Fiddling with Computers – Tim Tangherlini

One of my greatest frustrations when I first started working in folklore was the difficulty I had finding things. I’d read a story or, as was more often the case, hear a snippet of one, and think, “oh, I’ve heard that before.” And then I’d spend an inordinate amount of time looking for something that I almost never found.


The majority of my work has focused on story telling – legends, personal experience narratives and, to a lesser extent, fairy tales. The very large archival resources that exist for Scandinavia have been an excellent challenge for understanding the important relationship between people and places, and how people use their expressive repertoire to shape and understand the environment in which they live.  Of particular importance has been the large Evald Tang Kristensen (1843-1929) collection housed at the Danish Folklore archives. A recurrent and persistent challenge in my work has been how to relate the repertoires of individuals not only to that person’s own history and that of their local community, but also how to relate their stories and story parts to those of other storytellers. In short, the challenge is one of understanding the dynamics and contours of a tradition situated in time and place.

Over the past decade, I have begun refining what I have described elsewhere as the “folklore macroscope” which relies on digitization and computational techniques for working with an entire collection at once. The folklore macroscope relies on multi-scale analysis to model the complexity of the dynamic folkloric process, operationalizing the underlying observation that folklore, which emerges from the productive tension between the individual and tradition, can be understood as the dynamic flow of culturally expressive forms on and across social networks. Katy Börner (2011), writing in the Communications of the ACM, characterizes microscopes as:

“provid[ing] a ‘vision of the whole,’ helping us ‘synthesize’ the related elements and detect patterns, trends, and outliers while granting access to myriad details. Rather than make things larger or smaller, macroscopes let us observe what is at once too great, slow, or complex for the human eye and mind to notice and comprehend.”

Tang Kristensen, the Danish folklore collector with whose collection I work, was an intriguing character whose first love was music. Not surprisingly, many of his early collecting trips focused on ballads, and one of his lifelong irritations was the chilly reception his transcriptions of melodies received from the music establishment in Copenhagen. Unfortunately, as with most 19th century collections, there are very few field recordings of actual performances. Nevertheless as I worked through the collection, I stumbled on enough hints—and enough storytellers who also happened to be fiddlers—to realize that this musical dimension of the collection was a rich area for future study. Yet, I also knew that working with audio as part of a computational model is quite challenging.

When I was invited by Prof. Ian Russell to participate in the NAFCo workshop in April 2017, I figured this was my chance to get my toes wet with Nordic fiddling traditions. I was curious whether, given a collection of fiddle music that I knew little about (actually, a pretty low bar since I would know little to nothing about any collection of fiddle music), I could find various points of similarity across tunes. Finding those similarities could help me with the underlying challenge of discovering “things like this” given a group of performances that interested me.

A great deal of music search is based on meta-data—largely words or texts that help describe the piece; this is an obvious and often incredibly powerful approach and drives many music search engines. While this usually works well when the data is “well labeled”, it often fails when there is no or little meta-data, or if the existing meta-data is misleading. Since I intentionally chose works for which I had little meta-data, I wondered if I could use automatically discovered features to create a profile of a tune and then use that profile to discover other tunes that had similar profiles in the corpus.

Fortunately, the computational analysis of music is developing at a rapid pace, and there is a great deal of commercial interest in devising recommender systems that share many of the features of what we are interested in when we search for similarity across a corpus. Consequently, it seemed that running an experiment for the NAFCo workshop on a corpus of Nordic fiddle tunes would teach me a lot about the possibilities and pitfalls, yet wouldn’t be so difficult as to be intractable.

To run my experiment, I found a stable and relatively easy to use music audio analysis package, PyAudioAnalysis, that allowed me to select features, such as beats per minute, spectral spread, or features related to pitch, as well as features calculated in a moving window across each tune. Using this package allowed me to fairly rapidly generate an overall profile of each tune in a corpus of 500 or so Nordic fiddle tunes. From there, it was straight forward to generate a visualisation that facilitated the exploration similarities between tunes across the corpus. In this first visualisation, the artists are grouped by colour, and the individual tunes are linked with lines to other tunes that share similarities across a feature set

See first visualisation by clicking here

Once I found a group of interesting songs or artists, I selected those, and reran the proves to generate a more focused visualisation.

See second visualisation by clicking here

In this example, I could fairly quickly identify two songs that had intriguing similarities even though they came from two different fiddler repertoires. Interestingly, none of the very sparse meta-data suggested that these two tunes would be any more similar than any two other arbitrarily selected tunes in the corpus.

The first tune in this simple illustration is aPolska efter Jöns Persson” from Northern Sweden, performed by Anders Hällström and Agneta Wiberg-Hällström in 2004, while the second is a “Slängpolska efter Blomgren” from Southern Sweden, performed by Edward Anderzon and Elin Skoglund in 2005.

Having no meaningful musical training, I turned to my local experts. Georgia Broughton, a PhD ethnomusicology student at UCLA, told me that, while no one would ever confuse these tunes, they do share certain interesting features, noting that:

“they are the same type of tune, they are both in minor keys, and the melodic line starts low then ascends in a bell curve before returning to the root of the scale at the end of the last phrase.”

Importantly—and something that would confound many other ways of searching—the tunes are in different keys and they have differences in the melody, one with (interval) jumps as opposed to the step-wise motion of the other.

This very brief experiment raises some interesting possibilities that could allow us to rapidly integrate sound recordings into the macroscopic modeling of folklore collections, thereby bringing us a step closer to the promise that the folklore macroscope holds for generating thick description at large scale.

While it might be tempting to quote U2’s “I still haven’t found what I’m looking for” at the end of this piece, I do feel that these computational methods hold a great deal of promise. Now I can say, “I’m looking for things like this” and, if I have a general sense of what “like this” means, I can much more easily find things that will interest me.


I’d like to thank UCLA graduate student, Georgia Broughton, and my longtime collaborator, UCLA digital librarian (and musicologist) Peter M. Broadwell for helpful suggestions and stimulating conversations as I dipped my toes into trying to work on questions of classification and search across Nordic fiddle music corpora. I’d also like to thank all of the participants of the NAFCo workshop whose interest, good humor, amazing music and friendly conversation make me want to spend more time learning about what in Danish is called “spillemandsmusik”.


Timothy R. Tangherlini teaches folklore, literature and cultural studies at the University of California, where he is a professor in Scandinavian Section, and the Department of Asian Languages and Cultures. He is also an affiliate of the Center for Medieval and Renaissance Studies, the Religious Studies Program, and a faculty member in the Center for Korean Studies and the Center for European and Eurasian Studies. Find out more here:


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s