Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MINDS: A Cross-cultural Dialogue Corpus for Social Norm Classification and Adherence Detection

Pritish Sahu, Anirudh Som, Dimitra Vergyri, Ajay Divakaran

TL;DR

This work tackles the challenge of modeling culturally grounded social norms in multilingual, multi-turn dialogue by introducing Norm-RAG, a retrieval-augmented agentic framework that grounds turn-level norm inference in structured normative documentation. It also provides MINDS, a bilingual corpus of Mandarin-English and Spanish-English conversations, annotated for norm categories and adherence to enable cross-cultural evaluation. The approach decomposes norms into four interpretable attributes (CI, IF, LF, CTC), uses semantic chunking to build a semantically coherent norm index, and employs retrieval, re-ranking, and an agentic loop to perform robust, context-aware norm classification and adherence detection. Empirical results show that Norm-RAG improves norm detection and adherence accuracy across multiple LLMs and demonstrates strong generalization, with ablations validating the contributions of semantic chunking, attribute extraction, and feedback-driven reasoning. The work advances socially intelligent dialogue systems capable of culturally aware interpretation and response, and sets the stage for future exploration of higher-order norm dynamics in multilingual, multi-agent settings.

Abstract

Social norms are implicit, culturally grounded expectations that guide interpersonal communication. Unlike factual commonsense, norm reasoning is subjective, context-dependent, and varies across cultures, posing challenges for computational models. Prior works provide valuable normative annotations but mostly target isolated utterances or synthetic dialogues, limiting their ability to capture the fluid, multi-turn nature of real-world conversations. In this work, we present Norm-RAG, a retrieval-augmented, agentic framework for nuanced social norm inference in multi-turn dialogues. Norm-RAG models utterance-level attributes including communicative intent, speaker roles, interpersonal framing, and linguistic cues and grounds them in structured normative documentation retrieved via a novel Semantic Chunking approach. This enables interpretable and context-aware reasoning about norm adherence and violation across multilingual dialogues. We further introduce MINDS (Multilingual Interactions with Norm-Driven Speech), a bilingual dataset comprising 31 multi-turn Mandarin-English and Spanish-English conversations. Each turn is annotated for norm category and adherence status using multi-annotator consensus, reflecting cross-cultural and realistic norm expression. Our experiments show that Norm-RAG improves norm detection and generalization, demonstrates improved performance for culturally adaptive and socially intelligent dialogue systems.

MINDS: A Cross-cultural Dialogue Corpus for Social Norm Classification and Adherence Detection

TL;DR

This work tackles the challenge of modeling culturally grounded social norms in multilingual, multi-turn dialogue by introducing Norm-RAG, a retrieval-augmented agentic framework that grounds turn-level norm inference in structured normative documentation. It also provides MINDS, a bilingual corpus of Mandarin-English and Spanish-English conversations, annotated for norm categories and adherence to enable cross-cultural evaluation. The approach decomposes norms into four interpretable attributes (CI, IF, LF, CTC), uses semantic chunking to build a semantically coherent norm index, and employs retrieval, re-ranking, and an agentic loop to perform robust, context-aware norm classification and adherence detection. Empirical results show that Norm-RAG improves norm detection and adherence accuracy across multiple LLMs and demonstrates strong generalization, with ablations validating the contributions of semantic chunking, attribute extraction, and feedback-driven reasoning. The work advances socially intelligent dialogue systems capable of culturally aware interpretation and response, and sets the stage for future exploration of higher-order norm dynamics in multilingual, multi-agent settings.

Abstract

Social norms are implicit, culturally grounded expectations that guide interpersonal communication. Unlike factual commonsense, norm reasoning is subjective, context-dependent, and varies across cultures, posing challenges for computational models. Prior works provide valuable normative annotations but mostly target isolated utterances or synthetic dialogues, limiting their ability to capture the fluid, multi-turn nature of real-world conversations. In this work, we present Norm-RAG, a retrieval-augmented, agentic framework for nuanced social norm inference in multi-turn dialogues. Norm-RAG models utterance-level attributes including communicative intent, speaker roles, interpersonal framing, and linguistic cues and grounds them in structured normative documentation retrieved via a novel Semantic Chunking approach. This enables interpretable and context-aware reasoning about norm adherence and violation across multilingual dialogues. We further introduce MINDS (Multilingual Interactions with Norm-Driven Speech), a bilingual dataset comprising 31 multi-turn Mandarin-English and Spanish-English conversations. Each turn is annotated for norm category and adherence status using multi-annotator consensus, reflecting cross-cultural and realistic norm expression. Our experiments show that Norm-RAG improves norm detection and generalization, demonstrates improved performance for culturally adaptive and socially intelligent dialogue systems.

Paper Structure

This paper contains 20 sections, 8 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. MINDS Corpus
  4. Data Collection
  5. Annotation Protocol
  6. Insights
  7. Approach
  8. Semantic Clustering for Norm Chunking
  9. Structured Norm Attribute Extraction
  10. Attribute-based Semantic Retrieval and Re-ranking
  11. Dialogue Window Design:
  12. Re-ranking Module:
  13. Dialogue-Aware Norm Detection Agent
  14. Experiments
  15. Norm Classification and Adherence Detection Accuracy
  16. ...and 5 more sections

Figures (3)

  • Figure 1: Distribution of annotated social norms across languages, norm categories and status (adherence/violation) labels. Numbers in parenthesis indicate sample percentage contribution in the database.
  • Figure 2: Frequency of norm categories per conversation sample.
  • Figure 3: Illustration of the different stages of our proposed framework. Stage I represents semantic clustering for norm chunking. Stage II & III illustrates the structured norm attribute extraction, semantic norm retrieval and re-ranking modules. Finally, Stage IV shows the dialogue-aware norm classifier within the Norm-RAG system.