ExU: AI Models for Examining Multilingual Disinformation Narratives and Understanding their Spread

Rumor stance classification and claim retrieval are vital in combating disinformation. They help assess information credibility by determining its stance on a claim. These tasks streamline fact-checking, making it efficient to locate and analyze relevant claims, ultimately aiding researchers in maintaining information integrity against the challenges of disinformation.

Online disinformation poses a significant and multifaceted challenge to our society, capable of causing harm across economic, social, and health dimensions. A prime example is the COVID-19 pandemic, where the World Health Organization coined the term (mis)“infodemic“ to describe the rapid dissemination of information, sometimes even surpassing the pace of the virus itself. Disinformation can manifest organically, originating from civil individuals without explicit harmful intent, or it can be organized, strategically created, disseminated, and promoted by groups with specific agendas.

Disinformation narratives can evolve and reemerge over time and also be disseminated in multiple languages in multiple countries, transcending language barriers and national borders. This complexity adds a layer of difficulty for fact-checkers and journalists, who grapple with the overwhelming volume of information in need of verification.

Consequently, there is an urgent need for automated approaches that can process and analyze online disinformation narratives, accommodating diverse languages and placing the user at the center of their design. Multilingual capabilities are crucial in studying disinformation, given that false narratives can originate in one language and be adapted to others.

The ExU project (AI Models for Examining Multilingual Disinformation Narratives and Understanding their Spread) is dedicated to developing AI-based models specifically tailored for multilingual disinformation analysis. It focuses on crucial tasks such as rumor stance classification and claim retrieval, directly relevant to the fact-checking process.

ExU’s project partners, USFD and KInIT will develop cutting-edge AI tools, while also bringing together relevant journalists and fact-checkers contacts through their participation in EDMO hubs and other HORIZON EU projects. Beyond English, ExU will engage with a set of 20+ languages, providing evaluation frameworks for at least seven languages, including Portuguese, Spanish, French, Hindi, Polish, Slovak and Czech.

The novelty of ExU project is two-folded:

It is the first effort on multilingual disinformation analysis considering various language families, relevant to the European scenario.

It proposes a user-centric evaluation built on the top of state-of-the-art generative language models such as ChatGPT or GPT-4, with a primary focus on delivering relevant information to end-users for a comprehensive understanding of the ExU model outputs.

Notably, the ExU project goes beyond existing approaches by incorporating considerations for various language families, especially relevant in the European context, and by leveraging the explainability and knowledge base offered by state-of-the-art generative language models like ChatGPT or GPT-4. This multifaceted approach enhances the project’s potential to significantly contribute to the development of more effective and comprehensive solutions for tackling disinformation.

This project is highly relevant to the EU and EDMO hubs, since it addresses languages spoken in multiple EU countries. It is also relevant to other EU projects addressing disinformation analysis, (e.g. HORIZON EU vera.ai and VIGILANT), since its results will be freely available to the research community, facilitating adaptation to specific project domains and contributing to the collective efforts against disinformation.