A Cross-Linguistic (Chinese-English) Study of Agent Adaptation under Content Moderation

Wen-Tzu Huang, Yoyo Tsung-Yu Hou

Keywords: Large Language Models (LLMS), agent-based simulation, content moderation

TL;DR: This study investigates how writing-system affordances in Chinese and English shape the adaptive communication strategies and coordination costs of LLM agents under content moderation pressure.

Abstract: Policy-oriented LLM agent simulations often treat language as a neutral channel, but communication under regulation may depend on the linguistic system itself. We study content moderation as a policy intervention and ask how language form might change its downstream effects. In a Chinese–English comparison with matched tasks and calibrated enforcement, we examine how writing-system affordances shape indirect expression and how agents adapt over repeated interactions. We focus on governancerelevant outcomes, including coordination cost, convention formation, and information recovery accuracy under constraints.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 5