構文型と乱数型に基づく作詞生成モデルの研究 — A Hybrid Model of Lyric Generation Based on Structural and Stochastic Processes —

Abstract This paper proposes a novel generative framework for lyric composition based on the interaction between two complementary mechanisms: a structural model and a stochastic model. The structural model represents deliberate composition driven by semantic intention, narrative coherence, and thematic control, while the stochastic model represents spontaneous generation driven by randomness, emotional fluctuation, and unconscious associative processes. Through iterative experimentation using human–AI collaborative lyric generation, this study demonstrates that neither purely structural nor purely stochastic methods are sufficient to produce emotionally resonant lyrics. Instead, a hybrid process—characterized by dynamic oscillation between randomness and structure—emerges as the most effective creative mechanism. The paper further conceptualizes this process as a form of convergent creative dynamics, in which meaning arises through repeated cycles of divergence and reintegration rather than linear optimization. This model provides an alternative theoretical foundation for understanding creativity not only in lyric writing but also in broader domains of human–AI co-creation. The proposed framework contributes to creative AI research by offering a process-oriented theory grounded in practical generative experiments, positioning prompt-based AI interaction as an executable form of theoretical validation. 概要(日本語) 本研究は、作詞における創造過程を「構文型生成」と「乱数型生成」という二つの異なる生成機構の相互作用として捉え、それらを統合するハイブリッド生成モデルを提案するものである。 構文型生成は意味構造・物語性・意図性に基づく設計的生成を表し、乱数型生成は偶発性・感情揺らぎ・無意識的連想に基づく自発的生成を表す。本研究では、人間と生成AIによる協働的作詞実験を通じて、両者の往復運動によってのみ、感情的説得力を持つ歌詞が生成されることを示した。 この創造過程は線形最適化ではなく、「発散と回収を繰り返す収束力学」としてモデル化される。本理論は作詞に留まらず、人間とAIの協働的創造全般に対する新たな理論的枠組みを提供するものである。