When:
Wednesday, February 6, 2019
3:45 PM - 5:45 PM CT
Where: Frances Searle Building, 3-417, 2240 Campus Drive, Evanston, IL 60208 map it
Audience: Faculty/Staff - Student - Public - Post Docs/Docs - Graduate Students
Contact:
Irene Sakk
(847) 491-7020
Group: Linguistics Department
Category: Academic
https://nlp.stanford.edu/robvoigt/
Computing Social Meaning from Micro to Macro
Abstract
In this talk I describe two studies that illustrate how computational linguistics can be expanded from its traditional focus on text data and propositional meanings to address the diverse contexts, multiple modalities, and real-world interaction necessary to model social meaning.
In the first, we analyze police body camera footage to study racial disparities in policing. We implement pragmatic theories of politeness in a computational model of officer respect and apply it to a large dataset of traffic stops, finding that officers are less respectful to black than to white community members, even after controlling for contextual factors like location and social factors like officer race. The resulting model allows us to better understand how discourse-pragmatic aspects of respect operate in this unique domain.
In the second, we computationally extract head cant (side-to-side tilt of the head) in a large corpus of video data and use a quantitative-qualitative analysis of its relations with prosody and discourse particles to examine its meaning-making potentials across interactional contexts. We find that head cant serves a bracketing discourse function that can index a frame of shared understanding, and show how automatic annotation using techniques from computer vision opens new possibilities to scale up our understanding of gesture and embodiment in language.
My work shows how micro-scale interpretive meaning can be integrated with macro-scale aggregate and distributional variation, and demonstrates the opportunities for social impact afforded by this approach.