Multilinguality in the VIGILANT project

In June 2024, members of the VIGILANT consortium published a short paper titled “Multilinguality in the VIGILANT Project”, as an outcome of the 25th Annual Conference of the European Association for Machine Translation. The article outlines how VIGILANT is tackling one of its key challenges: ensuring that its disinformation-detection tools support the diverse languages used by European law enforcement agencies.

The authors explain why multilinguality is essential for investigating criminal disinformation, which spreads across borders and linguistic communities. Since the VIGILANT project works with Law Enforcement Agencies (LEAs) from various European countries, the platform must handle at least seven languages, with more expected through its Community of Early Adopters.

The paper also summarises the project’s technical strategy, including zero-shot and few-shot learning, machine translation, and data-augmentation approaches—methods that allow powerful English-trained models to be extended to other languages efficiently.

The article was authored by researchers from the University College Dublin, the University of Sheffield, Kempelen Institute of Intelligent Technologies, Sirma AI EAD, Trinity College Dublin, and Gerulata Technologies.

Link to Zenodo: https://zenodo.org/records/17722594