Building robust language technologies that serve multilingual communities requires a comprehensive understanding of people who speak more than one language. This talk examines how and why speakers code-switch between different languages in spontaneous, informal speech, drawing on evidence from several complementary lines of research. I discuss how code-switching carries distinct prosodic styles that differ from monolingual speech, with influences from speaker language background and proficiency. I also explore the affective and cognitive motivations behind code-switching, particularly studying how speakers use it to express empathy and to accomplish specific language production goals. Together, these studies build an integrated picture of code-switching as a dynamic, structured, and socially meaningful behavior. I conclude by outlining how these insights can inform the design of future multilingual speech generation systems that can produce code-switching in more natural, human-like ways.
Debasmita Bhattacharya is a PhD candidate in Computer Science at Columbia University, where she is advised by Dr. Julia Hirschberg. Her research interests span multilingual and dialectal speech, conversation dynamics, speaker traits, and computational psycholinguistics. Her current work focuses on understanding how and why speakers code-switch between languages in order to improve the generation of human-like multilingual speech. Her graduate work is currently supported by an NSF Small Grant (2024-2027), and was previously supported by an NSF EAGER Award (2023-2024).