The Information Bottleneck principle in Linguistics

Berkeley Statistics and Machine Learning Discussion Group

ML&Sci Forum

February 4, 2019
1:30pm to 2:30pm
1011 Evans Hall
Get Directions


Efficient coding and language evolution: the case of color naming
Abstract: Languages vary widely in the way they encode meanings into words, but at the same time there are also universal patterns in word meanings across languages. In this talk, I will present a principled information-theoretic account of cross-language semantic variation. Specifically, I will argue that languages efficiently compress meanings into words by optimizing the Information Bottleneck principle, and that languages evolve under pressure to remain near the theoretical limit of efficiency. In support of this proposal, I will show evidence from the domain of color naming. This is joint work with Charles Kemp, Terry Regier and Naftali Tishby. Paper:

Full details about this meeting will be posted here:

The Berkeley Statistics and Machine Learning Discussion Group meets weekly to discuss current applications across a wide variety of research domains and software methodologies. Register here to view, propose and vote for this group's upcoming discussion topics. All interested members of the UC Berkeley and LBL communities are welcome and encouraged to attend. Questions may be directed to François Lanusse.


Noga Zaslavsky

PhD candidate, Center for Brain Sciences, Hebrew University

Noga Zaslavsky is a PhD candidate at the Center for Brain Sciences at the Hebrew University (working with advisor Naftali Tishby), and a visiting graduate student at UC Berkeley (hosted by Terry Regier). Her main research interest is to understand high-level cognitive functions from first principles, building on ideas and methods from machine learning and information theory. In particular, she is interested in computational principles that can account for the ability to maintain efficient semantic representations for learning and communication in complex environments. She believes that such principles could advance our understanding of human cognition and guide the development of human-like artificial intelligence.