not correlate in a simple fashion. Audio models tend to give weight to timbre, for example, which does not feature in symbolic models at all. Even in the domain of pitch, audio models tend to effectively measure similarity in harmony whereas symbolic models concentrate on melody. By exploring models which integrate both kinds of data, we hope to achieve better models of musical similarity. This is a short-term project which runs to September 2015.

I have expressed doubt elsewhere that 'music similarity' is a unitary concept (see, for example, my article from 2012 in JNMR), and I have stressed that judgements of similarity only make sense in the context of a specific musical task (this was the main point of my contribution at the Lorentz workshop), but it is nevertheless clear that the concept is one with many musicological applications. I have never claimed that techniques for modelling similarity from one musical perspective cannot be useful when modelling from another perspective, and to me that seems like the most likely outcome of these projects: meta-theory about music similarity rather than models which capture every kind of similarity. So all power to these projects; no doubt we will be hearing more of this kind of thing in the future!

Alan Marsden is a co-investigator on Transforming Musicology.