AEGIS: Awareness-Enhanced Guidance for Iterative Safeguard

2026-07-20

AI Summary

This work investigates the conditional efficacy of span-level guidance in multilingual text detoxification and its impact on the trade-off between toxicity reduction and semantic preservation. To this end, the authors propose the AEGIS framework, which integrates structured guidance—including outputs from a span-level detector, toxic spans, intensity labels, and target attributes—to enable controlled rewriting atop a frozen generator backbone. The study presents the first systematic evaluation of span-based guidance across diverse generator architectures, model scales, and languages (English, Chinese, and Korean), demonstrating its ability to effectively modulate the balance between detoxification and semantic fidelity. However, the results also reveal that performance is highly dependent on both the generator architecture and the linguistic context, thereby highlighting both the potential and limitations of this approach.

Abstract

Span-level rationales are often assumed to improve controllability in text detoxification, but it remains unclear when such guidance helps and when it introduces trade-offs. We present Awareness-Enhanced Guidance for Iterative Safeguard (AEGIS) as an exploratory framework for studying span-guided multilingual detoxification across English, Mandarin Chinese, and Korean. AEGIS combines span-level detector outputs with frozen generator backbones, allowing harmful spans, intensity labels, and target attributes to be provided as structured guidance during rewriting. Rather than claiming state-of-the-art detoxification performance, we analyze how span guidance affects the balance between toxicity reduction and meaning preservation across generator families, model scales, and languages. Our results suggest that span-guided detoxification is conditionally useful: explicit rationales change the trade-off between toxicity reduction and meaning preservation, but their effects depend strongly on the generator backbone and the linguistic context. These findings highlight both the promise and the limitations of span-level control signals for multilingual detoxification.

Problem

• text detoxification
• span-level rationales
• multilingual
• toxicity reduction
• meaning preservation

Innovation

• span-level rationales
• multilingual detoxification
• structured guidance
• toxicity reduction
• meaning preservation