Beyond Surface Forms: A Comprehensive, Mechanism-Oriented Taxonomy of Indirect Linguistic Encoding for LLM-Based Coded Language Detection
arXiv:2606.27314v1 Announce Type: new Abstract: To avoid moderation and surveillance on social media, some users routinely invent indirect linguistic expressions (ILE) that camouflage sensitive meanings. Such expressions surface as algospeak, euphemisms, and adversarial obfuscation, depending on intent and context, and they involve recurring encoding mechanisms. We propose a comprehensive, mechanism-oriented taxonomy of ILE that abstracts away from communicative goals and instead categorizes the underlying operations through which meaning is encoded and recovered. We evaluate the taxonomy by i
The proliferation of LLMs and their increasing deployment in moderation systems necessitates advanced methods to detect subtly encoded, sensitive language on social media platforms.
This research provides a foundational taxonomy for understanding how LLM-based systems can identify indirect linguistic expressions, critical for content moderation, platform safety, and mitigating misinformation.
The ability to systematically categorize and detect indirect linguistic encoding will enhance the efficacy of AI-driven content moderation, leading to more robust detection of harmful or restricted content.
- · Social media platforms
- · Content moderation services
- · AI safety researchers
- · Users seeking safer online environments
- · Actors using algospeak for illicit purposes
- · Adversarial obfuscation techniques
Improved detection of 'coded' language by LLMs, leading to more effective content moderation.
Increased pressure on bad actors to develop even more sophisticated obfuscation techniques, spurring an AI-moderation arms race.
Potential for new forms of censorship or over-moderation if the taxonomy is applied too broadly or without nuanced understanding of cultural contexts.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL