Henrik Smith-Sivertsen and Jesper Steen Andersen – Royal Library of Copenhagen/University of Copenhagen
Then the science guys entered the room – on MIR and popular music studies
ABSTRACT: During the last years new approaches to the study of popular music have emerged. A whole new field labeled MIR (music information retrieval) has been established as recent technological developments have made it possible to analyze large datasets of music/related information in ways unthinkable ten years ago. Due to the technical skills needed, the scholars conducting such research are generally trained within computer scientific fields. Contrarily, the field of popular music studies (PMS) has traditionally been anchored within disciplines of humanities and social science. For a number of reasons the contact between the two traditions has been limited. In this paper we will present and discuss the results of different MIR-projects on music history and demonstrate that: the digitization of music and the analytical tools being developed on the one hand holds big potential for new approaches within PMS.the results presented until now on the other hand often clearly demonstrate limited musical/historical knowledge. both MIR and musicology can benefit from acknowledging each other’s core competencies and from transgressing differences in research cultures.
[JB note: I would travel a long way (and indeed, have done so) to hear Henrik talk. He is a true eccentric, an extraordinarily entertaining presenter, a deeply impassioned archivist, and a methodologically thorough researcher. I’ve not seen Jesper present yet, but have high hopes that he is an equally methodologically passionate Dane].
Henrik opens by playing a section of Bohemian Rhapsody, from which we the opening 30s. We then hear the entire song speeded up so that its entire duration is less than 30s, which Henrik uses to demonstrate some of the suspect methodologies of ‘scientific’ studies of popular music. Warming to his theme, he shows us a series of screaming mainstream media headlines along the lines of ‘scientists prove that popular music is <whatever>’. These studies are often data-driven. Henrik lists some of the major scholars in our field – Frith, Adorno, Mauch, Negus, Born, Fabbri etc – few of whom use data-driven analytical methods. Phillip Tagg is the only one who shows up in any data-driven metric searches. Henrik’s point seems to be that qualitative humanities research methods tend to make little impact on data-driven scientific research, and that the two disciplinary worlds have such different methodologies that collaborative research is difficult.
Jesper now talks about Music Information Retrieval, which is hard data attached to popular music, some of it algorithmic and applicable to large corpuses of work – this is implicitly broader than the ubiquitous/canonical ‘Bohemian Rhapsody’ approach that Henrik criticizes at the start. Mauch et al is cited, and its data findings are discussed. This is a controversial study (for me) because it does not define its terms. For example, Mauch et al report a decline in ‘dominant 7th chords’ over a couple of decades, but don’t say how these were identified and how they were used in the study. This throws up some musically pedantic questions for me (how many dom 7ths? In what context? As a I chord or a V chord? Throughout or occasional?).
For a second example, the research generates a ‘Danceability’ data output, which varies over the years in the corpus. But if we don’t know how they determined danceability, any changes in it over time are meaningless.
There were, as Jesper acknowledges, many constraints in Mauch’s epistemology, including issues of corpus, theoretical assumptions, and opaque digital methods [JB note – this last point was so irritating to me when Mauch et al published in 2015 that I felt I could never use it in future research – so I have deliberately ignored their findings because their categories are so unclear]. Jesper acknowledges the issue that irritated me so much – we have an input (audio file), a ‘black box’ (Mauch et al’s data processing) and an output (the ‘findings by decade’). But because, epistemologically, we don’t know what’s in the black box, the research is meaningless. [neither Jesper nor Henrik use that word, but I think they imply it].
But Jesper defends audio content analysis as an epistemological tool (and I agree with him). It has so much potential. It’s just a shame that Mauch et al had such a poorly explained methodology. [I wonder if they had access to a musicologist?]
Jesper ends with a deeper philosophical dive, citing Moretti’s (2000) idea of ‘distant reading’ as implied justification for the limitations of Mauch et al:
Distant reading: where distance, let me repeat it, is a condition of knowledge: it allows you to focus on units that are much smaller or much larger than the text: devices, themes, tropes—or genres and systems. And if, between the very small and the very large, the text itself disappears, well, it is one of those cases when one can justifiably say, Less is more. If we want to understand the system in its entirety, we must accept losing something. We always pay a price for theoretical knowledge: reality is infinitely rich; concepts are abstract, are poor. But it’s precisely this ‘poverty’ that makes it possible to handle them, and therefore to know. This is why less is actually more. (Moretti, 2000).