Bias And Ethical Risks In Automated Toxicity Detection Of Bodo Language Using Deep Learning Approaches
DOI:
https://doi.org/10.29284/18nvga04Keywords:
Toxicity detection, Less resource Bodo language, bias issues, deep learning, Natural Language Processing.Abstract
The exponential growth of the digital communication on social media platforms has increased the spreading of the toxic language including hate speech, abusive expressions and cyber-harassment. Automated toxicity detection using Natural Language Processing (NLP) and deep learning has been a great resource to resolve the issues of moderating online content at remarkable standard. However, most available toxicity detection systems are created for high resource languages, especially English, and fail to take linguistic diversity, cultural nuance and ethical conformity in mind. This limit is especially important for the underprivileged indigenous languages like Bodo which is majorly spoken in Bodoland Territorial Council, Assam, North Eastern region of India and some parts of the neighbouring country like Bangladesh and Nepal. This study explores biasses and ethical risks in automated toxicity detection of the Bodo text by the Deep learning approaches. A collection of 12,037 text samples in the Bodo language that were collected from social media platforms, Blogs, Bodo drama books and short stories that were manually annotated regarding toxic and non-toxic text categories by native speakers. We considered more than one deep learning architecture for e.g. Vanilla Recurrent Neural Network and its variants. Performance evaluation was measured using different metrics like accuracy, precision, recall, F1-score and bias-specific measures like false positive rate for identity related terms. While it has seen that Bi-Recurrent Neural Network managed to achieve superior performance (F1-score reached 0.975845), it is found that significant bias against multicultural identity references, culturally embedded metaphors and emotionally expressed language common in Bodo discourse were identified. These results do raise serious ethical questions about unfair security review, cultural marginalization and the hostility of annotator and algorithmic prejudices. By comparing the results with past research studies on identifying the presence of toxicity in other languages, in this paper we highlighted the urgent need for culturally-aware, fairness based, and ethically motivated toxicity detection approaches for low-resource language.
References
[1] S. Das, A. Dutta, K. Roy, A. Mondal, and A. Mukhopadhyay, “A Survey on Automatic Online Hate Speech Detection in Low-Resource Languages,” 28 Nov 2024. https://arxiv.org/html/2411.19017v1
[2] Nihar Sahoo, Himanshu Gupta, Pushpak Bhattacharyya, 2022, “Detecting Unintended Social Bias in Toxic Language Datasets”, Proceedings of the 26th Conference on Computational Natural Language Learning (CoNLL) , pages 132 – 143, December 7- 8, 2022 ©2022 Association for Computational Linguistics, DOI: https://doi.org/10.18653/v1/2022.conll-1.10
[3] Abhishek Velankar, Hrushikesh Patil, Amol Gore, Shubham Salunke and Raviraj Joshi, “Hate and Offensive Speech Detection in Hindi and Marathi,” CEUR Workshop Proceedings 2021, ISSN 1613-0073, Vol-3159:239-247, https://ceur-ws.org/Vol-3159/, https://doi.org/10.48550/arXiv.2110.12200
[4] H. Naveed et al., “A Comprehensive Overview of Large Language Models,” arXiv preprint arXiv:2307.06435v10, Oct. 2024. https://doi.org/10.48550/arXiv.2307.06435
[5] A. Kumar and P. Kumar, “Investigating Bias in Automatic Toxic Comment Detection: An Empirical Study,” arXiv preprint arXiv:2108.06487, 2021. https://doi.org/10.48550/arXiv.2108.06487
[6] S. Das et al., “A Survey on Automatic Online Hate Speech Detection in Low-Resource Languages,” arXiv preprint arXiv:2411.19017, 2024. https://arxiv.org/html/2411.19017v1
[7] S. Caton and C. Haas, “Fairness in Machine Learning: A Survey,” ACM Computing Surveys, vol. 56, no. 7, Article 166, Jul. 2024. https://doi.org/10.1145/3616865
[8] S. Ghosh, S. Lepcha, S. Sakshi, R. Shah, and U. Srinivasan, “DeToxy: A Large-Scale Multimodal Dataset for Toxicity Classification in Spoken Utterances,” in Proc. Interspeech, 2022, pp. 5185–5189. https://doi.org/10.21437/Interspeech.2022-10752
[9] T. Hartvigsen, S. Gabriel, H. Palangi, M. Sap, D. Ray, and E. Kamar, “ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detection,” arXiv preprint arXiv:2203.09509, 2022. https://doi.org/10.48550/arXiv.2203.09509
[10] F. Elsafoury, S. Katsigiannis, and N. Ramzan, “On Bias and Fairness in NLP: How to Have a Fairer Text Classification?” arXiv preprint arXiv:2305.12829, 2023. https://doi.org/10.48550/arXiv.2305.12829
[11] M. Sap et al., “Annotators with Attitudes: How Annotator Beliefs and Identities Bias Toxic Language Detection,” arXiv preprint arXiv:2111.07997, 2022. https://doi.org/10.48550/arXiv.2111.07997
[12] B. D. W. O. J. 4th, R. Nieva, S. S. Lee, and N. Elhadad, “A Scoping Review of Ethics Considerations in Clinical Natural Language Processing,” JAMIA Open, vol. 5, no. 2, ooac039, 2022. https://doi.org/10.1093/jamiaopen/ooac039
[13] Carter, Emily & Narayanan, Ravi & Samuel, Adebis. “Ethical Challenges and Bias in NLP Models: A Python-Based Investigation”. (May 2025). https://www.researchgate.net/publication/392159070_Ethical_Challenges_and_Bias_in_NLP_Models_A_Python-Based_Investigation
[14] N. Kordzadeh and M. Ghasemaghaei, “Algorithmic Bias: Review, Synthesis, and Future Research Directions,” European Journal of Information Systems, vol. 31, no. 3, pp. 388–409, 2022. https://doi.org/10.1080/0960085X.2021.1927212
[15] Atijit Anuchitanukul, Julia Ive, and Lucia Specia. 2022. “Revisiting Contextual Toxicity Detection in Conversations.” J. Data and Information Quality 15, 1, Article 6 (March 2023), 22 pages. https://doi.org/10.1145/3561390
https://dl.acm.org/doi/10.1145/3561390
[16] M. Warner, A. Strohmayer, M. Higgs, and L. Coventry, “A Critical Reflection on the Use of Toxicity Detection Algorithms in Proactive Content Moderation Systems,” International Journal of Human-Computer Studies, vol. 198, 103468, Apr. 2025. https://doi.org/10.1016/j.ijhcs.2025.103468
(https://www.sciencedirect.com/science/article/pii/S1071581925000254)
[17] S. Chaudhary, “Advanced Language Toxicity Detection Using Deep Learning,” International Journal of Electrical and Electronics Engineering, vol. 14, no. 1–2, pp. 18–21, Apr. 2025.
[18] A. Abbasi, A. R. Javed, and F. Iqbal et al., “Deep Learning for Religious and Continent-Based Toxic Content Detection and Classification,” Scientific Reports, vol. 12, 17478, 2022. https://doi.org/10.1038/s41598-022-22523-3
[19] M. Shahid et al., “Leveraging Deep Learning for Toxic Comment Detection in Cursive Languages,” PeerJ Computer Science, vol. 10, e2486, Dec. 2024. https://doi.org/10.7717/peerj-cs.2486
Downloads
Published
Issue
Section
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
This work is licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
