This paper presents FairLib, an open-source python library for assessing and improving model fairness. It provides a systematic framework for quickly accessing benchmark datasets, reproducing existing debiasing baseline models, developing new methods, evaluating models with different metrics, and visualizing their results. Its modularity and extensibility enable the framework to be used for diverse types of inputs, including natural language, images, and audio. We implement 14 debiasing methods, including pre-processing,at-training-time, and post-processing approaches. The built-in metrics cover the most commonly acknowledged fairness criteria and can be further generalized and customized for fairness evaluation.

Mitigating bias in training on biased datasets is an important open problem. Several techniques have been proposed, however the typical evaluation regime is very limited, considering very narrow data conditions. For instance, the effect of target class imbalance and stereotyping is under-studied. To address this gap, we examine the performance of various debiasing methods across multiple tasks, spanning binary classification (Twitter sentiment), multi-class classification (profession prediction), and regression (valence prediction). Through extensive experimentation, we find that data conditions have a strong influence on relative model performance, and that general conclusions cannot be drawn about method efficacy when evaluating only on standard datasets, as is current practice in fairness research.

NLP technologies can cause unintended harms if learned representations encode sensitive attributes of the author, or predictions systematically vary in quality across groups. Popular debiasing approaches, like adversarial training, remove sensitive information from representations in order to reduce disparate performance, however the relation between representational fairness and empirical (performance) fairness has not been systematically studied. This paper fills this gap, and proposes a novel debiasing method building on contrastive learning to encourage a latent space that separates instances based on target label, while mixing instances that share protected attributes. Our results show the effectiveness of our new method and, more importantly, show across a set of diverse debiasing methods that representational fairness does not imply empirical fairness. This work highlights the importance of aligning and understanding the relation of the optimization objective and final fairness target.

Real-world datasets often encode stereotypes and societal biases. Such biases can be implicitly captured by trained models, leading to biased predictions and exacerbating existing societal preconceptions. Existing debiasing methods, such as adversarial training and removing protected information from representations, have been shown to reduce bias. However, a disconnect between fairness criteria and training objectives makes it difficult to reason theoretically about the effectiveness of different techniques. In this work, we propose two novel training objectives which directly optimise for the widely-used criterion of equal opportunity, and show that they are effective in reducing bias while maintaining high performance over two classification tasks.

With the growing prevalence of large-scale language models, their energy footprint and potential to learn and amplify historical biases are two pressing challenges. Dataset distillation (DD) — a method for reducing the dataset size by learning a small number of synthetic samples which encode the information in the original dataset — is a method for reducing the cost of model training, however its impact on fairness has not been studied. We investigate how DD impacts on group bias, with experiments over two language classification tasks, concluding that vanilla DD preserves the bias of the dataset. We then show how existing debiasing methods can be combined with DD to produce models that are fair and accurate, at reduced training cost.

Hierarchical document categorisation is a special case of multi-label document categorisation, where there is a taxonomic hierarchy among the labels. While various approaches have been proposed for hierarchical document categorisation, there is no standard benchmark dataset, resulting in different methods being evaluated independently and there being no empirical consensus on what methods perform best. In this work, we examine different combinations of neural text encoders and hierarchical methods in an end-to-end framework, and evaluate over three datasets. We find that the performance of hierarchical document categorisation is determined not only by how the hierarchical information is modelled, but also the structure of the label hierarchy and class distribution.

Sentence ordering is the task of arranging a given bag of sentences so as to maximise the coherence of the overall text. In this work, we propose a simple yet effective training method that improves the capacity of models to capture overall text coherence based on training over pairs of sentences/segments. Experimental results show the superiority of our proposed method in in- and cross-domain settings. The utility of our method is also verified over a multi-document summarisation task.

Images are core components of multi-modal learning in natural language processing (NLP), and results have varied substantially as to whether images improve NLP tasks or not. One confounding effect has been that previous NLP research has generally focused on sophisticated tasks (in varying settings), generally applied to English only. We focus on text classification, in the context of assigning named entity classes to a given Wikipedia page, where images generally complement the text and the Wikipedia page can be in one of a number of different languages. Our experiments across a range of languages show that images complement NLP models (including BERT) trained without external pre-training, but when combined with BERT models pre-trained on large-scale external data, images contribute nothing.

While pretrained language models (LMs) have driven impressive gains over morpho-syntactic and semantic tasks, their ability to model discourse and pragmatic phenomena is less clear. As a step towards a better understanding of their discourse modeling capabilities, we propose a sentence intrusion detection task. We examine the performance of a broad range of pretrained LMs on this detection task for English. Lacking a dataset for the task, we introduce INSteD, a novel intruder sentence detection dataset, containing 170,000+ documents constructed from English Wikipedia and CNN news articles. Our experiments show that pretrained LMs perform impressively in in-domain evaluation, but experience a substantial drop in the cross-domain setting, indicating limited generalization capacity. Further results over a novel linguistic probe dataset show that there is substantial room for improvement, especially in the cross- domain setting.

In the context of document quality assessment, previous work has mainly focused on predicting the quality of a document relative to a putative gold standard, without paying attention to the subjectivity of this task. To imitate people’s disagreement over inherently subjective tasks such as rating the quality of a Wikipedia article, a document quality assessment system should provide not only a prediction of the article quality but also the uncertainty over its predictions. This motivates us to measure the uncertainty in document quality predictions, in addition to making the label prediction. Experimental results show that both Gaussian processes (GPs) and random forests (RFs) can yield competitive results in predicting the quality of Wikipedia articles, while providing an estimate of uncertainty when there is inconsistency in the quality labels from the Wikipedia contributors. We additionally evaluate our methods in the context of a semi-automated document quality class assignment decision-making process, where there is asymmetric risk associated with overestimates and underestimates of document quality. Our experiments suggest that GPs provide more reliable estimates in this context.