Social Norms in Cinema: A Cross-Cultural Analysis of Shame, Pride and Prejudice

TL;DR

This study introduces the first cross-cultural dataset of over 10k shame/pride-related expressions drawn from 5.4k Bollywood and Hollywood movie subtitles, and analyzes how social expectations behind these emotions differ between the USA and India. It combines psychology-informed LIWC analysis with prompting-based extraction via GPT-4 to identify who experiences shame or pride, their gender, and the reasons behind these emotions, supplemented by SBERT-based thematic clustering. The results reveal culture-specific patterns, such as self-focused, past-oriented shame in Hollywood versus other-oriented, present-focused shame in Bollywood, and distinct pride norms tied to duty and family in Hollywood versus ethnolinguistic identity in Bollywood. The work provides a valuable resource for culturally aware NLP and AI systems, highlighting gender sanctions and the need for careful source-data scrutiny when learning social norms for AI alignment and interaction. It also offers a rich set of themes and validated judgments that can guide future research on cross-cultural emotion expression and AI safety in sociocultural contexts.

Abstract

Shame and pride are social emotions expressed across cultures to motivate and regulate people's thoughts, feelings, and behaviors. In this paper, we introduce the first cross-cultural dataset of over 10k shame/pride-related expressions, with underlying social expectations from ~5.4K Bollywood and Hollywood movies. We examine how and why shame and pride are expressed across cultures using a blend of psychology-informed language analysis combined with large language models. We find significant cross-cultural differences in shame and pride expression aligning with known cultural tendencies of the USA and India -- e.g., in Hollywood, shame-expressions predominantly discuss self whereas shame is expressed toward others in Bollywood. Women are more sanctioned across cultures and for violating similar social expectations.

Figures (9)

• Figure 1: An overview of our approach comprising two key steps (a) Vocabulary approach and (b) Prompting a pre-trained LLM.
• Figure 2: Pearson r for LIWC categories significantly correlated ($p<0.05$) with shame for Hollywood and Bollywood after Benjamini-Hochberg p-value correction (See Appendix for confidence intervals and p values). Note, $1^{st}$ person sing pronouns are strongly correlated with Hollywood-shame whereas the correlation with $2^{nd}$ person pronoun and social references are up to 3 times stronger in Bollywood compared to Hollywood. See Table \ref{['tab:liwc_shame_bolly']} and \ref{['tab:liwc_shame_holly']} for the complete set of correlations.
• Figure 3: Pearson r for LIWC categories significantly correlated ($p<0.05$) with pride for Hollywood and Bollywood after Benjamini-Hochberg p-value correction (See Appendix for confidence intervals and p values). Note the contrast, Achievement-related & We-centered pride in Bollywood vs Social & Self-centered pride in Hollywood. See Table \ref{['tab:liwc_pride_bolly']} and \ref{['tab:liwc_pride_holly']} for the complete set of correlations.
• Figure 4: Relative association ($\Delta$) of Bollywood and Hollywood to themes obtained from agglomerative clustering performed on shame-related norms. $*$ indicates significant difference i.e., $p<0.05$. See Table \ref{['tab:shame_statTest']} for statistics and p-value.
• Figure 5: Relative association ($\Delta$) of Bollywood and Hollywood to themes obtained from agglomerative clustering performed on pride-related norms. $*$ indicates significant difference i.e., $p<0.05$. See Table \ref{['tab:pride_statTest']} for statistics and p-value.