Global Prediction of Time-Span Tree by Cloze Task
摘要
Time-span trees in A Generative Theory of Tonal Music (GTTM) have global and local relationships. However, no analysis based on global relationships has been done, and higher-order mechanisms have not been clarified. Therefore, in this research, we will clarify this mechanism by masking the time-span tree as it is and using it as a cloze task. To experiment with the cloze task, we vectorized and embedded the tree structure. We also extended the data by changing the pitch and masking. As a result of experiments, it is possible to predict higher layers when masking a small maximum time-span.