We study fine-grained stereotypes of LLMs activated by linguistic features representing demographic groups, developing a probe-based framework to extract 235 fine-grained traits (e.g., friendly, uneducated, honest) from model representations of prompts. Using these probes, we conduct two studies analyzing the underlying structure of traits and specific stereotypes of demographic groups. We find that fine-grained trait information is represented in LLMs similarly to coarse stereotype content dimensions (e.g., warmth and competence), and that the latent factors of trait probes (e.g., morality, competence, sociability) largely align with those in the psychology literature (Fiske et al., 2002) despite traits being modeled independently. Finally, we extract LLMs’ associations between traits and demographics signaled through language to surface granular stereotypes encoded in LLMs. We, for example, find that the evaluated models encode associations between African American Language and traits clumsy, uneducated, and criminal. These associations may reinforce harmful stereotypes, with implications for both general dialogue responses and in domains like hiring. Overall, we discuss the need to consider fine-grained traits in stereotype evaluations in order to distinguish stereotypes of out-groups and between models.

Dense retrievers decide which documents reach users and the language models that use them, yet they are typically evaluated with neutral queries. We ask whether the identity signals that real users express in their queries—political ideology and dialect—bias what a retriever returns. We design evaluations in two domains, political news and consumer-health questions, each pairing a controlled synthetic set that varies only the identity signal with naturalistic queries. Across five dense retrievers and a sparse baseline, every retriever (i) retrieves articles that align with the query’s own political lean and (ii) performs worse for questions written in African American Language (AAL) than in White Mainstream English (WME). Two analyses tie these gaps to queries’ identity signals beyond surface vocabulary: partialling out an aggregate lexical-asymmetry score leaves the synthetic gaps largely intact, and linear probes recover lean and dialect from the retrievers’ query embeddings beyond token-level features. Left unaddressed, such retrieval biases risk contributing to polarization and reinforcing the health disparities already faced by AAL speakers.

In our interdisciplinary, collaborative research, we have found that humanities and social science researchers would like to analyze data underlying LLMs as well as study how language tools (e.g., emotion detection) work on their data. Existing tools are often unintuitive for such researchers and do not provide them enough agency. In this paper, we introduce CorpusLens, an LLM-based tool to aid social, humanistic, and language science researchers in exploratory textual data analysis. We present the design of CorpusLens as well as a user study showing that researchers found CorpusLens to be more helpful than prior tools they had used, while also rating it as intuitive, easy to use, transparent and enabling.

Prior work evaluating emotion and affective understanding in large language models (LLMs) typically rely on predetermined label sets or focus on a singular evaluation task (e.g., emotion detection). We consider affective states, referring to the much broader variety of terms people use to label their emotional experiences. We evaluate multilingual language models’ understanding of affective states in English and Spanish through three different tasks: 1) identification, where models predict an affective state given text, 2) expression, where models generate text expressing a given affective state, and 3) verification, where models report whether a given term refers to an affective state. We show that performance on one task is not necessarily predictive of performance on another. Using these three tasks, we then begin to explore when and why models struggle to understand particular affective states compared to others. We examine systematic patterns in the affective state terms that are well and poorly understood by models, characterizing the working emotion vocabulary of LLMs.

Large Language Models (LLMs) have achieved impressive performance in text summarization and are increasingly deployed in real-world applications. However, these systems often inherit associative and framing biases from pre-training data, leading to inappropriate or unfair outputs in downstream tasks. In this work, we present AdvSumm (Adversarial Summarization), a domain-agnostic training framework designed to mitigate bias in text summarization through improved generalization. Inspired by adversarial robustness, AdvSumm introduces a novel Perturber component that applies gradient-guided perturbations at the embedding level of Sequence-to-Sequence models, enhancing the model’s robustness to input variations. We empirically demonstrate that AdvSumm effectively reduces different types of bias in summarization—specifically, name-nationality bias and political framing bias—without compromising summarization quality. Compared to standard transformers and data augmentation techniques like back-translation, AdvSumm achieves stronger bias mitigation performance across benchmark datasets.

LLMs can be unpredictable, as even slight alterations to the prompt can cause the output to change in unexpected ways. Thus, the ability of models to accurately explain their behavior is critical, especially in high-stakes settings. Counterfactual simulatability measures how well an explanation allows users to infer the model’s output on related counterfactuals and has been previously studied for yes/no question answering. We provide a general framework for extending this method to generation tasks, using news summarization and medical suggestion as example use cases. We find that while LLM explanations do enable users to better predict their outputs on counterfactuals in the summarization setting, there is significant room for improvement for medical suggestion. Furthermore, our results suggest that evaluating counterfactual simulatability may be more appropriate for skill-based tasks as opposed to knowledge-based tasks.

Preference alignment via reward models helps build safe, helpful, and reliable large language models (LLMs). However, subjectivity in preference judgments and the lack of representative sampling in preference data collection can introduce new biases, hindering reward models’ fairness and equity. In this work, we introduce a framework for evaluating dialect biases in reward models and conduct a case study on biases against African American Language (AAL) through several experiments comparing reward model preferences and behavior on paired White Mainstream English (WME) and both machine-translated and human-written AAL corpora. We show that reward models are less aligned with human preferences when processing AAL texts vs. WME ones (-4% accuracy on average), frequently disprefer AAL-aligned texts vs. WME-aligned ones, and steer conversations toward WME, even when prompted with AAL texts. Our findings provide a targeted analysis of anti-AAL biases at a relatively understudied stage in LLM development, highlighting representational harms and ethical questions about the desired behavior of LLMs concerning AAL.

Generating unbiased summaries in real-world settings such as political perspective summarization remains a crucial application of Large Language Models (LLMs). Yet, existing evaluation frameworks rely on traditional metrics for measuring key attributes such as coverage and faithfulness without verifying their applicability, and efforts to develop improved summarizers are still nascent. We address these gaps by (1) identifying reliable metrics for measuring perspective summary quality, and (2) investigating the efficacy of LLM-based methods beyond zero-shot inference. Namely, we build a test set for benchmarking metric reliability using human annotations and show that traditional metrics underperform compared to language model–based metrics, which prove to be strong evaluators. Using these metrics, we show that reranking-based methods yield strong results, and preference tuning with synthetically generated and reranking-labeled data further boosts performance. Our findings aim to contribute to the reliable evaluation and development of perspective summarization methods.

We introduce and study artificial impressions–patterns in LLMs’ internal representations of prompts that resemble human impressions and stereotypes based on language. We fit linear probes on generated prompts to predict impressions according to the two-dimensional Stereotype Content Model (SCM). Using these probes, we study the relationship between impressions and downstream model behavior as well as prompt features that may inform such impressions. We find that LLMs inconsistently report impressions when prompted, but also that impressions are more consistently linearly decodable from their hidden representations. Additionally, we show that artificial impressions of prompts are predictive of the quality and use of hedging in model responses. We also investigate how particular content, stylistic, and dialectal features in prompts impact LLM impressions.

Global partisan hostility and polarization has increased, and this polarization is heightened around presidential elections. Models capable of generating accurate summaries of diverse perspectives can help reduce such polarization by exposing users to alternative perspectives. In this work, we introduce a novel dataset and task for independently summarizing each political perspective in a set of passages from opinionated news articles. For this task, we propose a framework for evaluating different dimensions of perspective summary performance. We benchmark 11 summarization models and LLMs of varying sizes and architectures through both automatic and human evaluation. While recent models like GPT-4o perform well on this task, we find that all models struggle to generate summaries that are faithful to the intended perspective. Our analysis of summaries focuses on how extraction behavior is impacted by features of the input documents.

With a combination of quantitative experiments, human judgments, and qualitative analyses, we evaluate the quantity and quality of African American Language (AAL) representation in 12 predominantly English, open-source pretraining corpora. We specifically focus on the sources, variation, and naturalness of included AAL texts representing the AAL speaking community. We find that AAL is underrepresented in all evaluated pretraining corpora compared to US demographics, constituting as few as 0.007% and at most 0.18% of documents. We also find that more than 25% of AAL texts in C4 may be perceived as inappropriate for LLMs to generate and to reinforce harmful stereotypes. Finally, we find that most automated filters are more likely to conserve White Mainstream English (WME) texts over AAL in pretraining corpora.

In the field of emotion analysis, much NLP research focuses on identifying a limited number of discrete emotion categories, often applied across languages. These basic sets, however, are rarely designed with textual data in mind, and culture, language, and dialect can influence how particular emotions are interpreted. In this work, we broaden our scope to a practically unbounded set of affective states, which includes any terms that humans use to describe their experiences of feeling. We collect and publish MASIVE, a dataset of Reddit posts in English and Spanish containing over 1,000 unique affective states each. We then define the new problem of affective state identification for language generation models framed as a masked span prediction task. On this task, we find that smaller finetuned multilingual models outperform much larger LLMs, even on region-specific Spanish affective states. Additionally, we show that pretraining on MASIVE improves model performance on existing emotion benchmarks. Finally, through machine translation experiments, we find that native speaker-written data is vital to good performance on this task.

While biases disadvantaging African American Language (AAL) have been uncovered in models for tasks such as speech recognition and toxicity detection, there has been little investigation of these biases for language generation models like ChatGPT. We evaluate how well LLMs understand AAL in comparison to White Mainstream English (WME), the encouraged “standard” form of English taught in American classrooms. We measure large language model performance on two tasks: a counterpart generation task, where a model generates AAL given WME and vice versa, and a masked span prediction (MSP) task, where models predict a phrase hidden from their input. Using a novel dataset of AAL texts from a variety of regions and contexts, we present evidence of dialectal bias for six pre-trained LLMs through performance gaps on these tasks.