A growing part of my practice involves translating datasets (numerical, textual, visual) into music. I should say upfront: I don’t believe any of this is objective. I’m not sure I ever did, but for a while I behaved as though it might be, and that gap between what I sensed and what I told myself is probably where this piece comes from. Every decision in the translation process carries aesthetic and cultural weight. Donna Haraway called this situated knowledge : the recognition that no observation exists outside the observer’s position. Sonification is no exception.
The translation problem
When I translate data into sound, I am not revealing some hidden music waiting in the numbers. I am making a series of choices that feel every bit as subjective as choosing which chord follows another in a piano piece. The difference is that sonification can wrap those choices in something that resembles scientific authority: a dataset was involved, a mapping was applied, therefore the result must be neutral. It isn’t, of course. But the pull toward that framing is real, and I’ve felt it.
Consider something as basic as mapping a declining dataset to pitch. The moment I reach for a descending minor scale, I’ve imported centuries of Western cultural conditioning into what was supposed to be a neutral reading. Leonard Meyer made this case decades ago in Emotion and Meaning in Music: our emotional responses to sound are not hardwired but learned. Minor tonalities carry sadness because we’ve been taught they do. Frantic rhythms signal urgency because our listening culture says so. Dissonance unsettles Western ears, but that reaction is a convention, not a law of physics.
The translation is always partial, always positioned, always an interpretation dressed up as a reading.
I think what I’m trying to say is that the phrase “letting the data speak” never quite sat right with me. The data doesn’t speak; I speak through it, choosing the voice, the accent, the register, and then (if I’m not careful) attributing the result to the dataset rather than to myself. Recognising that tendency didn’t resolve anything neatly. It just made the work more honest.
Raw material, not finished product
Once I accepted this (and it took longer than I’d like to admit), the question shifted. If objectivity is off the table, what is data actually good for in a compositional context?
My answer, for now: raw material. I use data to generate the initial substance of a piece, often through parallel streams and systematic translation matrices that connect data dimensions to musical parameters. This process acts as a creative agent in this phase, not merely a tool… But what comes out tends to be shapeless, sometimes genuinely dubious in its musical merit! That’s the whole point.
The stage I find most satisfying (and, I think, most honest) is what I call computational editing : taking that algorithmically generated material and chiselling it into a form that works as music. Not abandoning the data-driven foundation, but not pretending it’s sufficient either. This is where I reclaim my share of authorial ownership over the piece. The data provided the block; the final shape is mine.
It’s a way of working that sits between science and art in the traditional senses of both words: a form of situated interpretation that makes no claims to universal truth while insisting on rigorous method. Whether it holds up as a framework beyond my own practice, I genuinely don’t know.
I’m still working out where the boundaries are. The honest answer is that I probably always will be; the goal posts move as we evolve, and what felt like the right balance a year ago already looks different from where I stand now. Which, if anything, only reinforces the point: different composers will draw those lines in different places, at different times, for different reasons. And that’s just fine. The absence of a universal answer is not a problem to solve. It’s the condition that makes the work worth doing.