Elisa Kreiss    


Cognitive scientist
with a focus on Linguistics, Psychology and Computer Science

About


I'm a PhD student in the Linguistics department at Stanford University. I study language from a cognitive and computational perspective.

Concretely, my work aims to improve our understanding of how computational models of language relate to our human psycholinguistic reality (i.e., our capabilities and shortcomings when we produce and comprehend language).

I use different linguistic phenomena to investigate this relationship. In one series of projects, we look at human referring expression production and comprehension, and its prediction in a probabilistic model of communication, the Rational Speech Act framework. In another project, we investigate how we perceive guilt in news stories about arrested suspects and in what ways these human intuitions relate to predictions of deep neural networks.

At the moment, I mainly work with Christopher Potts (Stanford University), Judith Degen (Stanford University), and Michael Franke (Osnabrück University). I'm also a member of the interActive Language Processing Lab at Stanford and I'm an affiliated researcher of the Causality in Cognition Lab.

Publications


Pending

Degen, J., Hawkins, R.D., Graf, C., Kreiss, E., and Goodman, N.D. (in press). When redundancy is rational: A Bayesian approach to 'overinformative' referring expressions. [pdf][code]

2019

Kreiss, E., Franke, M., Degen, J. (2019). Uncertain evidence statements and guilt perception in iterative reproductions of crime stories. Proceedings of the 41st Annual Conference of the Cognitive Science Society. [pdf][code]

2017

Kreiss, E. (2017). Modeling Natural Language in the RSA Framework: Typicality Effects in Overinformative Referring Expressions. Unpublished Bachelor's Thesis.

Resources


The Annotated Iterated Narration Corpus (AINC)

In this corpus we have collected reproductions of news stories from participants who read carefully designed crime reports about a committed crime and the arrest of a suspect. The original stories were manipulated as to how strong the evidence seemed to be. The reproductions were then again read and reproduced by other participants. Afterwards those stories were annotated probing the reader's belief about the suspect's guilt, but also other measures, such as their emotional affectedness for each reproduction. For more details, have a look at our corpus documentation or consult Kreiss (2019) where we describe the data collection process and present some first interesting analyses.

Talks and Posters


2019

Kreiss, E., Degen, J. Production expectations modulate contrastive inferences. CAMP 2019, Santa Cruz, Oct 26-27.

Kreiss, E., Franke, M., Degen, J. Uncertain evidence statements and guilt perception in iterative reproductions of crime stories. CogSci 2019, Montreal, Jul 24-27.

2017

Kreiss, E., Degen, J., Hawkins, R.X.D., and Goodman, N.D. Mentioning atypical properties of objects is communicatively efficient. CogSci 2017, London, Jul 26-29.

Kreiss, E., Degen, J., Hawkins, R.X.D., and Goodman, N.D. Mentioning atypical properties of objects is communicatively efficient. XPRAG 2017, Cologne, Jun 21-23.