Text generation for quests in multiplayer role-playing video games

In this research, the feasibility and limitations of fine-tuning large language models for the use of quest backstory generation are explored. Previous work in this section of computational creativity is analyzed, after which four datasets were gathered. The generation results are discussed, and evaluated with users through both a digital survey and a focus group interview. A framework for statistical evaluation of generated text quality in this domain is proposed, employed, and evaluated. While improvements were observed when compared to previous work in this field, the evaluation results still showed several limitations and shortcomings of the currently used methods. Several avenues for possible continuations of this work are proposed.