Paper accepted in ACL Findings

Large language models (LLMs) are known to exhibit gender bias, yet most evaluations focus on downstream stereotypes rather than the normative frameworks that structure model inference. We study whether LLMs rely on traditional masculinity norms (e.g. “real men are though”) as latent priors in gender-biased inference grounding evaluation in the Male Role Norms Inventory (MRNI), a validated psychological framework of prescriptive male role norms. Anchored in MRNI items, we probe models using two complementary approaches: (i) explicit Likert-style agreement with masculinity norms, and (ii) through a newly crafted English and Italian scenario-based inference dataset (MRNI-BB dataset), in which gender information and evidential support are systematically varied. Across models, explicit endorsement of masculinity norms is generally low. In contrast, in inference tasks, models systematically attribute MRNI-aligned behaviors to male agents, even when evidence is ambiguous or absent. This effect disappears when gender markers are removed, suggesting that masculinity norms are treated as gender-specific expectations about male agents. Increasing model scale reduces explicit norm endorsement but is associated with stronger male-directed bias under uncertainty.