The Computational Wealth of Linguistic Diversity

The more than 7,500 languages spoken today differ across many relevant dimensions: the meanings they encode, the social rules governing their use in interaction, and the patterns required to construct intelligible messages. In AI research, this diversity has largely been framed as a challenge—an obstacle to building systems capable of serving the full range of human linguistic communities.

Yet linguistic diversity also represents something else: a vast repository of computational solutions produced through thousands of generations of cultural evolution. Different languages require speakers to track and encode different kinds of information—such as evidential sources, spatial reference frames, numerical structure, or social relationships—thereby emphasizing distinct patterns of representation and inference.

In this presentation, I explore the idea that linguistic diversity constitutes a largely untapped computational resource for AI. I discuss how the computations required or emphasized across languages can lead to measurable differences in downstream reasoning tasks in both humans and large language models. Finally, I consider what the observable limits of linguistic diversity may reveal about the broader design space of possible reasoning systems.

Lunch will be provided.


Hardware/software prerequisites: None

Format: Presentation

Target audience: Students, researchers, faculty, staff

Research Computing | View More Events
Co-hosted with: GradFUTURES

Contact the organizers