DFKI and Inria Launch Project for Enhanced AI Sign Language Translation
DFKI and Inria have launched the RoGSiLT project, a joint initiative aimed at developing advanced AI for robust and natural sign language translation, specifically for German and French sign languages. The project will address limitations in existing systems by improving translation quality and naturalness, utilizing multimodal neural networks, self-supervised learning, and large language models, with the goal of creating a bidirectional translation prototype that realistically portrays content via avatars.
Key Takeaways
- DFKI and Inria are collaborating on RoGSiLT, a three-year project focusing on German Sign Language (DGS) and French Sign Language (LSF).
- The project aims to improve existing AI sign language systems by addressing issues with inaccurate facial expressions, body language, and unnatural representations.
- RoGSiLT will utilize multimodal neural networks, self-supervised learning, and large language models to enhance translation quality and overcome data limitations.
- The initiative will develop new parallel corpora of sign language videos and texts to support training data preparation.
- A bidirectional translation prototype is the project's culmination, featuring an avatar that realistically conveys content, including full-body movements and emotional expression.
Why It Matters
The RoGSiLT project signifies a focused effort to overcome critical limitations in current AI sign language translation, which often struggles with natural expression and nuanced meaning. Improved accuracy and naturalness in avatar-based translation could expand accessibility for deaf and hard-of-hearing individuals, particularly in public information and communication. The success of RoGSiLT will depend on its ability to create widely applicable models that effectively capture the complexities of natural sign language, setting a benchmark for future AI accessibility implementations in streaming and interactive media.
Additional Context
The development of AI for sign language translation is a growing area, with several recent initiatives paralleling RoGSiLT's goals. NYU Abu Dhabi launched ChatSign in May 2026, an AI system providing real-time bidirectional translation between speech, text, and sign languages including Arabic, English, Emirati, and US sign languages (per The National, May 2026). ChatSign aims to be a cost-effective alternative to human interpreters across various public and private sectors. Similarly, in May 2026, Nigerian and UK-based Talksign debuted Palm 1.0 and Echo 1.0, AI models for real-time American Sign Language (ASL) and text/speech translation (per TechCabal, May 2026). Palm 1.0 interprets ASL to text or speech with 84.2% semantic accuracy, while Echo 1.0 generates photorealistic ASL video avatars from written or spoken language. These models build on Talksign's earlier foundation model and address limitations in continuous sentence interpretation. Meanwhile, Kara Technologies and Auckland Transport are preparing AI New Zealand Sign Language avatars for public transport announcements, aiming for a public launch soon (per News Wire, May 2026). This project has spent over two years digitizing unique NZSL gestures and facial expressions, using motion capture and AI to create accurate digital libraries. These efforts underscore a collective push to bridge communication gaps for the deaf and hard-of-hearing community through advanced AI, highlighting a trend toward more natural, accurate, and context-aware sign language translation systems.
Read full article at idw-online.de
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source