Table of Links
Abstract and I. Introduction
II. Background and Related Work
III. Framework Design
IV. Evaluation
V. Conclusion and Future Work, Acknowledgement, and References
Abstract—Social media platforms such as Twitter, Reddit, and Sina Weibo play a crucial role in global communication but often encounter strict regulations in geopolitically sensitive regions. This situation has prompted users to ingeniously modify their way of communicating, frequently resorting to coded language in these regulated social media environments. This shift in communication is not merely a strategy to counteract regulation, but a vivid manifestation of language evolution, demonstrating how language naturally evolves under societal and technological pressures. Studying the evolution of language in regulated social media contexts is of significant importance for ensuring freedom of speech, optimizing content moderation, and advancing linguistic research. This paper proposes a multi-agent simulation framework using Large Language Models (LLMs) to explore the evolution of user language in regulated social media environments. The framework employs LLM-driven agents: supervisory agent who enforce dialogue supervision and participant agents who evolve their language strategies while engaging in conversation, simulating the evolution of communication styles under strict regulations aimed at evading social media regulation. The study evaluates the framework’s effectiveness through a range of scenarios from abstract scenarios to real-world situations. Key findings indicate that LLMs are capable of simulating nuanced language dynamics and interactions in constrained settings, showing improvement in both evading supervision and information accuracy as evolution progresses. Furthermore, it was found that LLM agents adopt different strategies for different scenarios. The reproduction kit can be accessed at https://github.com/BlueLinkX/GA-MAS.
I. INTRODUCTION
In the modern digital era, social networks like X (Twitter), Reddit, and Facebook have become pivotal in shaping human interaction, primarily through their ability to facilitate vast connectivity and instantaneous information exchange. Yet, in regions with heightened geopolitical or socio-political sensitivities, users often navigate complex user regulations. Their online expressions can lead to severe consequences, including censorship or account suspension, as documented in various news [1], [2]. While intended to curb misinformation and maintain social harmony, these regulations significantly constrain user expression. In response to these regulations, Corresponding Author: Jialong Li users on social networks have adapted by adopting a phenomenon known as “coded language.” [3] In linguistics, Coded Language typically refers to expressing information in a concealed or indirect manner. On social media platforms, this often manifests as the use of metaphors, slang, and creative wordplay.
This adaptation is not merely a circumvention strategy but a vivid example of “language evolution” in a digital context. In linguistics, language evolution refers to the progression and adaptation of languages over time, shaped by societal, cultural, and technological influences. Specifically, in social networks, this language evolution is demonstrated as users constantly adjust their communication styles to test whether they have circumvented oversight. Depending on the level of regulatory pressure and the nature of the audience, users engage in a strategic play with the platform. From indirect descriptions to the creation of new slang, users ultimately develop coded languages of varying degrees of abstraction.
This dynamic shift in communication methods offers deep insights from a sociological perspective, reflecting how societal norms and technological advancements shape language. For platforms and users alike, understanding this evolution is crucial for developing balanced content moderation policies and navigating regulated digital environments. For social media platforms and their users, grasping this concept is equally vital. Platforms need this knowledge to adapt to changing user behaviors, to create balanced content moderation policies, and to identify and counteract harmful or illegal activities. For users, an awareness of how language evolves is vital in navigating the intricacies of regulated digital environments. It helps in maintaining free speech and in developing communication strategies that are both effective and meaningful in fostering enhanced interactions
The emergence of Large Language Models (LLMs) like ChatGPT and Bard, represents a significant leap in Artificial intelligence (AI). These LLMs have demonstrated strong capabilities in (i) understanding intricate dialogues [4], generating coherent texts [5], and aligning to human ethical and value standards [6]–[8]. These capabilities position LLMs as ideal tools to simulate human’s decision-making and language representation, providing new potential in sociology. For instance, [9] investigated the ability of LLMs to comprehend the implicit information in social language. The study by [10] demonstrated the efficiency of LLMs in understanding and generating content that mimics the style of specific social network users. Furthermore, research by [11]–[13] integrated LLMs with Multi-Agent Systems to simulate micro-social networks, observing agent behaviors and strategies that reflect human interactions. Despite the extensive application of LLMs in understanding human intension and simulating social media dynamics, the use of LLMs in studying the specific phenomenon of language evolution under regulatory constraints has not been thoroughly explored. As mentioned above, such simulation could not only preempt criminal activities on social media but also provide technical support to uphold freedom of speech.
Addressing this gap, our research employs LLMs to simulate the nuanced interplay between language evolution and regulatory enforcement on social media. We introduce a simulation framework with two types of LLM-driven agents: (i) participant agents, who adapt their language to communicate concept ’B’ under restrictions, and (ii) supervisory agent, who enforce guidelines and react to these language evolutions. Our approach effectively simulates the dynamics model between both sides in language evolution, which allows us to observe the tension and adaptability inherent in language evolution in a controlled, simulated environment. To assess the framework’s effectiveness, we designed three diverse scenarios: “Guess the Number Game”, “Illegal Pet Trading”, and “Nuclear Wastewater Discharge”. These scenarios vary from abstract concepts to situations closely resembling real-world events, thereby progressively testing the framework from theoretical to practical applications.
The main contributions of this study are:
• We introduce a multi-agent simulation framework utilizing LLMs to simulate human linguistic behaviors in regulated social media environments. This framework offers a unique approach to studying language evolution within the confines of regulatory constraints.
• We conducted an extensive evaluation of LLMs in simulating language evolution and interaction efficacy in regulated social media settings. Through experiments on three distinct scenarios, we not only captured the process of language strategy evolution but also uncovered the varied evolutionary trajectories that LLMs follow under different conditions.
• The experiment reproduction kit, including the proposed simulation framework along with the results of our experiments, are made publicly accessible as open-source assets; The anonymized artifact can be accessed at: https://github.com/BlueLinkX/GA-MAS.
The rest of this paper is organized as follows: Section II provides essential background information and explores related work. Section III is dedicated to presenting our proposed simulation framework. Section IV details the experiment setting, presents results, and discusses a discussion. Finally, Section V concludes the paper and offers an outlook on potential future work.
