---
title: "Beyond Surface Form: A Comprehensive, Mechanism-Oriented Taxonomy of Indirect Language Encoding for LLM-Based Coded Language Detection"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2606.27314v1"
permalink: "https://feed.astralor.com/items/itm_37061fc0b41ab11beb2725fe"
published: "2026-06-25T17:29:45.000Z"
score: 53
channel: "ai"
tags: ["LLM","Content Moderation","Taxonomy","Social Media","Coded Language Detection"]
---

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

## At a glance

Social media users often employ indirect linguistic expressions (ILE) to evade content moderation, such as algospeak, euphemisms, and adversarial obfuscation. Existing taxonomies focus on communicative intent, leading to unstable detection. This paper proposes a comprehensive, mechanism-oriented taxonomy that abstracts underlying encoding operations and integrates it into LLM prompts. On 2,000 manually annotated TikTok and Bluesky posts, the new taxonomy achieves the strongest document- and span-level performance across three LLMs, with a 4.7% accuracy improvement and 5.4% F1 improvement, providing a stable scaffold for detecting emerging coded language.

## Key points

- Proposes a comprehensive, mechanism-oriented taxonomy of indirect linguistic expressions (ILE) focusing on underlying encoding operations rather than communicative intent.
- Integrates the taxonomy into LLM prompts and evaluates on 2,000 manually annotated social media posts.
- Achieves 4.7% accuracy and 5.4% F1 improvement over the best-performing benchmark compared to four existing taxonomies and a no-taxonomy baseline.
- Highlights the importance of a comprehensive, mechanism-oriented taxonomy as a stable scaffold for detecting emerging coded language.

## Full text

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.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_37061fc0b41ab11beb2725fe