Seminar: Molly Lewis
Research Scientist, Department of Psychology and Social & Decision Sciences
Seminar will be presented via Zoom
What are we learning from language? Cognitive and social biases are encoded in the structure of natural language
Natural language provides speakers with information about the world through both explicit messages (e.g., "Mongolia is cold"), and through implicit messages present in the distributional statistics of words. One type of implicit message potentially present in distributional language statistics is information about gender stereotypes. In this talk, I'll present data examining the hypothesis that distributional statistics (e.g., the co-occurrence of “nurse” with “she”) play a causal role in the formation of the stereotype that men are more suited for paid work while women are more suited for family life. We use word embedding models to measure bias in the distributional statistics of 25 languages and find that languages with larger biases tend to have speakers with larger implicit biases (N = 657,335). These biases are further related to the extent that languages mark gender in their lexical forms (e.g., “waiter”/“waitress”) hinting that linguistic biases may be causally related to biases shown in people’s implicit judgments. Together, this work suggests that language implicitly contains rich information about the world, and that the cognitive system is sensitive to this information.
Molly Lewis is Research Scientist (Special Faculty) at Carnegie Mellon University in the Department of Psychology, and Social and Decision Sciences. She received her PhD from Stanford University in Developmental Psychology in 2016, and her BA in Linguistics from Reed College in 2009. She is a cognitive scientist studying language use and language learning, and is particularly interested in understanding the relationship between cognitive processes and large scale linguistic regularities. Her research uses a combination of behavioral experiments and computational social science