Several recent studies have shown that strong natural language understanding (NLU) models are prone to relying on unwanted dataset biases without learning the underlying task, resulting in models that fail to generalize to out-of-domain datasets and are likely to perform poorly in real-world scenarios. We propose two learning strategies to train neural models, which are more robust to such biases and transfer better to out-of-domain datasets. The biases are specified in terms of one or more bias-only models, which learn to leverage the dataset biases. During training, the bias-only models' predictions are used to adjust the loss of the base model to reduce its reliance on biases by down-weighting the biased examples and focusing the training on the hard examples. We experiment on large-scale natural language inference and fact verification benchmarks, evaluating on out-of-domain datasets that are specifically designed to assess the robustness of models against known biases in the training data. Results show that our debiasing methods greatly improve robustness in all settings and better transfer to other textual entailment datasets. Our code and data are publicly available in \url{https://github.com/rabeehk/robust-nli}.
University of Oxford（牛津大学）；Rural Senses Ltd.（Rural Senses有限公司）；University of Cambridge（剑桥大学）
Most well-established data collection methods currently adopted in NLP depend on the assumption of speaker literacy. Consequently, the collected corpora largely fail to represent swathes of the global population, which tend to be some of the most vulnerable and marginalised people in society, and often live in rural developing areas. Such underrepresented groups are thus not only ignored when making modeling and system design decisions, but also prevented from benefiting from development outcomes achieved through data-driven NLP. This paper aims to address the under-representation of illiterat…
University of Qom（库姆大学）；Tehran Institute for Advanced Studies（德黑兰高等研究院）；Khatam University（哈塔姆大学）
Existing techniques for mitigating dataset bias often leverage a biased model to identify biased instances. The role of these biased instances is then reduced during the training of the main model to enhance its robustness to out-of-distribution data. A common core assumption of these techniques is that the main model handles biased instances similarly to the biased model, in that it will resort to biases whenever available. In this paper, we show that this assumption does not hold in general. We carry out a critical investigation on two well-known datasets in the domain, MNLI and FEVER, along…
Johannes Kepler University（约翰内斯·开普勒大学）；EPFL（洛桑联邦理工学院）；Idiap Research Institute（IDIAP研究所）；TU Wien（维也纳工业大学）；Complexity Science Hub Vienna（维也纳复杂性科学中心）
Text corpora are widely used resources for measuring societal biases and stereotypes. The common approach to measuring such biases using a corpus is by calculating the similarities between the embedding vector of a word (like nurse) and the vectors of the representative words of the concepts of interest (such as genders). In this study, we show that, depending on what one aims to quantify as bias, this commonly-used approach can introduce non-relevant concepts into bias measurement. We propose an alternative approach to bias measurement utilizing the smoothed first-order co-occurrence relation…
Princeton University（普林斯顿大学）；University of Bath（巴斯大学）
Artificial intelligence and machine learning are in a period of astounding growth. However, there are concerns that these technologies may be used, either with or without intention, to perpetuate the prejudice and unfairness that unfortunately characterizes many human institutions. Here we show for the first time that human-like semantic biases result from the application of standard machine learning to ordinary language---the same sort of language humans are exposed to every day. We replicate a spectrum of standard human biases as exposed by the Implicit Association Test and other well-known …
Fair Play in the Newsroom: Actor-Based Filtering Gender Discrimination in Text Corpora
Faculty for Computer Science and Mathematics, Hochschule München University of Applied Sciences（慕尼黑应用科学大学计算机科学与数学学院）；Department of Statistics, LMU Munich（慕尼黑大学统计学系）；Munich Center for Machine Learning (MCML), LMU Munich（慕尼黑大学机器学习中心）
Beijing Academy of Artificial Intelligence（北京人工智能研究院）；The University of Sheffield（谢菲尔德大学）；University of Michigan（密歇根大学）；University of Colorado Boulder（科罗拉多大学博尔德分校）；Sichuan University（四川大学）；McGill University（麦吉尔大学）
As natural language processing (NLP) for gender bias becomes a significant interdisciplinary topic, the prevalent data-driven techniques such as large-scale language models suffer from data inadequacy and biased corpus, especially for languages with insufficient resources such as Chinese. To this end, we propose a Chinese cOrpus foR Gender bIas Probing and Mitigation CORGI-PM, which contains 32.9k sentences with high-quality labels derived by following an annotation scheme specifically developed for gender bias in the Chinese context. Moreover, we address three challenges for automatic textual…
Indian Institute of Technology Kharagpur（印度理工学院克勒格布尔分校）；Université de Lorraine（洛林大学）；CNRS（法国国家科学研究中心）；Inria（法国国家信息与自动化研究所）；LORIA（洛林信息学及其应用实验室）
This work addresses the cross-corpora generalization issue for the low-resourced spoken language identification (LID) problem. We have conducted the experiments in the context of Indian LID and identified strikingly poor cross-corpora generalization due to corpora-dependent non-lingual biases. Our contribution to this work is twofold. First, we propose domain diversification, which diversifies the limited training data using different audio data augmentation methods. We then propose the concept of maximally diversity-aware cascaded augmentations and optimize the augmentation fold-factor for ef…
Data and Web Science Research Group（数据与网络科学研究组）；University of Mannheim（曼海姆大学）
Recent work has shown that distributional word vector spaces often encode human biases like sexism or racism. In this work, we conduct an extensive analysis of biases in Arabic word embeddings by applying a range of recently introduced bias tests on a variety of embedding spaces induced from corpora in Arabic. We measure the presence of biases across several dimensions, namely: embedding models (Skip-Gram, CBOW, and FastText) and vector sizes, types of text (encyclopedic text, and news vs. user-generated content), dialects (Egyptian Arabic vs. Modern Standard Arabic), and time (diachronic anal…
Masked language models pick up gender biases during pre-training. Such biases are usually attributed to a certain model architecture and its pre-training corpora, with the implicit assumption that other variations in the pre-training process, such as the choices of the random seed or the stopping point, have no effect on the biases measured. However, we show that severe fluctuations exist at the fundamental level of individual templates, invalidating the assumption. Further against the intuition of how humans acquire biases, these fluctuations are not correlated with the certainty of the predi…
Dissecting vocabulary biases datasets through statistical testing and automated data augmentation for artifact mitigation in Natural Language Inference
University of Illinois Chicago（伊利诺伊大学芝加哥分校）；University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）；Arizona State University（亚利桑那州立大学）
Natural Language Processing (NLP) models have been found discriminative against groups of different social identities such as gender and race. With the negative consequences of these undesired biases, researchers have responded with unprecedented effort and proposed promising approaches for bias mitigation. In spite of considerable practical importance, current algorithmic fairness literature lacks an in-depth understanding of the relations between different forms of biases. Social bias is complex by nature. Numerous studies in social psychology identify the "generalized prejudice", i.e., gene…
Despite widespread use in natural language processing (NLP) tasks, word embeddings have been criticized for inheriting unintended gender bias from training corpora. programmer is more closely associated with man and homemaker is more closely associated with woman. Such gender bias has also been shown to propagate in downstream tasks.
Bias Evaluation and Mitigation in Retrieval-Augmented Medical Question-Answering Systems
University of Pittsburgh（匹兹堡大学）；School of Computing and Information, University of Pittsburgh（匹兹堡大学计算与信息学院）；Clinical and Translational Science Institute, University of Pittsburgh（匹兹堡大学临床与转化科学研究所）；University of Pittsburgh Medical Center（匹兹堡大学医学中心）
It is well known that textual data on the internet and other digital platforms contain significant levels of bias and stereotypes. Although many such texts contain stereotypes and biases that inherently exist in natural language for reasons that are not necessarily malicious, there are crucial reasons to mitigate these biases. For one, these texts are being used as training corpus to train language models for salient applications like cv-screening, search engines, and chatbots; such applications are turning out to produce discriminatory results. Also, several research findings have concluded t…
The University of Tokyo（东京大学）；National Institute of Informatics（国立信息学研究所）
Debiasing language models from unwanted behaviors in Natural Language Understanding tasks is a topic with rapidly increasing interest in the NLP community. Spurious statistical correlations in the data allow models to perform shortcuts and avoid uncovering more advanced and desirable linguistic features. A multitude of effective debiasing approaches has been proposed, but flexibility remains a major issue. For the most part, models must be retrained to find a new set of weights with debiased behavior. We propose a new debiasing method in which we identify debiased pruning masks that can be app…
Word embedding spaces are powerful tools for capturing latent semantic relationships between terms in corpora, and have become widely popular for building state-of-the-art natural language processing algorithms. However, studies have shown that societal biases present in text corpora may be incorporated into the word embedding spaces learned from them. Thus, there is an ethical concern that human-like biases contained in the corpora and their derived embedding spaces might be propagated, or even amplified with the usage of the biased embedding spaces in downstream applications. In an attempt t…
BiMa: Towards Biases Mitigation for Text-Video Retrieval via Scene Element Guidance
FPT Software AI Center（FPT软件AI中心）；Aalborg University（奥尔堡大学）；Pioneer Centre for AI（先锋人工智能中心）；University of Liverpool（利物浦大学）；AICV Lab, University of Arkansas（AICV实验室，阿肯色大学）
University of Southern California（南加州大学）；Snap Inc.（Snap 公司）
Fine-tuned language models have been shown to exhibit biases against protected groups in a host of modeling tasks such as text classification and coreference resolution. Previous works focus on detecting these biases, reducing bias in data representations, and using auxiliary training objectives to mitigate bias during fine-tuning. Although these techniques achieve bias reduction for the task and domain at hand, the effects of bias mitigation may not directly transfer to new tasks, requiring additional data collection and customized annotation of sensitive attributes, and re-evaluation of appr…
Bocconi University（博科尼大学）；Politecnico di Torino（都灵理工大学）
Natural Language Processing (NLP) models risk overfitting to specific terms in the training data, thereby reducing their performance, fairness, and generalizability. E.g., neural hate speech detection models are strongly influenced by identity terms like gay, or women, resulting in false positives, severe unintended bias, and lower performance. Most mitigation techniques use lists of identity terms or samples from the target domain during training. However, this approach requires a-priori knowledge and introduces further bias if important terms are neglected. Instead, we propose a knowledge-fr…
University of Mannheim（曼海姆大学）；Technische Universität Darmstadt（达姆施塔特工业大学）
We analyze bias in historical corpora as encoded in diachronic distributional semantic models by focusing on two specific forms of bias, namely a political (i.e., anti-communism) and racist (i.e., antisemitism) one. For this, we use a new corpus of German parliamentary proceedings, DeuPARL, spanning the period 1867--2020. We complement this analysis of historical biases in diachronic word embeddings with a novel measure of bias on the basis of term co-occurrences and graph-based label propagation. The results of our bias measurements align with commonly perceived historical trends of antisemit…
Word embeddings have recently been shown to reflect many of the pronounced societal biases (e.g., gender bias or racial bias). Existing studies are, however, limited in scope and do not investigate the consistency of biases across relevant dimensions like embedding models, types of texts, and different languages. In this work, we present a systematic study of biases encoded in distributional word vector spaces: we analyze how consistent the bias effects are across languages, corpora, and embedding models. Furthermore, we analyze the cross-lingual biases encoded in bilingual embedding spaces, i…
As natural language processing systems become more widespread, it is necessary to address fairness issues in their implementation and deployment to ensure that their negative impacts on society are understood and minimized. However, there is limited work that studies fairness using a multilingual and intersectional framework or on downstream tasks. In this paper, we introduce four multilingual Equity Evaluation Corpora, supplementary test sets designed to measure social biases, and a novel statistical framework for studying unisectional and intersectional social biases in natural language proc…
Bias in Large Language Models: Origin, Evaluation, and Mitigation
Department of Statistics, George Washington University（乔治华盛顿大学统计学系）；University of Connecticut（康奈尔大学）；Saint Elizabeth University（圣艾利泽大学）；The Pennsylvania State University（宾夕法尼亚州立大学）
UNC Chapel Hill（北卡罗来纳大学教堂山分校）；University of California, Santa Cruz（加利福尼亚大学圣克鲁兹分校）；University of British Columbia（不列颠哥伦比亚大学）
Pre-trained language models learn socially harmful biases from their training corpora, and may repeat these biases when used for generation. We study gender biases associated with the protagonist in model-generated stories. Such biases may be expressed either explicitly ("women can't park") or implicitly (e.g. an unsolicited male character guides her into a parking space). We focus on implicit biases, and use a commonsense reasoning engine to uncover them. Specifically, we infer and analyze the protagonist's motivations, attributes, mental states, and implications on others. Our findings regar…
University of Southern California（南加利福尼亚大学）；University of California, Los Angeles（加利福尼亚大学洛杉矶分校）
We present a general approach towards controllable societal biases in natural language generation (NLG). Building upon the idea of adversarial triggers, we develop a method to induce societal biases in generated text when input prompts contain mentions of specific demographic groups. We then analyze two scenarios: 1) inducing negative biases for one demographic and positive biases for another demographic, and 2) equalizing biases between demographics. The former scenario enables us to detect the types of biases present in the model. Specifically, we show the effectiveness of our approach at fa…
Position: LLMs Must Use Functor-Based and RAG-Driven Bias Mitigation for Fairness
Knight Foundation School of Computing and Information Sciences（骑士基金会计算与信息科学学院）；Florida International University（佛罗里达国际大学）；College of Engineering（工程学院）；University of South Florida（佛罗里达州立大学）
School of Computing and Information Systems（计算与信息系统学院）；University of Melbourne（墨尔本大学）
Bias is pervasive in NLP models, motivating the development of automatic debiasing techniques. Evaluation of NLP debiasing methods has largely been limited to binary attributes in isolation, e.g., debiasing with respect to binary gender or race, however many corpora involve multiple such attributes, possibly with higher cardinality. In this paper we argue that a truly fair model must consider `gerrymandering' groups which comprise not only single attributes, but also intersectional groups. We evaluate a form of bias-constrained model which is new to NLP, as well an extension of the iterative n…
