Beyond Surface Forms: A Mechanism-Oriented Taxonomy of Indirect Linguistic Encoding for LLM-Based Coded Language Detection

arXiv·medium signal

Builds a comprehensive, mechanism-oriented taxonomy of indirect linguistic expressions — algospeak, euphemisms, and adversarial obfuscation — that users invent to evade moderation and surveillance. The framework organizes how meaning is camouflaged so LLM-based detectors can target the underlying encoding mechanism rather than surface forms. Directly applicable to trust-and-safety and content-moderation pipelines.