The Art of Taarof
Modeling ritual politeness in Persian for culturally aware AI.
Overview
Taarof is a complex system of ritual politeness in Persian that governs offers, refusals, greetings, hospitality, and social negotiation. Far beyond mere etiquette, Taarof encodes hierarchy, reciprocity, and indirect communication—making it both a linguistic and cultural phenomenon. While native speakers effortlessly navigate these cues, AI systems typically fail to recognize or respond appropriately to Taarof, leading to breakdowns in user experience, misinterpretation, or even offense.
This initiative investigates how Taarof can be computationally modeled and applied across AI systems, from language learning platforms to culturally aware virtual assistants. It sits at the intersection of pragmatics, sociolinguistics, and NLP.
Methodology
In collaboration with Brock University, we are combining linguistic fieldwork with computational modeling to understand and simulate Taarof behavior:
- Corpus analysis of real-life Taarof exchanges across media, conversation, and social platforms
- Annotation schema development for identifying Taarof acts (offers, refusals, honorifics, hedging, etc.)
- Dialogue modeling using rule-based and LLM-driven approaches for detecting and generating Taarof utterances
- Evaluation of cross-cultural interpretations and system responsiveness in interactional settings
- Pilot integration into conversational AI agents for Persian language learners and chatbots
Preliminary Results
- Taarof is frequently misclassified by standard NLP systems as contradiction, negation, or non-cooperation
- Even advanced LLMs struggle to correctly disambiguate sincerity from ritual speech without contextual prompts
- Speakers consistently use layered politeness strategies (e.g., multi-turn refusals), requiring discourse-level modeling
- Annotators found that emotion, tone, and relationship status heavily influence interpretation—highlighting the need for socio-pragmatic grounding
Use Case
This work has direct applications for:
- Conversational agents and chatbots interacting with Persian speakers
- Cross-cultural communication tools and translation systems needing nuance in social interaction
- Language learning platforms aiming to teach pragmatic competence, not just grammar
- AI ethics researchers exploring how to embed cultural variation and politeness norms in dialogue systems
- Social robotics and assistive technologies operating in Persian-speaking regions
Team
Karine Megerdoomian in collaboration with Nikta Gohari Sadr (Brock Univ.), Sahar Heidariasl (Brock Univ.), Laleh Seyyed-Kalantari (York Univ.), Ali Emami (Emory Univ.)
Latest publication or presentation
(if available)