Evaluations of social bias in LLMs largely focus on whether models generate or imply biased content. However, as LLMs are increasingly used as judges of bias, they may exhibit social biases in subtler ways in how they evaluate biased content, which current methods do not systematically capture. We call this second-order bias: social bias in an LLM’s judgment about social bias, which we evaluate through a novel, philosophically grounded reasoning task. Drawing on entitlement epistemology, we conceptualize bias as misplaced foundational knowledge that shapes an agent’s rational inquiry, and derive a logical reasoning task for LLMs to judge to whom a biased text is acceptable or non-acceptable. We develop two simple metrics to measure how biased LLM judges are in inferring demographics for acceptability without sufficient support, and how these inferences vary across groups targeted by biased texts. Evaluating open and closed models, we find that our task evades safety guardrails by surfacing bias in model judgment. It varies systematically across target groups, reflects implicit social maps, and shows how models are still triggered by demographic labels. Our work points to the need for LLM bias evaluation in judgment tasks and broadly, for more theoretically grounded approaches to bias evaluation in NLP. We release our code and model responses at https://github.com/uofthcdslab/second-order-bias.

The discourse around toxicity and LLMs in NLP largely revolves around detection tasks. This work shifts the focus to evaluating LLMs’ reasoning about toxicity—from their explanations that justify a stance—to enhance their trustworthiness in downstream tasks. Despite extensive research on explainability, it is not straightforward to adopt existing methods to evaluate free-form toxicity explanation due to their over-reliance on input text perturbations, among other challenges. To account for these, we propose a novel, theoretically-grounded multi-dimensional criterion, Argument-based Consistency (ArC), that measures the extent to which LLMs’ free-form toxicity explanations reflect an ideal and logical argumentation process. Based on uncertainty quantification, we develop six metrics for ArC to comprehensively evaluate the (in)consistencies in LLMs’ toxicity explanations. We conduct several experiments on three Llama models (of size up to 70B) and an 8B Ministral model on five diverse toxicity datasets. Our results show that while LLMs generate plausible explanations to simple prompts, their reasoning about toxicity breaks down when prompted about the nuanced relations between the complete set of reasons, the individual reasons, and their toxicity stances, resulting in inconsistent and irrelevant responses. We open-source our code and LLM-generated explanations for future works.

Though large language models (LLMs) are increasingly used in multilingual contexts, their political and sociocultural biases in low-resource languages remain critically underexplored. In this paper, we investigate how LLM-generated texts in Bengali shift in response to personas with varying political orientations (left vs. right), religious identities (Hindu vs. Muslim), and national affiliations (Bangladeshi vs. Indian). In a quasi-experimental study, we simulate these personas and prompt an LLM to respond to political discussions. Measuring the shifts relative to responses for a baseline Bengali persona, we examined how political orientation influences LLM outputs, how topical association shape the political leanings of outputs, and how demographic persona-induced changes align with differently politically oriented variations. Our findings highlight left-leaning political bias in Bengali text generation and its significant association with Muslim sociocultural and demographic identity. We also connect our findings with broader discussions around emancipatory politics, epistemological considerations, and alignment of multilingual models.

Critical studies found NLP systems to bias based on gender and racial identities. However, few studies focused on identities defined by cultural factors like religion and nationality. Compared to English, such research efforts are even further limited in major languages like Bengali due to the unavailability of labeled datasets. This paper describes a process for developing a bias evaluation dataset highlighting cultural influences on identity. We also provide a Bengali dataset as an artifact outcome that can contribute to future critical research.