Reinforcement-Learned Ethical Transformers for Socially Responsible Text Generation in Sensitive Domains

Read full text Summarise with VIRAT
Authors: Afaf Abdullah Ali Alalawi, Yasmeen Anjeer Alshehhi, Bidoor Noori Ishaq, Maen Qaddoura
Cited by: 0
References: 46

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Neural language models have revolutionized text creation, benefiting chatbots, content production, and automated support systems. However, these models frequently provide biased, incorrect, or harmful content in sensitive domains including healthcare, legal services, and mental health support. In many cases, these models are trained without clear ethical guidelines. This raises ethical and security concerns concerning automated text-generation systems. To address these difficulties, this research introduces Ethical Neural Models (ENM), a novel framework for socially acceptable text generation in sensitive domains. The ENM neural language model has an ethical framework. This technique uses dual-objective training by combining ethical alignment with traditional language modeling losses. Following domain-specific norms and human-curated ethical criteria may help the model avoid prejudice and damage while preserving linguistic fluency and relevance. ENM was compared to controlled generation baselines and fine-tuning on healthcare and legal advisory datasets. ENM reduces biased or inappropriate outputs by over 35%, according to statistical and human evaluations. ENM remains competitive in contextual relevance and fluency. For sensitive applications, ENM proves that ethical and responsible text generation is possible and efficient. Ethical principles in ENM increase the dependability and security of automated language systems in high-stakes social scenarios.

Fields of Study & Subjects

Ethics and Social Impacts of AI Topic Modeling Hate Speech and Cyberbullying Detection

Publication details

Journal / source
Procedia Computer Science
Volume
275
Pages
567–574
Publication date
1 January 2026
Language
EN