A Prototype Quest Generator for Simulating Human-Authored Narrative

Abstract

This thesis investigates several techniques of player-tailored procedural quest generation in terms of authenticity and immersion. This is done through the application of a shotgun hill climbing algorithm as well as surface realisation using a natural language generator. Thereafter, a quantitative fitness evaluation is applied using discoveries from previous research. Moreover, several qualitative assessment methods are proposed. From the findings, this thesis concludes that the utilisation of shotgun hill climbing yields high-quality results that can be applied to generate questlines at runtime. However, one should comprehend that the proposed natural language generation using transformer models needs extensive further refinement in order to provide holistically suitable narrative for games.