Zoom link: https://mit.zoom.us/j/2711902511

***
Speaker: Jacob Vigly

Affliation: Computational Psycholinguistics Laboratory

Abstract: 
How do we understand a sentence as it unfolds in real time?  One important clue about the human language processing mechanism comes from a well-documented observation that the effort to process a word scales with its surprisal (the negative log probability, given context). In a Bayesian inference setting, this makes sense if a bigger belief-update requires more effort. However, while this general phenomenon is well documented, and has strong justifications, it also has two important potential shortcomings: (1) We still don’t have a clear answer to the question of how the relationship arises, algorithmically, and (2) precisely equating belief update size to surprisal requires some assumptions, which may not always be licensed.  In this talk, I’ll outline how an approach to measuring processing cost by divergence between belief-distributions might help address these shortcomings, and discuss some prior and ongoing work I have in this space.

***
Speaker: Hokyung Sung

Affiliation: JazLab and Fiete Lab

Title: Marr's blind spot: normativity as the key to explaining the mind

Abstract:
How does the mind/brain know what it ought to do? Last year at Cog Lunch, I presented an experiment to begin exploring this question. This time, I hope to elaborate on the conceptual foundations of this project. Why is normativity essential for understanding the mind, and how should we approach it? First, I introduce intrinsic intentionality (i.e. "aboutness") as both a defining feature and core mystery of the mind. I then present arguments for why any science of intentionality must be grounded in a science of normativity. Finally, I explore how we can scientifically study normativity. While approaches based on Marr's levels of analysis begins by identifying what a system ought to do at the computational level, it assumes these norms without accounting for how they are generated or upheld. I will argue that this is Marr’s blind spot —- the failure of existing normative approaches to explain the very nature of normativity and, by extension, the mind itself.