We present bgGLUE (Bulgarian General Language Understanding Evaluation), a benchmark for evaluating language models on Natural Language Understanding (NLU) tasks in Bulgarian. Our benchmark includes NLU tasks targeting a variety of NLP problems (e.g., natural language inference, fact-checking, named entity recognition, sentiment analysis, question answering, etc.) and machine learning tasks (sequence labeling, document-level classification, and regression). We run the first systematic evaluation of pre-trained language models for Bulgarian, comparing and contrasting results across the nine tasks in the benchmark. The evaluation results show strong performance on sequence labeling tasks, but there is a lot of room for improvement for tasks that require more complex reasoning. We make bgGLUE publicly available together with the fine-tuning and the evaluation code, as well as a public leaderboard at https://bgglue.github.io, and we hope that it will enable further advancements in developing NLU models for Bulgarian.

The paper presents an approach for solving SemEval 2023 Task 7 - identifying the inference relation in a clinical trials dataset. The system has two levels for retrieving relevant clinical trial evidence for a statement and then classifying the inference relation based on the relevant sentences. In the first level, the system classifies the evidence-statement pairs as relevant or not using a BERT-based classifier and contextual data augmentation (subtask 2). Using the relevant parts of the clinical trial from the first level, the system uses an additional BERT-based classifier to determine whether the relation is entailment or contradiction (subtask 1). In both levels, the contextual data augmentation is showing a significant improvement in the F1 score on the test set of 3.7% for subtask 2 and 7.6% for subtask 1, achieving final F1 scores of 82.7% for subtask 2 and 64.4% for subtask 1.

While there has been substantial progress in developing systems to automate fact-checking, they still lack credibility in the eyes of the users. Thus, an interesting approach has emerged: to perform automatic fact-checking by verifying whether an input claim has been previously fact-checked by professional fact-checkers and to return back an article that explains their decision. This is a sensible approach as people trust manual fact-checking, and as many claims are repeated multiple times. Yet, a major issue when building such systems is the small number of known tweet–verifying article pairs available for training. Here, we aim to bridge this gap by making use of crowd fact-checking, i.e., mining claims in social media for which users have responded with a link to a fact-checking article. In particular, we mine a large-scale collection of 330,000 tweets paired with a corresponding fact-checking article. We further propose an end-to-end framework to learn from this noisy data based on modified self-adaptive training, in a distant supervision scenario. Our experiments on the CLEF’21 CheckThat! test set show improvements over the state of the art by two points absolute. Our code and datasets are available at https://github.com/mhardalov/crowdchecked-claims

We propose a novel framework for predicting the factuality of reporting of news media outlets by studying the user attention cycles in their YouTube channels. In particular, we design a rich set of features derived from the temporal evolution of the number of views, likes, dislikes, and comments for a video, which we then aggregate to the channel level. We develop and release a dataset for the task, containing observations of user attention on YouTube channels for 489 news media. Our experiments demonstrate both complementarity and sizable improvements over state-of-the-art textual representations.

The task of automatic diagnosis encoding into standard medical classifications and ontologies, is of great importance in medicine - both to support the daily tasks of physicians in the preparation and reporting of clinical documentation, and for automatic processing of clinical reports. In this paper we investigate the application and performance of different deep learning transformers for automatic encoding in ICD-10 of clinical texts in Bulgarian. The comparative analysis attempts to find which approach is more efficient to be used for fine-tuning of pretrained BERT family transformer to deal with a specific domain terminology on a rare language as Bulgarian. On the one side are used SlavicBERT and MultiligualBERT, that are pretrained for common vocabulary in Bulgarian, but lack medical terminology. On the other hand in the analysis are used BioBERT, ClinicalBERT, SapBERT, BlueBERT, that are pretrained for medical terminology in English, but lack training for language models in Bulgarian, and more over for vocabulary in Cyrillic. In our research study all BERT models are fine-tuned with additional medical texts in Bulgarian and then applied to the classification task for encoding medical diagnoses in Bulgarian into ICD-10 codes. Big corpora of diagnosis in Bulgarian annotated with ICD-10 codes is used for the classification task. Such an analysis gives a good idea of which of the models would be suitable for tasks of a similar type and domain. The experiments and evaluation results show that both approaches have comparable accuracy.

Extracting the most important part of legislation documents has great business value because the texts are usually very long and hard to understand. The aim of this article is to evaluate different algorithms for text summarization on EU legislation documents. The content contains domain-specific words. We collected a text summarization dataset of EU legal documents consisting of 1563 documents, in which the mean length of summaries is 424 words. Experiments were conducted with different algorithms using the new dataset. A simple extractive algorithm was selected as a baseline. Advanced extractive algorithms, which use encoders show better results than baseline. The best result measured by ROUGE scores was achieved by a fine-tuned abstractive T5 model, which was adapted to work with long texts.

In education, quiz questions have become an important tool for assessing the knowledge of students. Yet, manually preparing such questions is a tedious task, and thus automatic question generation has been proposed as a possible alternative. So far, the vast majority of research has focused on generating the question text, relying on question answering datasets with readily picked answers, and the problem of how to come up with answer candidates in the first place has been largely ignored. Here, we aim to bridge this gap. In particular, we propose a model that can generate a specified number of answer candidates for a given passage of text, which can then be used by instructors to write questions manually or can be passed as an input to automatic answer-aware question generators. Our experiments show that our proposed answer candidate generation model outperforms several baselines.

Vast amounts of data in healthcare are available in unstructured text format, usually in the local language of the countries. These documents contain valuable information. Secondary use of clinical narratives and information extraction of key facts and relations from them about the patient disease history can foster preventive medicine and improve healthcare. In this paper, we propose a hybrid method for the automatic transformation of clinical text into a structured format. The documents are automatically sectioned into the following parts: diagnosis, patient history, patient status, lab results. For the “Diagnosis” section a deep learning text-based encoding into ICD-10 codes is applied using MBG-ClinicalBERT - a fine-tuned ClinicalBERT model for Bulgarian medical text. From the “Patient History” section, we identify patient symptoms using a rule-based approach enhanced with similarity search based on MBG-ClinicalBERT word embeddings. We also identify symptom relations like negation. For the “Patient Status” description, binary classification is used to determine the status of each anatomic organ. In this paper, we demonstrate different methods for adapting NLP tools for English and other languages to a low resource language like Bulgarian.

We propose EXAMS – a new benchmark dataset for cross-lingual and multilingual question answering for high school examinations. We collected more than 24,000 high-quality high school exam questions in 16 languages, covering 8 language families and 24 school subjects from Natural Sciences and Social Sciences, among others.EXAMS offers unique fine-grained evaluation framework across multiple languages and subjects, which allows precise analysis and comparison of the proposed models. We perform various experiments with existing top-performing multilingual pre-trained models and show that EXAMS offers multiple challenges that require multilingual knowledge and reasoning in multiple domains. We hope that EXAMS will enable researchers to explore challenging reasoning and knowledge transfer methods and pre-trained models for school question answering in various languages which was not possible by now. The data, code, pre-trained models, and evaluation are available at http://github.com/mhardalov/exams-qa.

The recent explosion of false claims in social media and on the Web in general has given rise to a lot of manual fact-checking initiatives. Unfortunately, the number of claims that need to be fact-checked is several orders of magnitude larger than what humans can handle manually. Thus, there has been a lot of research aiming at automating the process. Interestingly, previous work has largely ignored the growing number of claims about images. This is despite the fact that visual imagery is more influential than text and naturally appears alongside fake news. Here we aim at bridging this gap. In particular, we create a new dataset for this problem, and we explore a variety of features modeling the claim, the image, and the relationship between the claim and the image. The evaluation results show sizable improvements over the baseline. We release our dataset, hoping to enable further research on fact-checking claims about images.

Online media aim for reaching ever bigger audience and for attracting ever longer attention span. This competition creates an environment that rewards sensational, fake, and toxic news. To help limit their spread and impact, we propose and develop a news toxicity detector that can recognize various types of toxic content. While previous research primarily focused on English, here we target Bulgarian. We created a new dataset by crawling a website that for five years has been collecting Bulgarian news articles that were manually categorized into eight toxicity groups. Then we trained a multi-class classifier with nine categories: eight toxic and one non-toxic. We experimented with different representations based on ElMo, BERT, and XLM, as well as with a variety of domain-specific features. Due to the small size of our dataset, we created a separate model for each feature type, and we ultimately combined these models into a meta-classifier. The evaluation results show an accuracy of 59.0% and a macro-F1 score of 39.7%, which represent sizable improvements over the majority-class baseline (Acc=30.3%, macro-F1=5.2%).

Recently, reading comprehension models achieved near-human performance on large-scale datasets such as SQuAD, CoQA, MS Macro, RACE, etc. This is largely due to the release of pre-trained contextualized representations such as BERT and ELMo, which can be fine-tuned for the target task. Despite those advances and the creation of more challenging datasets, most of the work is still done for English. Here, we study the effectiveness of multilingual BERT fine-tuned on large-scale English datasets for reading comprehension (e.g., for RACE), and we apply it to Bulgarian multiple-choice reading comprehension. We propose a new dataset containing 2,221 questions from matriculation exams for twelfth grade in various subjects —history, biology, geography and philosophy—, and 412 additional questions from online quizzes in history. While the quiz authors gave no relevant context, we incorporate knowledge from Wikipedia, retrieving documents matching the combination of question + each answer option. Moreover, we experiment with different indexing and pre-training strategies. The evaluation results show accuracy of 42.23%, which is well above the baseline of 24.89%.

This paper presents an approach for prediction of results for sport events. Usually the sport forecasting approaches are based on structured data. We test the hypothesis that the sports results can be predicted by using natural language processing and machine learning techniques applied over interviews with the players shortly before the sport events. The proposed method uses deep learning contextual models, applied over unstructured textual documents. Several experiments were performed for interviews with players in individual sports like boxing, martial arts, and tennis. The results from the conducted experiment confirmed our initial assumption that an interview from a sportsman before a match contains information that can be used for prediction the outcome from it. Furthermore, the results provide strong evidence in support of our research hypothesis, that is, we can predict the outcome from a sport match analyzing an interview, given before it.

We present the system built for SemEval-2018 Task 2 on Emoji Prediction. Although Twitter messages are very short we managed to design a wide variety of features: textual, semantic, sentiment, emotion-, and color-related ones. We investigated different methods of text preprocessing including replacing text emojis with respective tokens and splitting hashtags to capture more meaning. To represent text we used word n-grams and word embeddings. We experimented with a wide range of classifiers and our best results were achieved using a SVM-based classifier and a Hierarchical Attention Neural Network.

We propose to use question answering (QA) data from Web forums to train chat-bots from scratch, i.e., without dialog data. First, we extract pairs of question and answer sentences from the typically much longer texts of questions and answers in a forum. We then use these shorter texts to train seq2seq models in a more efficient way. We further improve the parameter optimization using a new model selection strategy based on QA measures. Finally, we propose to use extrinsic evaluation with respect to a QA task as an automatic evaluation method for chatbot systems. The evaluation shows that the model achieves a MAP of 63.5% on the extrinsic task. Moreover, our manual evaluation demonstrates that the model can answer correctly 49.5% of the questions when they are similar in style to how questions are asked in the forum, and 47.3% of the questions, when they are more conversational in style.

In the context of investigative journalism, we address the problem of automatically identifying which claims in a given document are most worthy and should be prioritized for fact-checking. Despite its importance, this is a relatively understudied problem. Thus, we create a new corpus of political debates, containing statements that have been fact-checked by nine reputable sources, and we train machine learning models to predict which claims should be prioritized for fact-checking, i.e., we model the problem as a ranking task. Unlike previous work, which has looked primarily at sentences in isolation, in this paper we focus on a rich input representation modeling the context: relationship between the target statement and the larger context of the debate, interaction between the opponents, and reaction by the moderator and by the public. Our experiments show state-of-the-art results, outperforming a strong rivaling system by a margin, while also confirming the importance of the contextual information.

It is completely amazing! Fake news and “click baits” have totally invaded the cyberspace. Let us face it: everybody hates them for three simple reasons. Reason #2 will absolutely amaze you. What these can achieve at the time of election will completely blow your mind! Now, we all agree, this cannot go on, you know, somebody has to stop it. So, we did this research, and trust us, it is totally great research, it really is! Make no mistake. This is the best research ever! Seriously, come have a look, we have it all: neural networks, attention mechanism, sentiment lexicons, author profiling, you name it. Lexical features, semantic features, we absolutely have it all. And we have totally tested it, trust us! We have results, and numbers, really big numbers. The best numbers ever! Oh, and analysis, absolutely top notch analysis. Interested? Come read the shocking truth about fake news and clickbait in the Bulgarian cyberspace. You won’t believe what we have found!

Given the constantly growing proliferation of false claims online in recent years, there has been also a growing research interest in automatically distinguishing false rumors from factually true claims. Here, we propose a general-purpose framework for fully-automatic fact checking using external sources, tapping the potential of the entire Web as a knowledge source to confirm or reject a claim. Our framework uses a deep neural network with LSTM text encoding to combine semantic kernels with task-specific embeddings that encode a claim together with pieces of potentially relevant text fragments from the Web, taking the source reliability into account. The evaluation results show good performance on two different tasks and datasets: (i) rumor detection and (ii) fact checking of the answers to a question in community question answering forums.

We address information credibility in community forums, in a setting in which the credibility of an answer posted in a question thread by a particular user has to be predicted. First, we motivate the problem and we create a publicly available annotated English corpus by crowdsourcing. Second, we propose a large set of features to predict the credibility of the answers. The features model the user, the answer, the question, the thread as a whole, and the interaction between them. Our experiments with ranking SVMs show that the credibility labels can be predicted with high performance according to several standard IR ranking metrics, thus supporting the potential usage of this layer of credibility information in practical applications. The features modeling the profile of the user (in particular trollness) turn out to be most important, but embedding features modeling the answer and the similarity between the question and the answer are also very relevant. Overall, half of the gap between the baseline performance and the perfect classifier can be covered using the proposed features.