Data, Annotation, and Meaning-Making: The Politics of Categorization in Annotating a Dataset of Faith-based Communal Violence

被引:1
作者
Rifat, Mohammad Rashidujjaman [1 ]
Safir, Abdullah Hasan [2 ]
Saha, Sourav [3 ]
Junaed, Jahedul Alam [3 ]
Saleki, Maryam [4 ]
Amin, Mohammad Ruhul [4 ]
Ahmed, Syed Ishtiaque [1 ]
机构
[1] Univ Toronto, Toronto, ON, Canada
[2] Univ Cambridge, Cambridge, England
[3] Shahjalal Univ Sci & Technol, Sylhet, Bangladesh
[4] Fordham Univ, New York, NY USA
来源
PROCEEDINGS OF THE 2024 ACM CONFERENCE ON FAIRNESS, ACCOUNTABILITY, AND TRANSPARENCY, ACM FACCT 2024 | 2024年
基金
加拿大自然科学与工程研究理事会; 加拿大创新基金会;
关键词
annotation; meaning-making; decolonizing knowledge practices; communal violence; Faith; Religion; and Spirituality; RELIGION; SYSTEMS;
D O I
10.1145/3630106.3659030
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Data annotation is a process of meaning-making and is inherently political. The literature on ethics in data-driven technologies explores these political aspects, primarily focusing on questions of bias and power. This paper argues that the politics of annotation often overemphasize secular and modern values and overlooks faith-based, religious, and spiritual aspects (FRS) in data annotation. This oversight particularly affects the postcolonial regions of the Global South, where FRS are intertwined with people's everyday experiences and ethics. We conducted a focus group discussion and contextual inquiries with six annotators who annotated a faith-related "violence" dataset from South Asian YouTube content. Our analysis reveals that FRS blindness in data annotation manifests through the politics of achieving objectivity and the "scientific" process of meaning-making. Due to these goals, which are predominantly shaped by Western values, FRS sensitivities are overlooked from the initial stages of data curation through annotation, ultimately leading to a context collapse within the annotation process. Finally, we advocate for the adaptation of FRS sensitivities into the annotation process and data infrastructure, particularly when the dataset clearly pertains to FRS, to promote greater cultural and contextual inclusivity in annotation practices.
引用
收藏
页码:2148 / 2156
页数:9
相关论文
共 73 条
[1]  
Alatas SF, 2000, AM STUD INT, V38, P80
[2]  
Alatas SyedFarid., 2003, CURR SOCIOL, V51, P599, DOI DOI 10.1177/00113921030516003
[3]  
Almanea D, 2022, LREC 2022: THIRTEEN INTERNATIONAL CONFERENCE ON LANGUAGE RESOURCES AND EVALUATION, P2282
[4]  
Alvares C., 2011, ECON POLIT WEEKLY, P72
[5]   A scalable machine learning approach for measuring violent and peaceful forms of political protest participation with social media data [J].
Anastasopoulos, Lefteris Jason ;
Williams, Jake Ryland .
PLOS ONE, 2019, 14 (03)
[6]   Special Issue on AI in HCI [J].
Antona, Margherita ;
Margetis, George ;
Ntoa, Stavroula ;
Degen, Helmut .
INTERNATIONAL JOURNAL OF HUMAN-COMPUTER INTERACTION, 2023, 39 (09) :1723-1726
[7]  
Bowker Geoffrey C, 1999, Sorting things out: Classification and its consequences, DOI DOI 10.7551/MITPRESS/6352.001.0001
[8]  
Brooke JohnHedley., 1991, SCI RELIG SOME HIST
[9]   Us and them: identifying cyber hate on Twitter across multiple protected characteristics [J].
Burnap, Pete ;
Williams, Matthew L. .
EPJ DATA SCIENCE, 2016, 5
[10]   Understanding the Relationship between Religion Online and Offline in a Networked Society [J].
Campbell, Heidi A. .
JOURNAL OF THE AMERICAN ACADEMY OF RELIGION, 2012, 80 (01) :64-93