P R E S E N T A T I O N B Y K A T H A R I N A S U H R BIAS IN THE REAL WORLD

UNIVERSITY OF Ethi cs and NLP Semi nar by Sebasti an Pado DISCUSSION QUESTIONS

#1 WHERE DOES THIS KIND OF BIAS COME FROM?

#2 IS THIS STUDY TRANSFERABLE TO

BIAS IN THE REAL WORLD | 2 MOTIVATION Increasing number of videos involving officers use of force with black suspects renewed conversations about modern-day race relations law enforcement agencies adopt body-corn cameras to increase accountability and transparency no systematic analysis of this massive amount of footage by the agencies common, everyday interactions not examined person's experience of respect or disrespect in interaction with police officers play a central role in judgment how fair the police as a institution is Blacks report more negative experiences in their interactions than other groups BIAS IN THE REAL WORLD | 3 RESEARCH QUESTION

DO OFFICERS TREAT WHITE COMMUNITY MEMBERS WITH A GREATER DEGREE OF RESPECT THAN THEY AFFORD TO BLACKS?

BIAS IN THE REAL WORLD | 4 APPROACH Analyzing officers language during vehicle stops of white and black community members Officers can communicate respect and understanding or contempt and disregard The language of those in positions of institutional power has greater influence over the course of the interaction than the language used by those with less power Quantitative lens on one key aspect of the quality or tone of police-community interactions offers new oppurtunity for advancing police training

BIAS IN THE REAL WORLD | 5 DATA TRANSCRIBED BODY CAMERA FOOTAGE FROM VEHICLE STOPS OF WHITE AND BLACK COMMUNITY MEMBERS Oakland Police Department April 2014 981 stops were examined 183h of footage 36.738 utterances Processed with Stanford CoreNLP to generate sentence and word segmentation, part-of-speech-tags and dependency parses

BIAS IN THE REAL WORLD | 6 STUDY STRUCTURE

STUDY 1 STUDY 2 STUDY 3 Human perceptions of Building statistical Racial Disparities in Officer Treatment models to predict Respect from Language aspects of respect Model from Study II was randomly selected CL model based on applied to all 36.738 utterances linguistic theories of utterances to predict a four point scale respect score for Respect and Respect and Formality extracting features with Formality ratings CL methods predict rating from Study I model-assigned ratings agree as well as humans agree with each other BIAS IN THE REAL WORLD | 7 RESULTS Respect higher: white or older community members, when a citation was issued White drivers: 57% more likely to hear the most respectful utterances. Black drivers: 61% more likely to hear the least respectful utterances. Did not contribute: officer race, severity of the crime, specific officers, location Formality higher: older and female community members, Location with higher crime rates Respect increased and Formality reduced as interaction progressed. For white drivers increase

faster than for black drivers BIAS IN THE REAL WORLD | 8 CONCLUSION body camera footage is a rich and important source of data racial disparities in officers respect are clear and consistent polices officers interpersonal interaction with blacks is more tense limitations: only transcribed text, no speech intonation, facial expression etc. opportunity: computational, large-scale analysis of language offers a new way to examine and improve police-community interaction

BIAS IN THE REAL WORLD | 9 DISCUSSION QUESTIONS

#1 WHERE DOES THIS KIND OF BIAS COME FROM?

#2 IS THIS STUDY TRANSFERABLE TO GERMANY

BIAS IN THE REAL WORLD | 10 LITERATURE & RESOURCES

Rob Voigt et al.: Language from police body camera footage shows racial disparities in officer respect. https://www.pnas.org/content/114/25/6521 [Retrieved: May 8th 2020]

Sharon LaFraniere and Andrew W. Lehren: The Disproportionate Risk of Driving While Black: An examination of traffic stops and arrests in Greensboro, N.C., uncovered wide racial differences in measure after measure of police conduct. https://www.nytimes.com/2015/10/25/us/racial-disparity-traffic-stops-driving- black.html [Retrieved: May 8th 2020]

Photos Page 8: Travis Dove for The New York Times All others: unsplash.com