Toward Automated Quest Generation in Text-Adventure Games

TL;DR

This paper tackles automated quest generation in text-adventure games by comparing Markov-chain and neural-language approaches within a cooking-recipe domain, grounded by semantic knowledge graphs. A Markov-based pipeline uses an ingredient graph and instruction generation, while a neural approach uses a 4-layer LSTM for ingredients and GPT-2 fine-tuning for title and steps, both integrated into a TextWorld-based world. Human studies (n=75) assess coherence, novelty, surprise, and value, revealing that neural methods yield higher coherence and value, Markov methods offer greater novelty, and all generated designs can match or exceed human-authored quests under certain conditions. The work demonstrates that quest content can be automatically generated and coherently embedded in a game world, with implications for scalable, creative interactive narratives. This has practical impact for developing automated, engaging text-adventure experiences without extensive human authoring.

Abstract

Interactive fictions, or text-adventures, are games in which a player interacts with a world entirely through textual descriptions and text actions. Text-adventure games are typically structured as puzzles or quests wherein the player must execute certain actions in a certain order to succeed. In this paper, we consider the problem of procedurally generating a quest, defined as a series of actions required to progress towards a goal, in a text-adventure game. Quest generation in text environments is challenging because they must be semantically coherent. We present and evaluate two quest generation techniques: (1) a Markov model, and (2) a neural generative model. We specifically look at generating quests about cooking and train our models on recipe data. We evaluate our techniques with human participant studies looking at perceived creativity and coherence.

Figures (10)

• Figure 1: Example of ingredient connections.
• Figure 2: Ingredient extraction process.
• Figure 3: Object graph in the one room map.
• Figure 4: Room layout in the five room map.
• Figure 5: Object graph in the five room map.