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

2026-06-25

AI Summary

This study addresses the challenge of detecting indirect language on social media, which users frequently employ in diverse and dynamically evolving forms to circumvent content moderation. Existing approaches struggle to identify such obfuscated expressions effectively. To overcome this limitation, the work proposes the first mechanism-oriented taxonomy for encoding and classifying indirect language, focusing on the underlying semantic encoding and decoding operations rather than surface forms or communicative intent. This framework is integrated into prompt design for large language models (LLMs). Evaluated on a human-annotated dataset through comparative experiments, the approach consistently outperforms baseline methods across three prominent LLMs, achieving a 4.7% absolute improvement in accuracy on document-level detection and a 5.4% gain in F1 score on span-level tasks, thereby offering a robust and scalable solution for identifying subtly encoded sensitive content.

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 incorporating it into LLM prompts and comparing it with four existing taxonomies and a no-taxonomy baseline, using 2,000 manually annotated TikTok and Bluesky posts. The proposed taxonomy attains the strongest document- and span-level performance across the three LLMs, achieving an improvement of 4.7% in accuracy and 5.4% in F1 over the best-performing benchmark. The empirical results reveal the importance of a comprehensive, mechanism-oriented taxonomy as a stable scaffold for detecting emerging coded language and a useful input to content moderation. Disclaimer: This paper contains content that may be profane, vulgar, or offensive.

Problem

• indirect linguistic encoding
• coded language detection
• content moderation
• LLM-based detection
• social media surveillance

Innovation

• indirect linguistic encoding
• mechanism-oriented taxonomy
• coded language detection
• LLM prompting
• content moderation