We present the results of the second Fact Extraction and VERification (FEVER2.0) Shared Task. The task challenged participants to both build systems to verify factoid claims using evidence retrieved from Wikipedia and to generate adversarial attacks against other participant’s systems. The shared task had three phases: building, breaking and fixing. There were 8 systems in the builder’s round, three of which were new qualifying submissions for this shared task, and 5 adversaries generated instances designed to induce classification errors and one builder submitted a fixed system which had higher FEVER score and resilience than their first submission. All but one newly submitted systems attained FEVER scores higher than the best performing system from the first shared task and under adversarial evaluation, all systems exhibited losses in FEVER score. There was a great variety in adversarial attack types as well as the techniques used to generate the attacks, In this paper, we present the results of the shared task and a summary of the systems, highlighting commonalities and innovations among participating systems.
The goal of our paper is to compare psycholinguistic text features with fact checking approaches to distinguish lies from true statements. We examine both methods using data from a large ongoing study on deception and deception detection covering a mixture of factual and opinionated topics that polarize public opinion. We conclude that fact checking approaches based on Wikipedia are too limited for this task, as only a few percent of sentences from our study has enough evidence to become supported or refuted. Psycholinguistic features turn out to outperform both fact checking and human baselines, but the accuracy is not high. Overall, it appears that deception detection applicable to less-than-obvious topics is a difficult task and a problem to be solved.
We present a multi-task learning model that leverages large amount of textual information from existing datasets to improve stance prediction. In particular, we utilize multiple NLP tasks under both unsupervised and supervised settings for the target stance prediction task. Our model obtains state-of-the-art performance on a public benchmark dataset, Fake News Challenge, outperforming current approaches by a wide margin.
We present our Generative Enhanced Model (GEM) that we used to create samples awarded the first prize on the FEVER 2.0 Breakers Task. GEM is the extended language model developed upon GPT-2 architecture. The addition of novel target vocabulary input to the already existing context input enabled controlled text generation. The training procedure resulted in creating a model that inherited the knowledge of pretrained GPT-2, and therefore was ready to generate natural-like English sentences in the task domain with some additional control. As a result, GEM generated malicious claims that mixed facts from various articles, so it became difficult to classify their truthfulness.
In this paper, we propose a new approach to learn multimodal multilingual embeddings for matching images and their relevant captions in two languages. We combine two existing objective functions to make images and captions close in a joint embedding space while adapting the alignment of word embeddings between existing languages in our model. We show that our approach enables better generalization, achieving state-of-the-art performance in text-to-image and image-to-text retrieval task, and caption-caption similarity task. Two multimodal multilingual datasets are used for evaluation: Multi30k with German and English captions and Microsoft-COCO with English and Japanese captions.
The recent demonstration of the power of huge language models such as GPT-2 to memorise the answers to factoid questions raises questions about the extent to which knowledge is being embedded directly within these large models. This short paper describes an architecture through which much smaller models can also answer such questions - by making use of ‘raw’ external knowledge. The contribution of this work is that the methods presented here rely on unsupervised learning techniques, complementing the unsupervised training of the Language Model. The goal of this line of research is to be able to add knowledge explicitly, without extensive training.
We present a scalable, open-source platform that “distills” a potentially large text collection into a knowledge graph. Our platform takes documents stored in Apache Solr and scales out the Stanford CoreNLP toolkit via Apache Spark integration to extract mentions and relations that are then ingested into the Neo4j graph database. The raw knowledge graph is then enriched with facts extracted from an external knowledge graph. The complete product can be manipulated by various applications using Neo4j’s native Cypher query language: We present a subgraph-matching approach to align extracted relations with external facts and show that fact verification, locating textual support for asserted facts, detecting inconsistent and missing facts, and extracting distantly-supervised training data can all be performed within the same framework.
Many previous studies on relation extrac-tion have been focused on finding only one relation between two entities in a single sentence. However, we can easily find the fact that multiple entities exist in a single sentence and the entities form multiple relations. To resolve this prob-lem, we propose a relation extraction model based on a dual pointer network with a multi-head attention mechanism. The proposed model finds n-to-1 subject-object relations by using a forward de-coder called an object decoder. Then, it finds 1-to-n subject-object relations by using a backward decoder called a sub-ject decoder. In the experiments with the ACE-05 dataset and the NYT dataset, the proposed model achieved the state-of-the-art performances (F1-score of 80.5% in the ACE-05 dataset, F1-score of 78.3% in the NYT dataset)
Recent Deep Learning (DL) models have succeeded in achieving human-level accuracy on various natural language tasks such as question-answering, natural language inference (NLI), and textual entailment. These tasks not only require the contextual knowledge but also the reasoning abilities to be solved efficiently. In this paper, we propose an unsupervised question-answering based approach for a similar task, fact-checking. We transform the FEVER dataset into a Cloze-task by masking named entities provided in the claims. To predict the answer token, we utilize pre-trained Bidirectional Encoder Representations from Transformers (BERT). The classifier computes label based on the correctly answered questions and a threshold. Currently, the classifier is able to classify the claims as “SUPPORTS” and “MANUAL_REVIEW”. This approach achieves a label accuracy of 80.2% on the development set and 80.25% on the test set of the transformed dataset.
Recognizing the implicit link between a claim and a piece of evidence (i.e. warrant) is the key to improving the performance of evidence detection. In this work, we explore the effectiveness of automatically extracted warrants for evidence detection. Given a claim and candidate evidence, our proposed method extracts multiple warrants via similarity search from an existing, structured corpus of arguments. We then attentively aggregate the extracted warrants, considering the consistency between the given argument and the acquired warrants. Although a qualitative analysis on the warrants shows that the extraction method needs to be improved, our results indicate that our method can still improve the performance of evidence detection.
One of the important tasks in opinion mining is to extract aspects of the opinion target. Aspects are features or characteristics of the opinion target that are being reviewed, which can be categorised into explicit and implicit aspects. Extracting aspects from opinions is essential in order to ensure accurate information about certain attributes of an opinion target is retrieved. For instance, a professional camera receives a positive feedback in terms of its functionalities in a review, but its overly high price receives negative feedback. Most of the existing solutions focus on explicit aspects. However, sentences in reviews normally do not state the aspects explicitly. In this research, two hybrid models are proposed to identify and extract both explicit and implicit aspects, namely TDM-DC and TDM-TED. The proposed models combine topic modelling and dictionary-based approach. The models are unsupervised as they do not require any labelled dataset. The experimental results show that TDM-DC achieves F1-measure of 58.70%, where it outperforms both the baseline topic model and dictionary-based approach. In comparison to other existing unsupervised techniques, the proposed models are able to achieve higher F1-measure by approximately 3%. Although the supervised techniques perform slightly better, the proposed models are domain-independent, and hence more versatile.
Triggered by Internet development, a large amount of information is published in online sources. However, it is a well-known fact that publications are inundated with inaccurate data. That is why fact-checking has become a significant topic in the last 5 years. It is widely accepted that factual data verification is a challenge even for the experts. This paper presents a domain-independent fact checking system. It can solve the fact verification problem entirely or at the individual stages. The proposed model combines various advanced methods of text data analysis, such as BERT and Infersent. The theoretical and empirical study of the system features is carried out. Based on FEVER and Fact Checking Challenge test-collections, experimental results demonstrate that our model can achieve the score on a par with state-of-the-art models designed by the specificity of particular datasets.
Finding evidence is of vital importance in research as well as fact checking and an evidence detection method would be useful in speeding up this process. However, when addressing a new topic there is no training data and there are two approaches to get started. One could use large amounts of out-of-domain data to train a state-of-the-art method, or to use the small data that a person creates while working on the topic. In this paper, we address this problem in two steps. First, by simulating users who read source documents and label sentences they can use as evidence, thereby creating small amounts of training data for an interactively trained evidence detection model; and second, by comparing such an interactively trained model against a pre-trained model that has been trained on large out-of-domain data. We found that an interactively trained model not only often out-performs a state-of-the-art model but also requires significantly lower amounts of computational resources. Therefore, especially when computational resources are scarce, e.g. no GPU available, training a smaller model on the fly is preferable to training a well generalising but resource hungry out-of-domain model.
Defined as the intentional or unintentionalspread of false information (K et al., 2019)through context and/or content manipulation,fake news has become one of the most seriousproblems associated with online information(Waldrop, 2017). Consequently, it comes asno surprise that Fake News Detection hasbecome one of the major foci of variousfields of machine learning and while machinelearning models have allowed individualsand companies to automate decision-basedprocesses that were once thought to be onlydoable by humans, it is no secret that thereal-life applications of such models are notviable without the existence of an adequatetraining dataset. In this paper we describethe Veritas Annotator, a web application formanually identifying the origin of a rumour.These rumours, often referred as claims,were previously checked for validity byFact-Checking Agencies.
We describe our submission for the Breaker phase of the second Fact Extraction and VERification (FEVER) Shared Task. Our adversarial data can be explained by two perspectives. First, we aimed at testing model’s ability to retrieve evidence, when appropriate query terms could not be easily generated from the claim. Second, we test model’s ability to precisely understand the implications of the texts, which we expect to be rare in FEVER 1.0 dataset. Overall, we suggested six types of adversarial attacks. The evaluation on the submitted systems showed that the systems were only able get both the evidence and label correct in 20% of the data. We also demonstrate our adversarial run analysis in the data development process.
This paper contains our system description for the second Fact Extraction and VERification (FEVER) challenge. We propose a two-staged sentence selection strategy to account for examples in the dataset where evidence is not only conditioned on the claim, but also on previously retrieved evidence. We use a publicly available document retrieval module and have fine-tuned BERT checkpoints for sentence se- lection and as the entailment classifier. We report a FEVER score of 68.46% on the blind testset.
Fever Shared 2.0 Task is a challenge meant for developing automated fact checking systems. Our approach for the Fever 2.0 is based on a previous proposal developed by Team Athene UKP TU Darmstadt. Our proposal modifies the sentence retrieval phase, using statement extraction and representation in the form of triplets (subject, object, action). Triplets are extracted from the claim and compare to triplets extracted from Wikipedia articles using semantic similarity. Our results are satisfactory but there is room for improvement.