Reconstructing Gender Values from Film: Digital Twins of Fictional Characters as Surveyable Agents

Vivienne Bihe Chi, Reyhan Jamalova, Lyle Ungar, Sharath Chandra Guntuku

Keywords: LLM Agents, Social Reasoning, Social Simulation, Narrative Analysis, Gender Attitudes, Digital Twins, Survey Emulation

Abstract: Do fictional narratives encode gender attitudes that resemble those held by real populations at the same historical moment? We present a proof-of-concept framework for measuring gender values as portrayed in narrative media by turning fictional film characters into surveyable LLM agents. Using movie scripts from 160 U.S. films (1990–2019), we build character ``digital twins'' grounded in dialogue and scene descriptions, condense their personas through expert-style reflections, and simulate responses to gender-attitude items from the World Values Survey. The resulting agents reproduce systematic gender differences, indicating that narrative context encodes attitudinal signals. However, compared to historical survey data, simulated responses exaggerate gender gaps and show greater volatility. These findings demonstrate both the promise and limitations of using narrative archives to measure culturally portrayed values at scale.

Paper Format Check: Yes

Submission Type: Short Paper (non-archival)

Attendance: I plan to present in person (strongly encouraged)

Presenter Name: Vivienne Bihe Chi

Presenter Email: [e-mail]

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 12