PNAS – Proceedings of the National Academy of Sciences of the United States of America
Dissecting landscape art history with information theory
Byunghwee Lee, Min Kyung Seo, Daniel Kim, In-seob Shin, Maximilian Schich, Hawoong Jeong, and Seung Kee Han
PNAS first published October 12, 2020. https://doi.org/10.1073/pnas.2011927117
A foundational question in art and aesthetics is if there are culturally and temporally transcendent characteristics in the organizing principles within art and if yes, how these principles evolve over time. We propose a simple yet coherent information-theoretic framework that captures compositional proportion as used for dissecting landscape paintings by artists. The analysis of 14,912 landscape paintings representing the canonical historiography of Western art uncovers that the preferred compositional proportion within the historiography of landscape paintings systematically evolves over time. The network analysis of similarity distributions reveals clear clustering structures of individual artist’s artworks, absolute time periods, and conventional style periods, suggesting meaningful supergroups of art historical concepts, which remained so far nonintuitive to a broader audience.
Painting has played a major role in human expression, evolving subject to a complex interplay of representational conventions, social interactions, and a process of historization. From individual qualitative work of art historians emerges a metanarrative that remains difficult to evaluate in its validity regarding emergent macroscopic and underlying microscopic dynamics. The full scope of granular data, the summary statistics, and consequently, also their bias simply lie beyond the cognitive limit of individual qualitative human scholarship. Yet, a more quantitative understanding is still lacking, driven by a lack of data and a persistent dominance of qualitative scholarship in art history. Here, we show that quantitative analyses of creative processes in landscape painting can shed light, provide a systematic verification, and allow for questioning the emerging metanarrative. Using a quasicanonical benchmark dataset of 14,912 landscape paintings, covering a period from the Western renaissance to contemporary art, we systematically analyze the evolution of compositional proportion via a simple yet coherent information-theoretic dissection method that captures iterations of the dominant horizontal and vertical partition directions. Tracing frequency distributions of seemingly preferred compositions across several conceptual dimensions, we find that dominant dissection ratios can serve as a meaningful signature to capture the unique compositional characteristics and systematic evolution of individual artist bodies of work, creation date time spans, and conventional style periods, while concepts of artist nationality remain problematic. Network analyses of individual artists and style periods clarify their rhizomatic confusion while uncovering three distinguished yet nonintuitive supergroups that are meaningfully clustered in time.