The new Cultural Analytics speaker series — a collaboration between the UC Berkeley School of Information and the Berkeley Institute for Data Science (BIDS) — launched on October 2 with its inaugural speaker, Manvir Singh, Assistant Professor of Anthropology at UC Davis. The series builds on inaugural talks in the past two years by David Blei (Columbia), Kristoffer Nielbo (Aarhus) and Tina Eliassi-Rad (Northeastern). A welcome from I School Dean Eric Meyer and Tim Tangherlini, BIDS Associate Faculty Director and professor in Scandinavian and the School of Information, provided the context for this five-year series focused on the data-driven analysis of cultural phenomena.
Photo: Dean Meyer introduces the Cultural Analytics speaker series
Singh’s research centers on culturally important domains from magic to religion to music. In his talk “Universality and Diversity in Story and Song”, he shared results from two large-scale comparative projects, the Natural History of Song (NHS) and Anansi, that investigate global patterns in music and storytelling, respectively (see abstract).
An intriguing result noted by the NHS project is the reliable association of certain behaviours with song, including dance, infant care (lullabies), and healing. In experiments with speakers of 28 languages in 49 countries and in three smaller-scale societies, they discovered that listeners can reliably identify the function of songs (here dance songs and lullabies) from excerpts as short as fourteen seconds. After identifying two primary dimensions along which musical features vary (rhythmic and musical complexity), the NHS group was also able to determine that dance songs and lullabies represent opposite points in musical space. Intriguingly, and importantly in the context of resisting the urge to find universals where none might exist, long-term fieldwork by anthropologist Kim Hill among the Northern Aché of Paraguay, which notes the absence both of dancing and infant-directed song, has pushed NHS researchers to appreciate how cultural transmission may be required to sustain these expressive forms.
Photo: Manvir Singh discusses his research and the question of how widespread cultural behaviors compare across societies
The Anansi project, which Singh leads, is similarly engaged in the interrogation of the tension between universals and cultural situatedness. Using a manually coded corpus of over 1500 folktales from 76 cultures, the team has been on the search for commonalities and variation in story structure and content. The emergence of LLMs over the past five years has now opened the possibility of greatly expanding the target corpus and shifting away from a reliance on English language translations of target culture resources. Watch the I School event recording for the full account!
A hallmark of the Cultural Analytics series is an afternoon deep-dive with the visitor, allowing for sustained and in-depth conversations. The AI Futures Lab, a crossroads of interdisciplinarity, is the perfect location for these events, as Singh’s afternoon conversation confirmed. Here, students, faculty, and industry partners began to explore critically and in-depth new perspectives of complex cultural phenomena and pave the way for future collaborations. Importantly, culture in this program is not presented solely as an engineering problem, but rather as an opportunity to engage with the expressive forms that emerge whenever people get together, and where data and computation can help with thick description and deeper understanding, a critical first step toward building culturally-aware AI models.
The series will continue throughout the academic year, with opportunities for UC Berkeley students and faculty, and industry partners, to interact with the invited guests.
Photo: A group gathers at the AI Futures Lab for a deep-dive discussion into Singh’s work
To stay in touch and join these far-ranging conversations of critical cultural importance, please join our Cultural Analytics mailing list by visiting this page or emailing bids-cultural-analytics+subscribe@lists.berkeley.edu.