Magazine
Eine Gruppe an Personen steht draußen auf dem Campus der Princeton University. Im Hintergrund sind grüne Wiesen und Bäume zu erkennen.

Das Music Cognition Lab der Princeton University, u. a. mit Hannah Wilkie, Elizabeth Hellmuth Margulis und Itamar Jalon. © Foto: Music Cognition Lab, Princeton University

In Conversation: Voices on “On Water” – Hannah Wilkie

From the Lab to the Museum

Could you briefly describe your role in the Music Cognition Lab and in Sounds of Water?

I’m a third-year graduate student in the Music Cognition Lab at Princeton, working with Elizabeth (Lisa) Margulis. My research focuses on the imaginings and streams of thought that arise while listening to music, with a focus on live and group listening contexts. With Sounds of Water, we are taking our research on music-evoked imaginings into a public-facing environment. Through the Humboldt Labor, we receive a large number of responses from people we otherwise wouldn’t reach, and visitors get the opportunity to take part in the lab’s research.

How did you come to work on music-evoked imaginings?

In a way, I came to it via a detour. My master’s research and some of my undergraduate research were actually on music and room acoustics, and how acoustics influence music perception. But Lisa’s research had fascinated me since my undergraduate studies, so when I decided to go to graduate school, I applied to Princeton to work in the music cognition lab. The lab was already researching musical imaginings, and a study had been lined up in which audiences at a live classical concert were asked to describe what they imagined. That became my project.
The idea itself goes back to an observation of Lisa’s. In a class, she played musical excerpts to her students and asked them to write down what they were thinking. She noticed that they were saying the same things. That led to the idea that music can give people shared experiences, even experiences that seem entirely private.

How does Sounds of Water build on that work?

At its core, we are again asking people to describe the story they imagine while listening. But the museum gives us access to a potentially much larger number of participants. And the station is embedded in a very specific setting, an exhibition about water. We think a lot about shared space, and about what influences people’s responses. We have some hypotheses, but I don’t want to give too much away about the study design.

What is the most unexpected imagining you can share with the public?

Several participants referenced a video game in response to one of the excerpts, which was interesting to see. And some descriptions are very elaborate, almost cinematic. Music really takes people to very different places.

The Museum as a Laboratory

What can you learn in the museum that you couldn’t learn in a lab?

In a lab setting, you usually work with a limited university subject pool. At Princeton, that’s mostly undergraduates aged 18 to 22. In the Humboldt Labor, we reach a much broader audience: people from all over the world and of all ages. That’s amazing.
There is also something special about the museum itself that we still need to unpack. People move through the exhibition, see the same things and experience them in different ways. They bring that frame of mind with them when they listen to the music.

You visited the station in Berlin. What was it like to see it in use?

When you design an experiment on your laptop, you can’t anticipate everything that members of the public will do. Seeing that helped us make changes to make the station easier to use. And it’s lovely to watch people take your experiment seriously, take their time and give thoughtful responses. At the same time, it’s a little strange, a bit like spying on people.

What does your day-to-day work on the project look like?

The project is in a long-running phase of data collection that will continue for many months. Every week, I check on the data, download it, back it up and take note of what we’ve received and how many responses there are. I also read through the free-text responses to see whether anything unusual is happening and whether we need to make any adjustments. On top of that, we’re running some preliminary analyses. Alongside this, I’m involved in other projects and in teaching.

What challenges come with collecting data in a museum?

In the lab, we usually work with people who have signed up for a study and are paid or receive course credit, so they’re incentivized to give it their full attention. Moving to a general audience is a big shift, and working with the curators of the Humboldt Labor was really helpful, because they have much more expertise in that.
One challenge is presenting the study in the right way, with the right signage, so that it looks like something people want to do. Another is getting people to take part attentively. Visitors are there to have fun, and some just want to click through to see what it’s all about. So we try to keep them engaged. It was also important to make the interface accessible, and we made modifications to that end.
And people come with friends and family. They sit together, look at each other’s screens and talk. In a lab study, you’re isolated in a booth. Here we can’t control for that, so we need to be aware of it in our data. It’s a trade-off: we have less control, but we get so much data from so many different people. That’s a genuinely new kind of dataset.

Hannah Wilkie

The Experiment and First Results

Some visitors see a word cloud of previous participants’ associations while listening. What do you want to find out with this design?

I can’t say too much about that yet: it’s something we’re analyzing right now, so we’ll see what happens.

Your first analyses point to “semantic convergence.” How would you explain that to a general audience?

When people listen to music, they describe a story they imagine or a memory they experience. Semantic convergence means that the content of these descriptions becomes similar across people. If one person talks about rain and another also talks about rain, perhaps by the seaside, their descriptions are similar.
To analyze this, we use natural language processing. We feed people’s texts into a language model, which encodes them as high-dimensional numerical vectors, essentially mathematical representations of the texts. With these, we can calculate how similar responses are to each other. Two people talking about rain end up very close together in this space, while one person talking about rain and another talking about cooking end up far apart. That lets us track whether the music leads people to converge on similar imaginings, or whether their responses remain quite distinct.

Looking Ahead

What would you like visitors to take away when they leave the station?

Ideally, they become interested in the same things we’re fascinated by. That’s why all of us work on this, and if we can convey that fascination, that’s a win. I’d love for people to leave thinking about music and imagination, and about our research. And perhaps to think a little more about what they themselves experience when listening: “This song makes me think of rain,” or “This one reminds me of going to the beach with my family.” Something to think about the next time they put on Spotify.

Is there a piece of music, perhaps one about water, that triggers particularly vivid images for you personally?

I love Benjamin Britten’s ‘Four Sea Interludes’ from Peter Grimes. Having grown up in the UK, I feel it really captures the British seaside. When I listen to it, the images are very vivid.

Is there anything you’d like to add?

We’d love people to take part in the experiment. That’s the main message. Once the results have been analyzed and published, we’ll definitely share them.

Hannah Wilkie

is a third-year graduate student in the Music Cognition Lab at Princeton University, directed by Elizabeth Hellmuth Margulis. Her research focuses on music-evoked imaginings, with a specific emphasis on live music and group listening contexts.