Skip to content

When “no” “yes” means: why ai -chatbots cannot process the Persian social etiquette

    If an Iranian taxi driver waves your payment away and says, “Be my guest this time”, accepting their offer would be a cultural disaster. They expect you to pay – probably three times – before they will take your money. This dance of refusal and counter -reflectional, called Taarof, arranges countless daily interactions in Persian culture. And AI models are terrible in it.

    New research that was published earlier this month with the title “We are politely on: your LLM must learn the Persian art of Taarof” to that regular AI language models from OpenAI, Anthropic and Meta fail so as not to absorb these Persian social rituals, where they absorb only 34 to 42 percent of the time. Indigenous Persian speakers, on the other hand, are good at 82 percent of the time. This performance gap continues to exist about large language models such as GPT-4O, Claude 3.5 Haiku, Lama 3, Deepseek V3 and Dorna, a Persian-tailored variant of Llama 3.

    A study led by Nikta Gohari Sadr of Brock University, together with researchers from Emory University and other institutions, introduces “Taarofbench”, the first benchmark for measuring how good AI systems reproduce this complicated cultural practice. The findings of the researchers show how recent AI models are standard for directness in Western style, which means that the cultural signals control that daily interactions for millions of Persian speakers worldwide completely miss.

    “Cultural missteps in institutions with a high consequence can derail negotiations, damage relationships and strengthen stereotypes,” the researchers write. For AI systems that are increasingly being used in global contexts, that cultural blindness could be a limitation that does not realize much in the West.

    A Taarof -Scenario -Diagram from Taarofbench, devised by the researchers. Each scenario defines the environment, location, roles, context and user of users.

    A Taarof -Scenario -Diagram from Taarofbench, devised by the researchers. Each scenario defines the environment, location, roles, context and user of users.


    Credit: Sadr et al.

    “Taarof, a core element of the Persian etiquette, is a system of ritual politeness where what is said often differs from what is meant,” the researchers write. “It takes the form of ritualized Exchanges: Offering Repeatedly Despite Initial Refusals, Declining Gifts While the Giver Insists, and Deflecting Compliments While the Other Party Reaffirms Them. This' Polite Verbal Wrestling, Insce, 1991) Involves Resistance, which Shapes Everyday Interactions in Iranian Culture, Creating Implicit Rules for How Generosity, Gratitude, and requests are expressed. “