What Dance Scholars Can Learn From Warehouse Surveillance

Why It Matters

Applying warehouse‑surveillance analytics to traditional performance data gives scholars a scalable tool to track cultural change, safeguard intangible heritage, and inform policy on arts funding.

Key Takeaways

• Wayang kulit performances generate extensive YouTube data for analysis.
• Temporal action segmentation can map centuries-long performance structures.
• Social media captures ~90% of Java’s wayang events accurately.
• Pandemic boosted online viewership, revealing resilient cultural demand.
• Multimodal models enable computational study of traditional arts.

Summary

Miguel Escobar Varela, a computational folklorist, presented how warehouse surveillance techniques—specifically temporal action segmentation—can illuminate the evolving practice of wayang kulit, the Indonesian shadow‑puppet theater. He framed the talk in three “pathets,” mirroring the art form’s own structural divisions, and argued that digital recordings function like surveillance footage, offering granular timestamps for analysis.

He demonstrated that YouTube now hosts 5,000‑7,000 wayang recordings annually, with median video lengths growing from snippets a decade ago to full five‑hour performances today. By cross‑referencing social‑media mentions, his team estimates roughly 90 % of live shows are captured online, revealing seasonal spikes and regional language effects on performance density.

Historical visual records from 1846 show striking continuity with modern audiences—still 1,000 people, similar seating, even sleeping spectators. A notable surge in median view counts to 20,000 in 2019 coincided with the COVID‑19 lockdown, underscoring the tradition’s resilience when live attendance was impossible. The pathet‑adegan hierarchy provided a natural labeling scheme for training multimodal models.

These findings suggest that dance and performance scholars can adopt surveillance‑style segmentation to quantify stylistic shifts, audience engagement, and transmission pathways across centuries. Computational pipelines promise richer heritage preservation, real‑time monitoring, and new interdisciplinary collaborations between humanities and AI.

Original Description

Emic Approaches to Temporal Action Segmentation

Intangible cultural heritage (ICH) presents unique challenges for scholarly analysis. Unlike material heritage with its monuments and artifacts, ICH exists primarily in performance, practice, and living tradition. Video recordings offer some of the most comprehensive documentation of these ephemeral expressions, yet analyzing large video archives has remained a daunting challenge.

In this talk, I will describe how temporal action segmentation (TAS), a computational approach developed for activity recognition, can be adapted to ICH video analysis, enabling researchers to identify culturally meaningful performance segments at scale. Through a case study of Javanese puppet theater comprising over 10,000 recordings, I will demonstrate how TAS can address longstanding questions about performance evolution while respecting culture-specific narrative structures.

This approach advances what I term emic segmentation, the computational study of semantic units as they are conceived and valued within their originating cultural contexts. I will also present open-source tools and a complete methodological pipeline to extend this work to other cultural contexts.

Co-sponsored by the Berkeley Institute for Data Science, the School of Information, and the Department of Scandinavian.

Speaker

Miguel Escobar Varela

Miguel Escobar Varela is associate professor at the Department of English, Linguistics and Theatre Studies and deputy director of the Centre for Computational Social Science and Humanities at the National University of Singapore. In his research, he studies the changing landscape of Southeast Asian cultural heritage by combining fieldwork with computational methods (such as natural language processing, computer vision and network analysis). He is the author of Theatre as Data (University of Michigan Press, 2021). His papers, datasets and research software are available at https://miguelescobar.com.

More info:

https://www.ischool.berkeley.edu/events/2026/what-dance-scholars-can-learn-warehouse-surveillance-emic-approaches-temporal-action