Beyond Pre-Defined Scripts: Player Perceptions on Generative Non-Player Character Dialogues

Interactions with non-player characters (NPCs) are ubiquitous in video games but traditionally rely on scripted dialogue. The advent of large language models (LLMs) has opened up new opportunities to converse with NPCs and, in turn, contribute to player experience. However, integrating LLM-driven dialogue in games is not necessarily straightforward and needs an understanding of how this new technique is perceived by players and shapes their game experience. In this paper we contribute to this emerging line of inquiry and report on a study investigating players' perceptions on LLM-generated dialogues in a bespoke game. Towards this end, we have conducted an online survey, containing both quantitative measures and qualitative open-ended questions, with 62 participants. Our results indicate that LLM-generated dialogues have several benefits such as enhanced input flexibility or leading to more natural conversations but can create a variety of undesired side-effects that might be hard to anticipate beforehand or control for.