Starting in the 1970s, a team of researchers at IBM spearheaded efforts to reorient the field of automatic speech recognition, turning it away from study and simulation of human perception, reason, and expertise and toward a startling new mandate: “There’s no data like more data.” These efforts set off a comprehensive transformation throughout speech and natural language processing, one pivotal to the rise of data-intensive machine learning in the 21st century, and laid the epistemic, commercial, and technical groundwork for mainstream artificial intelligence to be remade in the image of data science. This talk examines the history of automatic speech recognition’s “statistical turn,” which reformulated language modeling from a task of representing linguistic knowledge to one of large-scale statistical pattern recognition. It offers a crucial glimpse into how we became data-driven, highlighting how the making of language into data in turn made data into an imperative, opening the door for the unchecked incursion of algorithmic authority into everyday life.
Xiaochang Li is an Assistant Professor in the Department of Communication at Stanford University, and affiliated faculty in Science and Technology Studies and the program in Modern Thought and Literature. Her forthcoming book, *Divination Engines: Natural Language Processing, Artificial Intelligence, and the Making of Algorithmic Culture* (University of Chicago Press, available in July 2026) examines the history of speech and natural language processing and how the problem of mapping communication to computation fueled the rise of “data-driven” machine learning as a privileged and pervasive form of knowledge.
We will also host a lunch, co-sponsored by the History of Science, Medicine and Technology Program, for any UW-Madison graduate students any faculty in this seminar to meet Professor Li at Noon, To RSVP please sign-up here: https://forms.gle/rbyn6sz1Bizfu3is8
