We All Love You But… Redesigning Translation Quality Frameworks for Multimodal AI-Generated Content


Track: Multilingual AI | AI8 |   Intermediate |
Wednesday, October 30, 2024, 2:30pm – 3:15pm
Held in: Steinbeck 3
Presenters:
Agustín Da Fieno Delucchi - Microsoft 
Katerina Gašová - Argos Multilingual 
Gilberto Segura - PGLS
Moderator: Pavel Soukenik

This panel will discuss the fundamental concept of quality for AI-generated content. This is not about a simplistic set of metrics; we need a holistic framework that scales and is specific to the task. We see new technologies that rely on well-established quality frameworks (e.g., MQM). But these can’t be readily applied to a world where there’s no source, the volumes are enormous, it’s not text only, and everything happens immediately. We’ll compare traditional approaches to translation quality with how to effectively assess quality of multimodal AI-generated content. Can our industry build a new framework: what can be reused and what needs to be invented? When, if not now?

Takeaways

Examples of evaluating Automated Post Editing and AI-assisted LQA; How to start developing a holistic standard for multimodal communication [speech, text, video] without a source; How to build a new world that merges old concepts with new technologies, in various contexts.