TY - BOOK

T1 - NLP Across Social Groups

AU - van Zee, Anna Katrine

PY - 2024

Y1 - 2024

N2 - The widespread adoption of Natural Language Processing (NLP) tools has become increasingly prevalent. From customer service, social media and news reporting to healthcare, education and virtual assistants. However, the uneven performance of these tools across different social groups raises significant concerns and amplifies the potential detrimental impact on society. This thesis delves into the investigation of this pertinent issue by systematically analyzing the performance of NLP tools across a spectrum of tasks, ranging from syntactic analysis to speech recognition, with a particular emphasis on their implications for various social groups.Our exploration extends to a comprehensive examination of the concept of fairness within the context of NLP. We scrutinize the intricate relationship between bias and fairness, and illustrate how these are indeed separate phenomena. A critical aspect of our inquiry involves considering the Rawlsian fairness principle and its prevalence in the current research field of NLP, and we propose a nuanced discussion on alternative definitions of fairness, emphasizing the necessity of evolving conceptual frameworks to address the pressing issue of increasing performance disparities within the field of NLP. By doing so, we aim to contribute to the ongoing conversation about inclusive and equitable NLP technologies and their application across diverse social groups.

AB - The widespread adoption of Natural Language Processing (NLP) tools has become increasingly prevalent. From customer service, social media and news reporting to healthcare, education and virtual assistants. However, the uneven performance of these tools across different social groups raises significant concerns and amplifies the potential detrimental impact on society. This thesis delves into the investigation of this pertinent issue by systematically analyzing the performance of NLP tools across a spectrum of tasks, ranging from syntactic analysis to speech recognition, with a particular emphasis on their implications for various social groups.Our exploration extends to a comprehensive examination of the concept of fairness within the context of NLP. We scrutinize the intricate relationship between bias and fairness, and illustrate how these are indeed separate phenomena. A critical aspect of our inquiry involves considering the Rawlsian fairness principle and its prevalence in the current research field of NLP, and we propose a nuanced discussion on alternative definitions of fairness, emphasizing the necessity of evolving conceptual frameworks to address the pressing issue of increasing performance disparities within the field of NLP. By doing so, we aim to contribute to the ongoing conversation about inclusive and equitable NLP technologies and their application across diverse social groups.

M3 - Ph.D. thesis

BT - NLP Across Social Groups

PB - Department of Computer Science, Faculty of Science, University of Copenhagen

ER -