A Structural Analysis of Immersive Descriptions in Game Manuals

Applying a Pseudo-Labeling Method with Large Language Models

Abstract

This study focuses on the immersive function of game manuals, one of their dual roles alongside instruction. To address the lack of systematic analysis of the narrative elements in manuals, this research proposes a method for automatically structuring text into sections such as "narrative" and "character" using pseudo-labeling with Large Language Models (LLMs). A case study analyzing 30 titles from the "Nintendo Classic Mini" revealed that even early manuals contain rich narrative descriptions, and identified genre patterns such as fantasy and science fiction. This approach enables the large-scale digital archiving of manuals and automatic metadata generation, enhancing their value as primary sources for game history research.