back to list

Project: Playing Text-based Games with Large Language Models


The project aims to explore the utilization of sophisticated language models in the domain of text-based games. This endeavor seeks to harness the capabilities of large language models, such as GPT (Generative Pre-trained Transformer), in the context of interactive narratives, text adventures, and other similar game formats. The project revolves around the intersection of reinforcement learning, natural language processing, and gaming, showcasing the potential for these models to engage, comprehend, and respond dynamically within the framework of game-based storytelling.

The objectives of this project are multifaceted:

Model Integration and Adaptation: Integrating and adapting pre-trained language models to comprehend the context, nuances, and complexities inherent in text-based games. This involves fine-tuning and adjusting the model's parameters to suit the requirements of interactive gaming environments.

Interactive Gameplay Development: Creating a system that enables players to engage in immersive, interactive gameplay experiences where the language model serves as an AI-driven game master. This system will understand and generate responses based on the narrative flow and player inputs, ensuring a coherent and engaging gaming experience.

Adaptive Storytelling and Dynamic Narratives: Implementing a mechanism where the language model dynamically adjusts the storyline based on player choices, creating a personalized and evolving narrative experience. This adaptive storytelling approach aims to enhance player immersion and engagement by providing a sense of agency and personalization within the game's world.

Evaluation and User Experience Analysis: Conducting rigorous evaluations to assess the effectiveness, responsiveness, and overall gaming experience facilitated by the integration of large language models. Gathering feedback from users to understand their experiences and perceptions of playing text-based games driven by AI models.

This project seeks to not only showcase the technical capabilities of large language models in a gaming context but also to explore the potential of AI-driven storytelling and the enhancement of user experiences within interactive narrative environments. The outcomes of this endeavor could have implications for the gaming industry, interactive storytelling platforms, and broader applications in AI-driven entertainment and user engagement.


Stay Moral and Explore: Learn to Behave Morally in Text-based Games

Zijing Shi*, Meng Fang*, Yunqiu Xu, Ling Chen, Yali Du

In ICLR 2023

Self-imitation Learning for Action Generation in Text-based Games

Zijing Shi, Yunqiu Xu, Meng Fang, Ling Chen

In EACL 2023

Keep CALM and Explore: Language Models for Action Generation in Text-based Games

Shunyu Yao, Rohan Rao, Matthew Hausknecht, Karthik Narasimhan

In EMNLP 2020

Meng Fang
Secondary supervisor
Yudi Zhang
Get in contact