Towards a Unified Model of Musical Form

Bridging Music Theory, Digital Corpus Research, and Computation

The system of tonal music has evolved a range of characteristic ways in which music is organized such that it affords its interpretability by a listening mind. As a central part of that system, musical form characterizes the large-scale organization of musical structure. This involves the segmentation into parts and their hierarchical grouping, repetition of musical material, and form-functionality, which are interconnected with other structural domains, such as (hyper)meter, harmony, melody, and voice-leading structure. Their interplay enables, for instance, the interpretation of formal units in terms of nested formal functions, such as the “beginning of the end (section)” or “the end of a beginning (section).”

Over the past ca. 25 years, the study of musical form has developed into a vibrant field. So far, most research endeavours have concentrated on describing a small number of prominent, concrete forms such as sonata form, rondo, variation, or minuet, focusing on confined historical repertoires. At present, the field still lacks an overarching theory that (a) models form and its components flexibly across styles, (b) is empirically grounded in large-scale corpus studies, and (c) is mathematically precise such that it lends itself for (d) the design of efficient algorithms for computational models of form. Further, there is still no large digital corpus of expert form analyses and no computational model achieving high-quality form analyses to aid researchers and musicians.The chief purpose of our interdisciplinary project is to address these lacunae.

Firstly, we will develop a generalized formal theory of musical form that brings together state-of-the-art music theory with the theory of formal languages in order to devise a formalism and a notation language that characterizes human expert intuition. Secondly, we will develop an annotation standard for form analyses in digital scores, compile a large corpus with expert analyses, and develop web-based annotation and visualization tools. Thirdly, we will develop computational models of musical form based on corpus analysis, parsing methods, and machine learning heuristics. The formal model poses the major algorithmic challenge of automatic deduction of form. We will tackle these challenges with state-of-the-art methods to achieve tractability of our model while maintaining the expressive power to produce results that are useful for humans. Finally, a generalized model of form and a working computational implementation will likely add immense value for musicological research, and enable multiple applications in Music Information Retrieval and automatic music generation.