Sentiment Analysis (SA) models harbor inherent social biases that can be harmful in real-world applications. These biases are identified by examining the output of SA models for sentences that only vary in the identity groups of the subjects. Constructing natural, linguistically rich, relevant, and diverse sets of sentences that provide sufficient coverage over the domain is expensive, especially when addressing a wide range of biases: it requires domain experts and/or crowd-sourcing. In this paper, we present a novel bias testing framework, BTC-SAM, which generates high-quality test cases for bias testing in SA models with minimal specification using Large Language Models (LLMs) for the controllable generation of test sentences. Our experiments show that relying on LLMs can provide high linguistic variation and diversity in the test sentences, thereby offering better test coverage compared to base prompting methods even for previously unseen biases.

Large Language Models (LLMs) frequently reproduce the gender- and sexual-identity prejudices embedded in their training corpora, leading to outputs that marginalize LGBTQIA+ users. Hence, reducing such biases is of great importance. To achieve this, we evaluate two parameter-efficient fine-tuning (PEFT) techniques - Low-Rank Adaptation (LoRA) and soft-prompt tuning - as lightweight alternatives to full-model fine-tuning for mitigating such biases. Using the WinoQueer benchmark, we quantify bias in three open-source LLMs and observe baseline bias scores reaching up to 98 (out of 100) across a range of queer identities defined by gender and/or sexual orientation, where 50 would indicate neutrality. Fine-tuning with LoRA (< 0.1% additional parameters) on a curated QueerNews corpus reduces those scores by up to 50 points and raises neutrality from virtually 0% to as much as 36%. Soft-prompt tuning (10 virtual tokens) delivers only marginal improvements. These findings show that LoRA can deliver meaningful fairness gains with minimal computation. We advocate broader adoption of community-informed PEFT, the creation of larger queer-authored corpora, and richer evaluation suites beyond WinoQueer, coupled with ongoing audits to keep LLMs inclusive.

In an effort to mitigate the harms of large language models (LLMs), learning from human feedback (LHF) has been used to steer LLMs towards outputs that are intended to be both less harmful and more helpful. Despite the widespread adoption of LHF in practice, the quality of this feedback and its effectiveness as a safety mitigation technique remain unclear. This study addresses these issues by auditing the widely-used Helpful and Harmless (HH) dataset by Anthropic. Our work includes: (1) a thorough investigation of the dataset's content through both manual and automated evaluation; (2) experiments demonstrating the dataset's impact on models' safety; and (3) an analysis of the 100 most influential papers citing this dataset. Through our audit, we showcase how conceptualization failures and quality issues identified in the HH dataset can create additional harms by leading to disparate safety behaviors across demographic groups. Our findings highlight the need for more nuanced, context-sensitive approaches to safety mitigation in LLMs.

LLMs are increasingly transforming our digital ecosystem, but they often inherit societal biases learned from their training data, for instance stereotypes associating certain attributes with specific identity groups. While whether and how these biases are mitigated may depend on the specific use cases, being able to effectively detect instances of stereotype perpetuation is a crucial first step. Current methods to assess presence of stereotypes in generated language rely on simple template or co-occurrence based measures, without accounting for the variety of sentential contexts they manifest in. We argue that understanding the sentential context is crucial for detecting instances of generalization. We distinguish two types of generalizations: (1) language that merely mentions the presence of a generalization ("people think the French are very rude"), and (2) language that reinforces such a generalization ("as French they must be rude"), from non-generalizing context ("My French friends think I am rude"). For meaningful stereotype evaluations, we need to reliably distinguish such instances of generalizations. We introduce the new task of detecting generalization in language, and build GeniL, a multilingual dataset of over 50K sentences from 9 languages (English, Arabic, Bengali, Spanish, French, Hindi, Indonesian, Malay, and Portuguese) annotated for instances of generalizations. We demonstrate that the likelihood of a co-occurrence being an instance of generalization is usually low, and varies across different languages, identity groups, and attributes. We build classifiers to detect generalization in language with an overall PR-AUC of 58.7, with varying degrees of performance across languages. Our research provides data and tools to enable a nuanced understanding of stereotype perpetuation, a crucial step towards more inclusive and responsible language technologies.

The pervasive influence of social biases in language data has sparked the need for benchmark datasets that capture and evaluate these biases in Large Language Models (LLMs). Existing efforts predominantly focus on English language and the Western context, leaving a void for a reliable dataset that encapsulates India's unique socio-cultural nuances. To bridge this gap, we introduce IndiBias, a comprehensive benchmarking dataset designed specifically for evaluating social biases in the Indian context. We filter and translate the existing CrowS-Pairs dataset to create a benchmark dataset suited to the Indian context in Hindi language. Additionally, we leverage LLMs including ChatGPT and InstructGPT to augment our dataset with diverse societal biases and stereotypes prevalent in India. The included bias dimensions encompass gender, religion, caste, age, region, physical appearance, and occupation. We also build a resource to address intersectional biases along three intersectional dimensions. Our dataset contains 800 sentence pairs and 300 tuples for bias measurement across different demographics. The dataset is available in English and Hindi, providing a size comparable to existing benchmark datasets. Furthermore, using IndiBias we compare ten different language models on multiple bias measurement metrics. We observed that the language models exhibit more bias across a majority of the intersectional groups.

Machine learning models can perpetuate unintended biases from unfair and imbalanced datasets. Evaluating and debiasing these datasets and models is especially hard in text datasets where sensitive attributes such as race, gender, and sexual orientation may not be available. When these models are deployed into society, they can lead to unfair outcomes for historically underrepresented groups. In this paper, we present a dataset coupled with an approach to improve text fairness in classifiers and language models. We create a new, more comprehensive identity lexicon, TIDAL, which includes 15,123 identity terms and associated sense context across three demographic categories. We leverage TIDAL to develop an identity annotation and augmentation tool that can be used to improve the availability of identity context and the effectiveness of ML fairness techniques. We evaluate our approaches using human contributors, and additionally run experiments focused on dataset and model debiasing. Results show our assistive annotation technique improves the reliability and velocity of human-in-the-loop processes. Our dataset and methods uncover more disparities during evaluation, and also produce more fair models during remediation. These approaches provide a practical path forward for scaling classifier and generative model fairness in real-world settings.

The unstructured nature of data used in foundation model development is a challenge to systematic analyses for making data use and documentation decisions. From a Responsible AI perspective, these decisions often rely upon understanding how people are represented in data. We propose a framework designed to guide analysis of human representation in unstructured data and identify downstream risks. We apply the framework in two toy examples using the Common Crawl web text corpus (C4) and LAION-400M. We also propose a set of hypothetical action steps in service of dataset use, development, and documentation.

Online communities of involuntary celibates (incels) are a prominent source of misogynist hate speech. In this paper, we use quantitative text and network analysis approaches to examine how identity groups are discussed on incels-dot-is, the largest black-pilled incels forum. We find that this community produces a wide range of novel identity terms and, while terms for women are most common, mentions of other minoritized identities are increasing. An analysis of the associations made with identity groups suggests an essentialist ideology where physical appearance, as well as gender and racial hierarchies, determine human value. We discuss implications for research into automated misogynist hate speech detection.