Unmasking the Factual-Conceptual Gap in Persian Language Models

TL;DR

This paper evaluates Persian language models on a new benchmark, DivanBench, revealing significant biases and limitations in reasoning about cultural norms, which are not improved by increased pretraining.

Contribution

Introduces DivanBench, a diagnostic benchmark for Persian NLP focusing on social norms and cultural knowledge, and analyzes the shortcomings of current models in this domain.

Findings

Models exhibit severe acquiescence bias.

Pretraining amplifies bias and degrades reasoning.

Performance gap between factual retrieval and scenario application.

Abstract

While emerging Persian NLP benchmarks have expanded into pragmatics and politeness, they rarely distinguish between memorized cultural facts and the ability to reason about implicit social norms. We introduce DivanBench, a diagnostic benchmark focused on superstitions and customs, arbitrary, context-dependent rules that resist simple logical deduction. Through 315 questions across three task types (factual retrieval, paired scenario verification, and situational reasoning), we evaluate seven Persian LLMs and reveal three critical failures: most models exhibit severe acquiescence bias, correctly identifying appropriate behaviors but failing to reject clear violations; continuous Persian pretraining amplifies this bias rather than improving reasoning, often degrading the model's ability to discern contradictions; and all models show a 21\% performance gap between retrieving factual knowledge and applying it in scenarios. These findings demonstrate that cultural competence requires more than scaling monolingual data, as current models learn to mimic cultural patterns without internalizing the underlying schemas.