Program

Keynote

Noah Smith (University of Washington, Allen Institute for Artificial Intelligence):
Imagining a Music Language Model

Language models and their multimodal variants now present many exciting new opportunities for advancing music processing applications. As a researcher in natural language processing for over twenty-five years, and as a musician for even longer, I’ve recently started learning about music IR and related challenges with some of my students.  In this talk, I’ll offer some opinionated observations, technical ideas, and lessons learned from NLP that I believe could be useful for the music processing community.  These include matters of evaluation methodology, the roles of data and theory, the framing of problems, and guiding questions about who we are building technology for.  I’ll also reflect a bit on our efforts to improve the state of the art in open (multimodal) language models and speculate about a grand challenge: building a (hopefully open) language model for MIR.