Social Story Frames: Contextual Reasoning about Narrative Intent and Reception

A research team from institutions including Carnegie Mellon University and the University of Toronto has published a new AI framework called SocialStoryFrames, designed to computationally model the nuanced ways humans interpret stories. The work, accepted to ACL 2026, addresses a major gap: current AI models are poor at inferring an author's intent or predicting a reader's affective and evaluative response to a narrative. The researchers developed a formal taxonomy grounded in narrative theory and psychology to categorize these responses, then built two AI models to apply it.

The first model, SSF-Generator, produces plausible inferences about a story, such as perceived author intent or a reader's moral judgment. Its outputs were validated through large-scale human surveys involving 382 participants. The second, SSF-Classifier, categorizes stories according to the framework and was validated by expert annotators. To demonstrate utility, the team applied these models to the SSF-Corpus, a curated dataset of 6,140 stories from diverse social media contexts like Reddit and Twitter.

This pilot analysis allowed the researchers to characterize the frequency of different storytelling intents and compare narrative practices across online communities at scale. By linking fine-grained, context-sensitive AI modeling with a generic taxonomy, SocialStoryFrames provides a new tool for computational social science and digital humanities, enabling systematic research into how stories function in the wilds of the internet.