Generating dynamic and lifelike NPC dialogs in role-playing games using large language model
Year: 2024
Type: dissertation
Abstract: Traditional non-player character dialog systems rely on preset lines, which lack realism and dynamics and reduce player immersion. This study aims to explore overcoming these limitations through large language models. \n \nThis paper reviews existing non-player character dialogue systems and emphasizes the need to generate dynamic, context-sensitive dialogs. The capabilities of large language models, especially GPT-3 and GPT-4, in generating coherent, contextually relevant content are explored. In this paper, I propose a framework for non-player role dialog based on large language model. The framework includes key components such as intent recognition, memory retrieval, dialog management, and response generation. The framework was tested in practice in the text-based role-playing game Call of Cthulhu. Evaluations from game participants indicated that the dialog based on the large language model was more natural and contextualized, enhancing the overall game experience. \n \nThis study demonstrates the feasibility and advantages of integrating large language models into non-player character dialog systems, providing valuable exploration and practice for developing smarter and more responsive game characters. However, there are still challenges in managing long-term memory and context in non-player character dialog systems based on large language models, which require further research.
Source: LUTPub (LUT University)
Author: Huang, Junyang
Institution: Lappeenranta-Lahti University of Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7028647704