From Friends to Modern Family to Saturday Night Live, these are just a few examples of famous and popular TV comedies. Even though they all belong to the comedy genre, these shows differ greatly in structure, including camera techniques, scripting, and overall style. There are different ways to identify what a comedy series is, such as the use of a laugh track or a live audience. However, another way to analyze comedy is through computational methods. A Professor of Data Science at the University of Richmond, Taylor Arnold, argued in his Cultural Analytics talk from March 2026 that different comedy subgenres have distinct structural styles, and these can be detected using computational techniques.
Questioning How We Define Comedy
Is it possible for comedy forms to have their own measurable style? What structural elements help us tell different genres apart? Arnold explores these questions and presents multiple examples, such as older multi-camera sitcoms, which are known for using several cameras at once, live audiences, and laugh tracks. Additionally, he introduces French shortcoms, which are extremely short sitcom-like programs. Arnold highlights that even within this format, there are major stylistic differences in the comedy genre. Some are among the slowest in the dataset because the camera barely changes, while others are among the fastest due to rapid cuts. This demonstrates that comedy genres exist across a range of forms rather than in one fixed category.
Photo: Taylor Arnold answering questions during the Cultural Analytics week of multimodal models for the study of culture talks.
Turning Comedy Into Data
The structural elements of comedy, such as camera techniques, editing pace, and visual style, are then translated into measurable features, including median and average shot length, whether a frame is a close-up, how often a single person appears in the frame, brightness levels, and the amount of movement, speech, and speaking turns. These features allow the computer to analyze the structural style of each episode.
What the Data Reveals About Comedy
Rather than focusing on traditional statistical methods like p-values, Arnold emphasizes measuring stylistic separation using predictive classification. He uses a model called QDA (Quadratic Discriminant Analysis) to estimate how distinguishable different genres and series are. His findings show that traditional measures like shot length alone are not very effective, and instead, combining features like visual and audio elements significantly improves performance. Additionally, allowing the model to make two guesses instead of one greatly improves performance and highlights that genres can overlap.
Arnold suggests that this analysis can help us understand how genres are formed, how they change over time, and how they can overlap. The combination of visual and sonic characteristics reveals different patterns and the points at which genres intermingle. Comedy programs do not belong to a strict category, but rather, they share traits with one another. These findings help connect data analysis to a deeper understanding of the mechanics of television genres and styles.
The structure of comedy is not random, and it is not invisible. As Arnold explains, “These genre formations are not binary things, but more of a continuum”, showing that comedy genres often overlap rather than existing as strict categories. Although this research is still a work in progress, Arnold concludes that television comedy has different forms in organizing pacing, speech, and visual rhythm in distinct ways. These differences can be studied systematically using computational methods, offering new ways to understand media and genre.
Watch Taylor Arnold’s cultural analytics talk on the I School website hereTo stay in touch and join 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.