| Acronym: Ouvia |
| Main Objective: This project aims to address disparities in speech translation (ST) technologies across different demographic groups by: 1. Creating a new multilingual speech translation benchmark that evaluates both technical performance and real-world usability 2. Investigating how ST tools perform across different sociodemographic groups, particularly focusing on gender and ethnicity 3. Simulating both high-stakes and low-stakes interaction scenarios 4. Collecting 3000 segments (25 hours) of English→Portuguese translations with quality and usability assessments from 450+ diverse participants The research specifically examines whether users find ST tools usable in real scenarios, how well automatic metrics reflect actual usability, and whether there are systematic performance gaps across demographic groups. |
| Reference: Non Applicable (NA) |
| Funding: European Association for Machine Translation |
| Start Date: 02-05-2025 |
| End Date: 02-05-2026 |
| Team: Giuseppe Attanasio, André Filipe Torres Martins |
| Groups: Pattern and Image Analysis – Lx |
| Partners: Google |
| Local Coordinator: Giuseppe Attanasio |
|
|
|
Associated Publications
|
