[Colloquium] Jenna Cryan Dissertation Defense/June 30, 2023

[Megan Woodward]

This is an announcement of Jennifer Cryan's Dissertation Defense.
===============================================
Candidate: Jennifer Cryan

Date: Friday, June 30, 2023

Time: 12 pm CDT

Remote Location:  https://uchicago.zoom.us/j/96960878286?pwd=ZHN2NXFRWVduQlB2SVpMSkhEb3krZz09<https://urldefense.com/v3/__https://uchicago.zoom.us/j/96960878286?pwd=ZHN2NXFRWVduQlB2SVpMSkhEb3krZz09__;!!BpyFHLRN4TMTrA!6aYJVkAf1eCoBrqhk7vcIgZ6RIfq6ntR1Eaf1VqxxD0n39NZQwtnVoZMJFXMRzczL524mNIkAhfi6vatbY9GalUKz_kSU3F5$> (Meeting ID: 969 6087 8286, Passcode: 669048)

Location: JCL 298

Title: Measuring Perceptions and Mitigating Bias in Text and Voice

Abstract: Our ability as humans to effectively communicate depends heavily on the language we use and the way we speak to one another. The values of our society are both reflected in and reinforced by our use of language. Detecting how language could reflect biases needs to remain effective as these values evolve over time. This dissertation evaluates methods to measure how people perceive written text and spoken voice, how these perceptions may perpetuate societal stereotypes, and ways to prevent biased perceptions from being formed.
Specifically, gendered language in text often affirms gender stereotypes and perpetuates bias and discrimination.
As readers absorb written content, gendered language used settings such as biographies, recommendation letters, and job advertisements can negatively impact the subjects. Gender stereotypes have been studied extensively, however, the current methods used today still rely on word banks from nearly 50 years ago. Since then, societal views have continued to evolve and it's important to be able to reflect these changes. Additionally, significant advances have been made in developing new methods for analyzing how words are used in larger bodies of text. To address this, I first examine how descriptive language reflects societal perceptions of gender roles. Then, I demonstrate a crowd-sourced method for updating gender lexicons to reflect modern language and train deep learning models to detect gendered language more efficiently.
In addition to written text, efficient and unbiased communication depends upon not only the content, but the manner in which it is presented. The tone of voice of a speaker can heavily influence how they are perceived (e.g., perceived trustworthiness, competence). Further, changes in emotional tone of voice can reduce biases and allow for more effective communication.
This work seeks to understand: can ML-based tools correctly detect human emotion and manipulate human emotion in speech; how do these alterations impact the perceptions of the speaker by human listeners; and how willing are users to accept ML-based voice alteration software as tools for reducing voice bias.

Advisors: Ben Zhao and Heather Zheng

Committee Members: Ben Zhao, Marshini Chetty, and Heather Zheng
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20230620/347b10e5/attachment.html>