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 —

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.