{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_52679"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_52679","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"INTERDISCIPLINARY DECODING OF SOCIAL BIASES ACROSS DOMAINS","abstract":"Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2025","abstract_html":"Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2025","abstract_has_math":false,"creators":["Kim, Michelle"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Johnson, Kristen","Kordjamshidi, Parisa","Ghassemi, Mohammad","Roberson, Quinetta"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025","date_published":"2025","updated_at":"2026-11-18T18:59:26Z","subjects":["Computer science"],"languages":["eng"],"rights":["Attribution-NonCommercial 4.0 International"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:52679","local:Kim_grad.msu_0128D_21648"],"render_values":[{"text":"etd:52679","href":null,"code":true},{"text":"local:Kim_grad.msu_0128D_21648","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/pg40-5b33","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/pg40-5b33","record_label":"Repository record","record_kind":"doi","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://d.lib.msu.edu/oai?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Ad.lib.msu.edu%3Aetd_52679","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Johnson, Kristen","Kordjamshidi, Parisa","Ghassemi, Mohammad","Roberson, Quinetta"]},{"key":"dc:creator","label":"Author","values":["Kim, Michelle"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2025"]},{"key":"dc:relation","label":"Dc Relation","values":["Electronic Theses & Dissertations"]},{"key":"dc:type","label":"Dc Type","values":["text","Electronic dissertations"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Computer science"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["Attribution-NonCommercial 4.0 International"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:52679","local:Kim_grad.msu_0128D_21648","https://doi.org/doi:10.25335/pg40-5b33"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2025","Language models reflect social biases. Language models, particularly Large Language Models (LLMs), have made remarkable advancements in processing, understanding, and generating text that closely resembles human communication. This progress has led to their widespread integration into various systems, ranging from conversational AI like ChatGPT to healthcare assistant tools that influence our social interactions and decision-making. However, a significant concern arises from the fact that LLMs reflect the social biases present in the data on which they are trained. As language models can learn and amplify harmful social biases, their increasing use across various contexts raises the likelihood of these biases causing harm. Research has shown that language models can reproduce social biases in downstream tasks such as language generation and coreference resolution. When deployed in real-world applications, these models pose risks to marginalized individuals and groups, raising important ethical considerations regarding their integration into society. Addressing these biases is crucial to ensure that the use of language models does not reinforce existing inequalities.Recognizing the significance of developing fair language models, we investigate the presence and impact of social biases in language models. We adopt an interdisciplinary perspective that integrates computer science with insights from other fields, rather than only focusing on a technical viewpoint. Rather than viewing these biases as errors to be fixed, we propose leveraging established theories from other fields, such as social psychology. We develop theory-driven computational algorithms that can better comprehend and interpret these biases to improve our understanding of their implications and foster more responsible AI development.This dissertation presents a comprehensive multilevel analysis of social biases and stereotypes in Natural Language Processing (NLP). It is organized into three chapters, each building on the insights of the previous one to provide a holistic perspective on bias in NLP. Chapter 2 explores social biases across various NLP tasks by examining political bias in news media framing and critical healthcare bias in clinical diagnosis prediction. This exploration reveals the real-world implications of these biases and their broader societal impact. Chapter 3 shifts focus from identifying problems to developing solutions by introducing a generalizable stereotype detection model. This model is designed for versatility, making it applicable across different NLP tasks and adaptable to various dimensions of bias, such as race and gender. Finally, Chapter 4 examines the utility of the developed stereotype detection model across various languages and cultures. This chapter demonstrates the potential of a pan-cultural stereotype detection model and broadens the focus of bias research from being English-centric to include additional languages and cultures.In this dissertation, we demonstrate a strong commitment to understanding and addressing fairness in language models. It employs a multilevel analysis approach to investigate biases in NLP at various levels, from exploring the specific manifestations of biases in different tasks to developing a universal mechanism for detecting stereotypes across diverse linguistic and cultural contexts. Through this systematic approach, we aim to provide both a deep and broad understanding of bias and contribute to AI ethics by advocating for more equitable and culturally aware NLP systems.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf","1 online resource (121 p.) : ill."]},{"key":"dc:title","label":"Title","values":["INTERDISCIPLINARY DECODING OF SOCIAL BIASES ACROSS DOMAINS"]}]}],"canonical_facts":{"dc:contributor":["Johnson, Kristen","Kordjamshidi, Parisa","Ghassemi, Mohammad","Roberson, Quinetta"],"dc:creator":["Kim, Michelle"],"dc:date":["2025"],"dc:description":["Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2025","Language models reflect social biases. Language models, particularly Large Language Models (LLMs), have made remarkable advancements in processing, understanding, and generating text that closely resembles human communication. This progress has led to their widespread integration into various systems, ranging from conversational AI like ChatGPT to healthcare assistant tools that influence our social interactions and decision-making. However, a significant concern arises from the fact that LLMs reflect the social biases present in the data on which they are trained. As language models can learn and amplify harmful social biases, their increasing use across various contexts raises the likelihood of these biases causing harm. Research has shown that language models can reproduce social biases in downstream tasks such as language generation and coreference resolution. When deployed in real-world applications, these models pose risks to marginalized individuals and groups, raising important ethical considerations regarding their integration into society. Addressing these biases is crucial to ensure that the use of language models does not reinforce existing inequalities.Recognizing the significance of developing fair language models, we investigate the presence and impact of social biases in language models. We adopt an interdisciplinary perspective that integrates computer science with insights from other fields, rather than only focusing on a technical viewpoint. Rather than viewing these biases as errors to be fixed, we propose leveraging established theories from other fields, such as social psychology. We develop theory-driven computational algorithms that can better comprehend and interpret these biases to improve our understanding of their implications and foster more responsible AI development.This dissertation presents a comprehensive multilevel analysis of social biases and stereotypes in Natural Language Processing (NLP). It is organized into three chapters, each building on the insights of the previous one to provide a holistic perspective on bias in NLP. Chapter 2 explores social biases across various NLP tasks by examining political bias in news media framing and critical healthcare bias in clinical diagnosis prediction. This exploration reveals the real-world implications of these biases and their broader societal impact. Chapter 3 shifts focus from identifying problems to developing solutions by introducing a generalizable stereotype detection model. This model is designed for versatility, making it applicable across different NLP tasks and adaptable to various dimensions of bias, such as race and gender. Finally, Chapter 4 examines the utility of the developed stereotype detection model across various languages and cultures. This chapter demonstrates the potential of a pan-cultural stereotype detection model and broadens the focus of bias research from being English-centric to include additional languages and cultures.In this dissertation, we demonstrate a strong commitment to understanding and addressing fairness in language models. It employs a multilevel analysis approach to investigate biases in NLP at various levels, from exploring the specific manifestations of biases in different tasks to developing a universal mechanism for detecting stereotypes across diverse linguistic and cultural contexts. Through this systematic approach, we aim to provide both a deep and broad understanding of bias and contribute to AI ethics by advocating for more equitable and culturally aware NLP systems.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."],"dc:format":["application/pdf","1 online resource (121 p.) : ill."],"dc:identifier":["etd:52679","local:Kim_grad.msu_0128D_21648","https://doi.org/doi:10.25335/pg40-5b33"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["Attribution-NonCommercial 4.0 International"],"dc:subject":["Computer science"],"dc:title":["INTERDISCIPLINARY DECODING OF SOCIAL BIASES ACROSS DOMAINS"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:26Z"}