We present polyNarrative, a new multilingual dataset of news articles, annotated for narratives. Narratives are overt or implicit claims, recurring across articles and languages, promoting a specific interpretation or viewpoint on an ongoing topic, often propagating mis/disinformation. We developed two-level taxonomies with coarse- and fine-grained narrative labels for two domains: (i) climate change and (ii) the military conflict between Ukraine and Russia. We collected news articles in four languages (Bulgarian, English, Portuguese, and Russian) related to the two domains and manually annotated them at the paragraph level. We make the dataset publicly available, along with experimental results of several strong baselines that assign narrative labels to news articles at the paragraph or the document level. We believe that this dataset will foster research in narrative detection and enable new research directions towards more multi-domain and highly granular narrative related tasks.

We introduce a novel multilingual and hierarchical corpus annotated for entity framing and role portrayal in news articles. The dataset uses a unique taxonomy inspired by storytelling elements, comprising 22 fine-grained roles, or archetypes, nested within three main categories: protagonist, antagonist, and innocent. Each archetype is carefully defined, capturing nuanced portrayals of entities such as guardian, martyr, and underdog for protagonists; tyrant, deceiver, and bigot for antagonists; and victim, scapegoat, and exploited for innocents. The dataset includes 1,378 recent news articles in five languages (Bulgarian, English, Hindi, European Portuguese, and Russian) focusing on two critical domains of global significance: the Ukraine-Russia War and Climate Change. Over 5,800 entity mentions have been annotated with role labels. This dataset serves as a valuable resource for research into role portrayal and has broader implications for news analysis. We describe the characteristics of the dataset and the annotation process, and we report evaluation results on fine-tuned state-of-the-art multilingual transformers and hierarchical zero-shot learning using LLMs at the level of a document, a paragraph, and a sentence.

We present NarratEX, a dataset designed for the task of explaining the choice of the Dominant Narrative in a news article, and intended to support the research community in addressing challenges such as discourse polarization and propaganda detection. Our dataset comprises 1,056 news articles in four languages, Bulgarian, English, Portuguese, and Russian, covering two globally significant topics: the Ukraine-Russia War (URW) and Climate Change (CC). Each article is manually annotated with a dominant narrative and sub-narrative labels, and an explanation justifying the chosen labels. We describe the dataset, the process of its creation, and its characteristics. We present experiments with two new proposed tasks: Explaining Dominant Narrative based on Text, which involves writing a concise paragraph to justify the choice of the dominant narrative and sub-narrative of a given text, and Inferring Dominant Narrative from Explanation, which involves predicting the appropriate dominant narrative category based on an explanatory text. The proposed dataset is a valuable resource for advancing research on detecting and mitigating manipulative content, while promoting a deeper understanding of how narratives influence public discourse.

We introduce SemEval-2025 Task 10 on Multilingual Characterization and Extraction of Narratives from Online News, which focuses on the identification and analysis of narratives in online news media. The task is structured into three subtasks: (1) Entity Framing, to identify the roles that relevant entities play within narratives, (2) Narrative Classification, to assign documents fine-grained narratives according to a given, topic-specific taxonomy of narrative labels, and (3) Narrative Extraction, to provide a justification for the dominant narrative of the document. To this end, we analyze news articles across two critical domains, Ukraine-Russia War and Climate Change, in five languages: Bulgarian, English, Hindi, Portuguese, and Russian. This task introduces a novel multilingual and multifaceted framework for studying how online news media construct and disseminate manipulative narratives. By addressing these challenges, our work contributes to the broader effort of detecting, understanding, and mitigating the spread of propaganda and disinformation. The task attracted a lot of interest: 310 teams registered, with 66 submitting official results on the test set.

The abundance of news sources and the urgent demand for reliable information have led to serious concerns about the threat of misleading information. In this paper, we present FRAPPE, a FRAming, Persuasion, and Propaganda Explorer system. FRAPPE goes beyond conventional news analysis of articles and unveils the intricate linguistic techniques used to shape readers’ opinions and emotions. Our system allows users not only to analyze individual articles for their genre, framings, and use of persuasion techniques, but also to draw comparisons between the strategies of persuasion and framing adopted by a diverse pool of news outlets and countries across multiple languages for different topics, thus providing a comprehensive understanding of how information is presented and manipulated. FRAPPE is publicly accessible at https://frappe.streamlit.app/ and a video explaining our system is available at https://www.youtube.com/watch?v=3RlTfSVnZmk

We systematically explore the predictive power of features derived from Persuasion Techniques detected in texts, for solving different tasks of interest for media analysis; notably: detecting mis/disinformation, fake news, propaganda, partisan news and conspiracy theories. Firstly, we propose a set of meaningful features, aiming to capture the persuasiveness of a text. Secondly, we assess the discriminatory power of these features in different text classification tasks on 8 selected datasets from the literature using two metrics. We also evaluate the per-task discriminatory power of each Persuasion Technique and report on different insights. We find out that most of these features have a noticeable potential to distinguish conspiracy theories, hyperpartisan news and propaganda, while we observed mixed results in the context of fake news detection.

We present a new multilingual multifacet dataset of news articles, each annotated for genre (objective news reporting vs. opinion vs. satire), framing (what key aspects are highlighted), and persuasion techniques (logical fallacies, emotional appeals, ad hominem attacks, etc.). The persuasion techniques are annotated at the span level, using a taxonomy of 23 fine-grained techniques grouped into 6 coarse categories. The dataset contains 1,612 news articles covering recent news on current topics of public interest in six European languages (English, French, German, Italian, Polish, and Russian), with more than 37k annotated spans of persuasion techniques. We describe the dataset and the annotation process, and we report the evaluation results of multilabel classification experiments using state-of-the-art multilingual transformers at different levels of granularity: token-level, sentence-level, paragraph-level, and document-level.

This paper reports on the results of preliminary experiments on the detection of persuasion techniques in online news in Polish and Russian, using a taxonomy of 23 persuasion techniques. The evaluation addresses different aspects, namely, the granularity of the persuasion technique category, i.e., coarse- (6 labels) versus fine-grained (23 labels), and the focus of the classification, i.e., at which level the labels are detected (subword, sentence, or paragraph). We compare the performance of mono- verus multi-lingual-trained state-of-the-art transformed-based models in this context.

The purpose of the shared task 2 at the Challenges and Applications of Automated Extraction of Socio-political Events from Text (CASE) 2023 workshop was to test the abilities of the participating models and systems to detect and geocode armed conflicts events in social media messages from Telegram channels reporting on the Russo Ukrainian war. The evaluation followed an approach which was introduced in CASE 2021 (Giorgi et al., 2021): For each system we consider the correlation of the spatio-temporal distribution of its detected events and the events identified for the same period in the ACLED (Armed Conflict Location and Event Data Project) database (Raleigh et al., 2010). We use ACLED for the ground truth, since it is a well established standard in the field of event extraction and political trend analysis, which relies on human annotators for the encoding of security events using a fine grained taxonomy. Two systems participated in this shared task, we report in this paper on both the shared task and the participating systems.

In this paper we report on the complexity of persuasion technique annotation in the context of a large multilingual annotation campaign involving 6 languages and approximately 40 annotators. We highlight the techniques that appear to be difficult for humans to annotate and elaborate on our findings on the causes of this phenomenon. We introduce Holistic IAA, a new word embedding-based annotator agreement metric and we report on various experiments using this metric and its correlation with the traditional Inter Annotator Agreement (IAA) metrics. However, given somewhat limited and loose interaction between annotators, i.e., only a few annotators annotate the same document subsets, we try to devise a way to assess the coherence of the entire dataset and strive to find a good proxy for IAA between annotators tasked to annotate different documents and in different languages, for which classical IAA metrics can not be applied.

This paper describes the system we used to participate in the shared task, as well as additional experiments beyond the scope of the shared task, but using its data. Our primary goal is to compare the effectiveness of transformers model compared to low-resource dictionaries. Secondly, we compare the difference in performance of a learned dictionary and of a dictionary designed by experts in the field of values. Our findings surprisingly show that transformers perform on par with a dictionary containing less than 1k words, when evaluated with 19 fine-grained categories, and only outperform a dictionary-based approach in a coarse setting with 10 categories. Interestingly, the expert dictionary has a precision on par with the learned one, while its recall is clearly lower, potentially an indication of overfitting of topics to values in the shared task’s dataset. Our findings should be of interest to both the NLP and Value scientific communities on the use of automated approaches for value classification

We describe SemEval-2023 task 3 on Detecting the Category, the Framing, and the Persuasion Techniques in Online News in a Multilingual Setup: the dataset, the task organization process, the evaluation setup, the results, and the participating systems. The task focused on news articles in nine languages (six known to the participants upfront: English, French, German, Italian, Polish, and Russian), and three additional ones revealed to the participants at the testing phase: Spanish, Greek, and Georgian). The task featured three subtasks: (1) determining the genre of the article (opinion, reporting, or satire), (2) identifying one or more frames used in an article from a pool of 14 generic frames, and (3) identify the persuasion techniques used in each paragraph of the article, using a taxonomy of 23 persuasion techniques. This was a very popular task: a total of 181 teams registered to participate, and 41 eventually made an official submission on the test set.

In this paper we present an approach to efficiently recover texts from corrupted documents of endangered languages. Textual resources for such languages are scarce, and sometimes the few available resources are corrupted PDF documents. Endangered languages are not supported by standard tools and present even the additional difficulties of not possessing any corpus over which to train language models to assist with the recovery. The approach presented is able to fully recover born digital PDF documents with minimal effort, thereby helping the preservation effort of endangered languages, by extending the range of documents usable for corpus building.

This paper presents, to the best of our knowledge, the first ever publicly available annotated dataset for sentiment classification and semantic polarity dictionary for Georgian. The characteristics of these resources and the process of their creation are described in detail. The results of various experiments on the performance of both lexicon- and machine learning-based models for Georgian sentiment classification are also reported. Both 3-label (positive, neutral, negative) and 4-label settings (same labels + mixed) are considered. The machine learning models explored include, i.a., logistic regression, SVMs, and transformed-based models. We also explore transfer learning- and translation-based (to a well-supported language) approaches. The obtained results for Georgian are on par with the state-of-the-art results in sentiment classification for well studied languages when using training data of comparable size.

We present our contribution to the SemEval 22 Share Task 8: Multilingual news article similarity. The approach is lightweight and language-agnostic, it is based on the computation of several lexicographic and embedding-based features, and the use of a simple ML approach: random forests. In a notable departure from the task formulation, which is a ranking task, we tackled this task as a classification one. We present a detailed analysis of the behaviour of our system under different settings.

This paper describes the Shared Task on Fine-grained Event Classification in News-like Text Snippets. The Shared Task is divided into three sub-tasks: (a) classification of text snippets reporting socio-political events (25 classes) for which vast amount of training data exists, although exhibiting different structure and style vis-a-vis test data, (b) enhancement to a generalized zero-shot learning problem, where 3 additional event types were introduced in advance, but without any training data (‘unseen’ classes), and (c) further extension, which introduced 2 additional event types, announced shortly prior to the evaluation phase. The reported Shared Task focuses on classification of events in English texts and is organized as part of the Workshop on Challenges and Applications of Automated Extraction of Socio-political Events from Text (CASE 2021), co-located with the ACL-IJCNLP 2021 Conference. Four teams participated in the task. Best performing systems for the three aforementioned sub-tasks achieved 83.9%, 79.7% and 77.1% weighted F1 scores respectively.

Evaluating the state-of-the-art event detection systems on determining spatio-temporal distribution of the events on the ground is performed unfrequently. But, the ability to both (1) extract events “in the wild” from text and (2) properly evaluate event detection systems has potential to support a wide variety of tasks such as monitoring the activity of socio-political movements, examining media coverage and public support of these movements, and informing policy decisions. Therefore, we study performance of the best event detection systems on detecting Black Lives Matter (BLM) events from tweets and news articles. The murder of George Floyd, an unarmed Black man, at the hands of police officers received global attention throughout the second half of 2020. Protests against police violence emerged worldwide and the BLM movement, which was once mostly regulated to the United States, was now seeing activity globally. This shared task asks participants to identify BLM related events from large unstructured data sources, using systems pretrained to extract socio-political events from text. We evaluate several metrics, accessing each system’s ability to identify protest events both temporally and spatially. Results show that identifying daily protest counts is an easier task than classifying spatial and temporal protest trends simultaneously, with maximum performance of 0.745 and 0.210 (Pearson r), respectively. Additionally, all baselines and participant systems suffered from low recall, with a maximum recall of 5.08.

This paper presents a study of state-of-the-art unsupervised and linguistically unsophisticated keyword extraction algorithms, based on statistic-, graph-, and embedding-based approaches, including, i.a., Total Keyword Frequency, TF-IDF, RAKE, KPMiner, YAKE, KeyBERT, and variants of TextRank-based keyword extraction algorithms. The study was motivated by the need to select the most appropriate technique to extract keywords for indexing news articles in a real-world large-scale news analysis engine. The algorithms were evaluated on a corpus of circa 330 news articles in 7 languages. The overall best F1 scores for all languages on average were obtained using a combination of the recently introduced YAKE algorithm and KPMiner (20.1%, 46.6% and 47.2% for exact, partial and fuzzy matching resp.).