A Researcher’s Day: From Matcha to Machine Learning
Ann Clifton is a Senior Research Scientist in Spotify’s New York office, where she has worked for just over a year. Her lab focuses on language technologies, applying machine learning to natural language data. She shares her Fort Greene apartment with her husband and two rescue dogs, Birdie and Buttercup.
Slow Mornings, Steady Mornings
Clifton is not a morning person. Her husband walks the dogs while she wakes up slowly with a matcha or kombucha. By 8:30, she checks email and Slack for anything needing immediate attention. Her role sits on the Tech Research team within the wider Personalization organization, a job that draws on her background in computer science, natural language processing, and linguistics.
Mornings tend to be stacked with meetings. Afternoons, when she can concentrate without interruption, are reserved for developing and delivering actual research work. She has to be proactive about setting boundaries and carving out focused time, otherwise the day fragments and the core work never gets done.
The Podcast Project
For the past year, Clifton has worked on a major undertaking: publicly releasing a huge data set of podcasts with transcripts and organizing a public challenge around it. The effort runs in conjunction with the National Institute of Standards and Technology (NIST) and the Text Retrieval Conference (TREC).
Podcasts are a relatively new medium, so the goal is to engage with the broader research community and attract more people to working with podcast data, while also highlighting Spotify’s research. The project has required a large organizational effort: assembling the data, formulating the challenge tasks, and starting to build models. On a personal level, it has connected her with people across product, design, PR, and legal. The work has the potential to move podcast research forward, as well as Spotify’s research in the natural language space.
In the final weeks before release, Clifton’s schedule tightened considerably, with daily stand-ups and regular meetings with external collaborators. But that intensity is not the norm. Much of the work in research is self-directed and follows a different rhythm than an engineering or product team’s.
Working from Home, Collaborating Across Time Zones
Internally, the team is distributed across New York, Boston, and Stockholm. That distributed setup means remote and asynchronous ways of working were familiar long before the shift to working from home. Clifton says she genuinely enjoys it — at lunchtime she takes the dogs out for another run around the local park.
Beyond the Main Project
Alongside her research and the podcast project, Clifton is organizing the first-ever industry track at COLING’2020. The conference was originally scheduled for the summer but is now expected to take place in December. She sees the industry in a transition: historically top-tier research happened in academia, but that is no longer the case. Industry researchers are doing as much of the cutting-edge work defining what comes next. The track aims to explore that shift, along with industry issues like privacy, fairness, and algorithmic responsibility.
Logging Off
Clifton admits she is terrible at stopping work and closing her laptop at a sensible hour. She tries to wrap up at 6pm, though that discipline slipped during the push to release the podcast project. She credits Spotify for recognizing the importance of work/life balance and hopes to return to a normal routine soon. She is looking forward to having evenings back — her favorite ballet studio has started streaming classes online, and she plans to join after work.
