언어모델을 이용한 반전이 있는 스토리 생성 및 평가 = Generation and Evaluation of Twisted-Plot Stories Using Large Language Models

다국어 초록 (Multilingual Abstract)

Peripeteia, introduced by Aristotle in Poetics, is considered the archetype of plot twists, playing a critical role in heightening narrative tension and engaging readers. While plot twists contribute to the depth and meaning of narratives across various media, research on generating plot twists using language models remains limited. This study explores effective prompt engineering techniques for generating plot-twist narratives using Large Language Models (LLMs). Narratives were evaluated based on twist quality, coherence, grammatical accuracy, verisimilitude, and enjoyment, utilizing GPT-4o-mini, Llama-3-8B, and Claude-3-Haiku. The results indicate that prompt designs tailored to the characteristics and objectives of narrative elements significantly influence the quality of generated twists. By generating and evaluating plot-twist narratives using language models, this study provides insights into enhancing the narrative generation capabilities of LLMs.