The past decade has seen a substantial rise in the amount of mis- and disinformation online, from targeted disinformation campaigns to influence politics, to the unintentional spreading of misinformation about public health. This development has spurred research in the area of automatic fact checking, from approaches to detect check-worthy claims and determining the stance of tweets towards claims, to methods to determine the veracity of claims given evidence documents. These automatic methods are often content-based, using natural language processing methods, which in turn utilise deep neural networks to learn higher-order features from text in order to make predictions. As deep neural networks are black-box models, their inner workings cannot be easily explained. At the same time, it is desirable to explain how they arrive at certain decisions, especially if they are to be used for decision making. While this has been known for some time, the issues this raises have been exacerbated by models increasing in size, and by EU legislation requiring models to be used for decision making to provide explanations, and, very recently, by legislation requiring online platforms operating in the EU to provide transparent reporting on their services. Despite this, current solutions for explainability are still lacking in the area of fact checking. This thesis presents my research on automatic fact checking, including claim check-worthiness detection, stance detection and veracity prediction. Its contributions go beyond fact checking, with the thesis proposing more general machine learning solutions for natural language processing in the area of learning with limited labelled data. Finally, the thesis presents some first solutions for explainable fact checking.
Alongside huge volumes of research on deep learning models in NLP in the recent years, there has been also much work on benchmark datasets needed to track modeling progress. Question answering and reading comprehension have been particularly prolific in this regard, with over 80 new datasets appearing in the past two years. This study is the largest survey of the field to date. We provide an overview of the various formats and domains of the current resources, highlighting the current lacunae for future work. We further discuss the current classifications of ``reasoning types" in question answering and propose a new taxonomy. We also discuss the implications of over-focusing on English, and survey the current monolingual resources for other languages and multilingual resources. The study is aimed at both practitioners looking for pointers to the wealth of existing data, and at researchers working on new resources.
Sparse attention has been claimed to increase model interpretability under the assumption that it highlights influential inputs. Yet the attention distribution is typically over representations internal to the model rather than the inputs themselves, suggesting this assumption may not have merit. We build on the recent work exploring the interpretability of attention; we design a set of experiments to help us understand how sparsity affects our ability to use attention as an explainability tool. On three text classification tasks, we verify that only a weak relationship between inputs and co-indexed intermediate representations exists -- under sparse attention and otherwise. Further, we do not find any plausible mappings from sparse attention distributions to a sparse set of influential inputs through other avenues. Rather, we observe in this setting that inducing sparsity may make it less plausible that attention can be used as a tool for understanding model behavior.
Most work on scholarly document processing assumes that the information processed is trustworthy and factually correct. However, this is not always the case. There are two core challenges, which should be addressed: 1) ensuring that scientific publications are credible -- e.g. that claims are not made without supporting evidence, and that all relevant supporting evidence is provided; and 2) that scientific findings are not misrepresented, distorted or outright misreported when communicated by journalists or the general public. I will present some first steps towards addressing these problems and outline remaining challenges.
Scientific document understanding is challenging as the data is highly domain specific and diverse. However, datasets for tasks with scientific text require expensive manual annotation and tend to be small and limited to only one or a few fields. At the same time, scientific documents contain many potential training signals, such as citations, which can be used to build large labelled datasets. Given this, we present an in-depth study of cite-worthiness detection in English, where a sentence is labelled for whether or not it cites an external source. To accomplish this, we introduce CiteWorth, a large, contextualized, rigorously cleaned labelled dataset for cite-worthiness detection built from a massive corpus of extracted plain-text scientific documents. We show that CiteWorth is high-quality, challenging, and suitable for studying problems such as domain adaptation. Our best performing cite-worthiness detection model is a paragraph-level contextualized sentence labelling model based on Longformer, exhibiting a 5 F1 point improvement over SciBERT which considers only individual sentences. Finally, we demonstrate that language model fine-tuning with cite-worthiness as a secondary task leads to improved performance on downstream scientific document understanding tasks.
While the prevalence of large pre-trained language models has led to significant improvements in the performance of NLP systems, recent research has demonstrated that these models inherit societal biases extant in natural language. In this paper, we explore a simple method to probe pre-trained language models for gender bias, which we use to effect a multi-lingual study of gender bias towards politicians. We construct a dataset of 250k politicians from most countries in the world and quantify adjective and verb usage around those politicians' names as a function of their gender. We conduct our study in 7 languages across 6 different language modeling architectures. Our results demonstrate that stance towards politicians in pre-trained language models is highly dependent on the language used. Finally, contrary to previous findings, our study suggests that larger language models do not tend to be significantly more gender-biased than smaller ones.
Stance detection concerns the classification of a writer's viewpoint towards a target. There are different task variants, e.g., stance of a tweet vs. a full article, or stance with respect to a claim vs. an (implicit) topic. Moreover, task definitions vary, which includes the label inventory, the data collection, and the annotation protocol. All these aspects hinder cross-domain studies, as they require changes to standard domain adaptation approaches. In this paper, we perform an in-depth analysis of 16 stance detection datasets, and we explore the possibility for cross-domain learning from them. Moreover, we propose an end-to-end unsupervised framework for out-of-domain prediction of unseen, user-defined labels. In particular, we combine domain adaptation techniques such as mixture of experts and domain-adversarial training with label embeddings, and we demonstrate sizable performance gains over strong baselines -- both (i) in-domain, i.e., for seen targets, and (ii) out-of-domain, i.e., for unseen targets. Finally, we perform an exhaustive analysis of the cross-domain results, and we highlight the important factors influencing the model performance.
We propose a novel interpretable framework for cross-lingual content flagging, which significantly outperforms prior work both in terms of predictive performance and average inference time. The framework is based on a nearest-neighbour architecture and is interpretable by design. Moreover, it can easily adapt to new instances without the need to retrain it from scratch. Unlike prior work, (i) we encode not only the texts, but also the labels in the neighbourhood space (which yields better accuracy), and (ii) we use a bi-encoder instead of a cross-encoder (which saves computation time). Our evaluation results on ten different datasets for abusive language detection in eight languages shows sizable improvements over the state of the art, as well as a speed-up at inference time.
In this paper, we describe our participation in the TREC Health Misinformation Track 2020. We submitted $11$ runs to the Total Recall Task and 13 runs to the Ad Hoc task. Our approach consists of 3 steps: (1) we create an initial run with BM25 and RM3; (2) we estimate credibility and misinformation scores for the documents in the initial run; (3) we merge the relevance, credibility and misinformation scores to re-rank documents in the initial run. To estimate credibility scores, we implement a classifier which exploits features based on the content and the popularity of a document. To compute the misinformation score, we apply a stance detection approach with a pretrained Transformer language model. Finally, we use different approaches to merge scores: weighted average, the distance among score vectors and rank fusion.