University of Sheffield presents work on synthetic text detection, credibility signals and OCR in EBU webinar series no. 3

In the third EBU webinar of our vera.ai series that ran on 26 November 2024, Olesya Razuvayevskaya and Carolina Scarton from the Natural Language Processing Research Group at the University of Sheffield (USFD) discussed their work on developing AI-based text analysis tools to verify information.

Textual misinformation can be detected through various semantic, stylistic, and linguistic clues, often referred to as so-called credibility signals. 

In this webinar, USFD explored three types of such signals:

• framing (presenting information from a particular perspective), 
• persuasion techniques (pragmatic strategies used to influence and potentially mislead the audience), and 
• genre (covering types such as objective, opinionated, and satirical news). 

Olesya and Carolina delved into the methods developed within the vera.ai project to identify these signals. They furthermore demonstrated publicly available tools for solving these tasks. 

Additionally, USFD introduced their work on a multilingual Optical Character Recognition (OCR) tool, commonly utilized by media professionals to convert text within images into machine-encoded text for easier analysis and understanding.

Below you can view a recording of the entire event.

This is the last webinar of the series in 2024. However, we won't stop and continue in 2025. 

So please join us on 20 January 2025 when the series resumes for the new year with a webinar showcasing KInIT’s work with Large Language Models (LLMs) to tackle online disinformation. 

Register here and mark your calendars - and join us in January.

Until then we wish you a happy holiday period and hope to see you again in 2025!

Author: Lalya Gaye (EBU)

Editor: Jochen Spangenberg (DW)