Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Procedurally generating rules to adapt difficulty for narrative puzzle games

Thomas Volden, Djordje Grbic, Paolo Burelli

TL;DR

The paper tackles adaptive difficulty in kindergarten narrative puzzle games by procedurally generating rules with a genetic algorithm and framing them through large language model storytelling. The approach combines a rule-concept based PCG framework with a tangible YOLI Board testbed and demonstrates that target difficulty can be closely approximated within ~22 generations. A narrative framing step using an LLM is employed to justify rules contextually, aiming to improve comprehension for children. The work lays a foundation for scalable, child-friendly adaptive learning on physical-digital puzzle platforms and points to future directions in child-tailored LLMs and multi-modal evaluation.

Abstract

This paper focuses on procedurally generating rules and communicating them to players to adjust the difficulty. This is part of a larger project to collect and adapt games in educational games for young children using a digital puzzle game designed for kindergarten. A genetic algorithm is used together with a difficulty measure to find a target number of solution sets and a large language model is used to communicate the rules in a narrative context. During testing the approach was able to find rules that approximate any given target difficulty within two dozen generations on average. The approach was combined with a large language model to create a narrative puzzle game where players have to host a dinner for animals that can't get along. Future experiments will try to improve evaluation, specialize the language model on children's literature, and collect multi-modal data from players to guide adaptation.

Procedurally generating rules to adapt difficulty for narrative puzzle games

TL;DR

The paper tackles adaptive difficulty in kindergarten narrative puzzle games by procedurally generating rules with a genetic algorithm and framing them through large language model storytelling. The approach combines a rule-concept based PCG framework with a tangible YOLI Board testbed and demonstrates that target difficulty can be closely approximated within ~22 generations. A narrative framing step using an LLM is employed to justify rules contextually, aiming to improve comprehension for children. The work lays a foundation for scalable, child-friendly adaptive learning on physical-digital puzzle platforms and points to future directions in child-tailored LLMs and multi-modal evaluation.

Abstract

This paper focuses on procedurally generating rules and communicating them to players to adjust the difficulty. This is part of a larger project to collect and adapt games in educational games for young children using a digital puzzle game designed for kindergarten. A genetic algorithm is used together with a difficulty measure to find a target number of solution sets and a large language model is used to communicate the rules in a narrative context. During testing the approach was able to find rules that approximate any given target difficulty within two dozen generations on average. The approach was combined with a large language model to create a narrative puzzle game where players have to host a dinner for animals that can't get along. Future experiments will try to improve evaluation, specialize the language model on children's literature, and collect multi-modal data from players to guide adaptation.
Paper Structure (6 sections, 3 figures)

This paper contains 6 sections, 3 figures.

Table of Contents

  1. Introduction
  2. YOLI Board and simulator
  3. Procedural rule generator
  4. Framing rules in a narrative context
  5. Current results
  6. Future plans

Figures (3)

  • Figure 1: Example of a tile game with thirty tiles, where a player has to match tiles that represent the same number or the same symbol in numerical order.
  • Figure 2: Thirty generic tiles generated, with the properties group, type, and color, to use as a sample for the random rule generator.
  • Figure 3: Prompt input to generate a story from an initial game definition and new prompt after the player has placed five tiles. The outputs are from the language model.