# NormGenesis: Multicultural Dialogue Generation via Exemplar-Guided Social Norm Modeling and Violation Recovery

Source: [NormGenesis: Multicultural Dialogue Generation via Exemplar-Guided Social Norm Modeling and Violation Recovery](<https://www.sota2.com/research/paper/68fa7b7f1e209131cea88b10>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-14T05:00:03.952Z

Authors: Minki Hong, Jangho Choi, Jihie Kim

Published: 2025-09-22T20:29:25.000Z

## Abstract

Social norms govern culturally appropriate behavior in communication, enabling dialogue systems to produce responses that are not only coherent but also socially acceptable. We present NormGenesis, a multicultural framework for generating and annotating socially grounded dialogues across English, Chinese, and Korean. To model the dynamics of social interaction beyond static norm classification, we propose a novel dialogue type, Violation-to-Resolution (V2R), which models the progression of conversations following norm violations through recognition and socially appropriate repair. To improve pragmatic consistency in underrepresented languages, we implement an exemplar-based iterative refinement early in the dialogue synthesis process. This design introduces alignment with linguistic, emotional, and sociocultural expectations before full dialogue generation begins. Using this framework, we construct a dataset of 10,800 multi-turn dialogues annotated at the turn level for norm adherence, speaker intent, and emotional response. Human and LLM-based evaluations demonstrate that NormGenesis significantly outperforms existing datasets in refinement quality, dialogue naturalness, and generalization performance. We show that models trained on our V2R-augmented data exhibit improved pragmatic competence in ethically sensitive contexts. Our work establishes a new benchmark for culturally adaptive dialogue modeling and provides a scalable methodology for norm-aware generation across linguistically and culturally diverse languages.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Dialogue Generation on Norm-grounded Dialogue English (test)](<https://www.sota2.com/research/sota/dialogue-generation-on-norm-grounded-dialogue-english-test.md>): Dialogue Generation; Norm-grounded Dialogue English (test); Win Rate vs. NormDial: 65.

- [Dialogue Generation on Norm-grounded Dialogue (Chinese) (test)](<https://www.sota2.com/research/sota/dialogue-generation-on-norm-grounded-dialogue-chinese-test.md>): Dialogue Generation; Norm-grounded Dialogue (Chinese) (test); Win Rate vs NormDial: 79.

- [Social Norm Dialogue Generation on Prosocial-Dialog Chinese](<https://www.sota2.com/research/sota/social-norm-dialogue-generation-on-prosocial-dialog-chinese.md>): Social Norm Dialogue Generation; Prosocial-Dialog Chinese; Human Preference Rate: 77.

- [Human Preference Evaluation on PROSOCIALDIALOG](<https://www.sota2.com/research/sota/human-preference-evaluation-on-prosocialdialog.md>): Human Preference Evaluation; PROSOCIALDIALOG; Preference: 82.

- [Social Norm Dialogue Generation on Prosocial-Dialog Korean](<https://www.sota2.com/research/sota/social-norm-dialogue-generation-on-prosocial-dialog-korean.md>): Social Norm Dialogue Generation; Prosocial-Dialog Korean; Human Preference: 68.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2509.18395>)

[PDF](<https://arxiv.org/pdf/2509.18395.pdf>)