Authors:
(1) Jinyu Cai, Waseda University (bluelink@toki.waseda.jp);
(2) Jialong Li, Waseda University (lijialong@fuji.waseda.jp);
(3) Mingyue Zhang, Southwest University (myzhangswu@swu.edu.cn);
(4) Munan Li, Dalian Maritime University (limunan@dlmu.edu.cn);
(5) Chen-Shu Wang, National Taipei University of Technology (wangcs@ntut.edu.tw);
(6) Kenji Tei, Tokyo Institute of Technology (tei@c.titech.ac.jp). Authors: Authors: (1) Jinyu Cai, Waseda University (bluelink@toki.waseda.jp); (2) Jialong Li, Waseda University (lijialong@fuji.waseda.jp); (3) Mingyue Zhang, Southwest University (myzhangswu@swu.edu.cn); (4) Munan Li, Dalian Maritime University (limunan@dlmu.edu.cn); (5) Chen-Shu Wang, National Taipei University of Technology (wangcs@ntut.edu.tw); (6) Kenji Tei, Tokyo Institute of Technology (tei@c.titech.ac.jp). Table of Links Abstract and I. Introduction Abstract and I. Introduction II. Background and Related Work II. Background and Related Work III. Framework Design III. Framework Design IV. Evaluation IV. Evaluation V. Conclusion and Future Work, Acknowledgement, and References V. Conclusion and Future Work, Acknowledgement, and References II. BACKGROUND AND RELATED WORK This section offers an extensive background and overview of related work in areas relevant to this study, starting with foundational information on LLMs, then exploring studies in slang detection and identification as they relate to language evolution, and concluding with a discussion on recent research applying LLMs to evolutionary game theory and social simulations. A. Large Language Models A. Large Language Models Large Language Models like the GPT series [14], [15], LLaMA series [16], [17], PaLM series [18], [19], GLM [20]and Bard [21] represent a significant advancement in the field of natural language processing. Fundamentally, these models are based on the Transformer [22] architecture, a type of neural network that excels in processing sequential data through self-attention mechanisms. This architecture enables LLMs to understand and predict linguistic patterns effectively. They are trained on extensive text datasets, allowing them to grasp a wide range of linguistic nuances from syntax to contextual meaning. These models exhibit remarkable zero-shot learning abilities, enabling them to perform tasks they were not explicitly trained for, like understanding and generating content in new contexts or languages [4], [5], [23]–[25]. A critical aspect of their training involves Reinforcement Learning from Human Feedback [26] (RLHF), where human reviewers guide the model to produce more accurate, contextually relevant, and ethically aligned responses. This method not only enhances the model’s language generation capabilities but also aligns its outputs with human values and ethical standards, making them more suitable for diverse, real-world applications. B. Slang Detection and Identification B. Slang Detection and Identification In the field of Natural Language Processing (NLP), the evolution of language has always been a subject of significant interest. Existing studies have primarily focused on utilizing various machine-learning techniques to recognize informal expressions within text [27]. These methods often include rule-based systems, statistical models, and early machine-learning technologies. For instance, [28] has employed predefined slang dictionaries and heuristic rules to identify and categorize informal language, proving effective on specific datasets but generally lacking the flexibility to adapt to emerging expressions and changing contexts. On the other hand, explorations have been made into using statistical models, such as Naive Bayes classifiers and Support Vector Machines (SVMs) [29], for the automatic detection of slang in text. These approaches rely on extensive annotated data but still face limitations when dealing with newly emerged slang or evolving forms of language. [30] views the generation of slang as a problem of selecting vocabulary to represent new concepts or referents, categorizing them accordingly. Subsequently, it predicts slang through the use of various cognitive categorization models. The study finds that these models greatly surpass random guessing in their ability to predict slang word choices. [31] proposed a Semantically Informed Slang Interpretation (SSI) framework, applying cognitive theory perspectives to the interpretation and prediction of slang. This approach not only considers contextual information but also includes the understanding of semantic changes and cognitive processes in the generation of slang. It is noteworthy that these traditional research methods have mainly focused on detecting or predicting existing slang and keywords, rather than generating slang expressions. This stands in stark contrast to the research focus of this paper. C. Evolutionary Game and Social Simulation with LLMs C. Evolutionary Game and Social Simulation with LLMs Merging evolutionary game theory with LLMs has unlocked innovative pathways for simulating complex game dynamics, extending beyond simple dialogue generation to the development and progression of game strategies. LLMs are employed to engage and refine strategic play within game-theoretical frameworks, as demonstrated by [32], which delves into the application of LLMs in negotiation-based games. This study underscores the ability of LLMs to advance their negotiation skills through continuous self-play and feedback loops with AI. LLMs also show proficiency in social deduction games such as Werewolf, as explored by [33]. In this context, a specialized framework leverages historical communication patterns to enhance LLM performance, exemplifying how LLMs can evolve intricate game strategies autonomously. Building on this, [34] combines reinforcement learning with LLMs, utilizing LLMs to output action spaces and employing reinforcement learning models for final decision-making. This enables the agents to maintain competitiveness while outputting reasonable actions, even outperforming human adversaries in games like Werewolf. This growing trend of employing LLMs in diverse simulation scenarios extends beyond game theory into broader aspects of social interactions and historical analysis. LLMs have proven to be versatile tools in simulating social dynamics and historical events, offering insights into complex human behaviors and societal patterns. [12] introduces a Wild Westinspired environment inhabited by LLM agents that display a wide array of behaviors without relying on external real-world data. Simultaneously, S3 [13] mirrors user interactions within social networks, crafting an authentic simulation space through the incorporation of user demographic prediction. The influence of LLM-driven social robots on digital communities is thoroughly examined in [35], which identifies distinct macrolevel behavioral trends. Furthermore, [11] employs LLMbased multi-agent frameworks to recreate historic military confrontations, offering a window into the decision-making processes and strategic maneuvers that have directed significant historical conflicts. This avenue of research accentuates the utility of LLMs in computational historiography, providing a deeper comprehension of historical events and their relevance to contemporary and future societal trajectories. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Part of HackerNoon's growing list of open-source research papers, promoting free access to academic material.

Your Next Slang Phrase Might be Created by an AI

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

A Comprehensive Overview of Image Generation Models: From GANs to Diffusion Techniques

Speaking in Code: How AI Simulates Language Evolution on Regulated Social Media

Your Social Media Supervisor Could Soon Be an AI, And It's Getting Smarter

Can GPT Outsmart Social Media Regulations? Inside an AI Language Evolution Experiment

What’s Next for AI in Regulated Social Media?

How Language Became Infrastructure—and Why That Should Worry Us

A Comprehensive Overview of Image Generation Models: From GANs to Diffusion Techniques

Speaking in Code: How AI Simulates Language Evolution on Regulated Social Media

Your Social Media Supervisor Could Soon Be an AI, And It's Getting Smarter

Can GPT Outsmart Social Media Regulations? Inside an AI Language Evolution Experiment

What’s Next for AI in Regulated Social Media?

How Language Became Infrastructure—and Why That Should Worry Us

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps