Search Papers | Poster Sessions | All Posters
Poster C47 in Poster Session C - Friday, August 9, 2024, 11:15 am – 1:15 pm, Johnson Ice Rink
Event similarity and word-level salience predict how humans summarize information from complex naturalistic narratives
Claire Sun1 (), Coraline Rinn Iordan1; 1University of Rochester
We continuously encounter multitude sources of complex, multisensory information. To navigate this deluge, we summarize the contents of our experiences into manageable chunks that help us encode them into memory. Here, we seek to identify the features of complex narratives that predict how information will be summarized, and thereby provide insights into the cognitive mechanisms of summarization. We show that (1) transformer models specifically trained for document summarization generate internal features that are relevant for how humans summarize information from similar content; and (2) that the process of summarizing complex narratives can be described partially through its interaction with how narratives are segmented into individual events.
Keywords: Transformers Event Perception Summarization Memory