The Signapse AI Roadmap: Our Journey to Human-Level Sign Language Translation
Signapse
Our goal at Signapse is to use state-of-the-art AI to make sign language high-quality, accessible, and available wherever it's needed. Although we're immensely proud of the progress we've made thus far, we're also honest about where we are, where we're headed, and how far we still have to go.
We developed a Sign Language Production Framework (SLPF) to help direct that journey. It measures how well digital signers, such as our AI signer Rae, use sign language in authentic situations. Consider it a sat nav for AI development that indicates our starting point, current location, and necessary steps to get there.
Because, let's be honest, Deaf communities are entitled to more than keyword-based videos and robotic avatars. We're creating something better.
Understanding the Framework: What Is a Tier?
Our AI signer is assessed on a seven-tier scale by the Sign Language Production Framework (SLPF). Each tier and the associated definition and attributes can be used to describe both technological advances (existing and in development) along with the key requirements from technology as identified from user feedback. It is not a substitute for accuracy and understandability, which needs to be measured separately. Tier 0: No Automation
Using a human translator with no AI assistance. Quality is guaranteed, but access to translators means creating accessible content is slow and expensive.
Tier 1: Just Getting Started
Here the translation is still performed by a human, but they can use AI tools to make their task easier.
Tier 2: Beginning to Make Sense (Our Current Level)
This is where things start to improve. Our AI signer looks photo-realistic, moves fluidly, and uses some facial expressions and mouthings. It can sign short phrases and has begun applying sign language grammar, although it still leans heavily on English word order (also known as Sign Supported English – SSE).
Challenges at this stage include:
- Understanding BSL grammar rules like topic-comment structure and role shifts
- Consistent use of space and direction for clarity
- Reducing the need for manual checking (currently essential)
Right now, our AI translations sit at around 60% understandability** on their own. To ensure quality, every video produced through SignStudio is reviewed by a human. This review and quality control process brings the final output to 100% understandability, combining the speed of AI with the expertise of Deaf translators. A typical 10-minute video currently takes under 3 days from input to final output.
** How We Measure Understandability
Our accuracy score reflects how closely Rae’s signing matches natural, fluent BSL. To measure it, we work with Deaf sign language experts who:
- Watch AI-generated videos
- Compare them to the intended meaning
- Rate the signing on clarity, grammar, and expression.
Each video is scored using a detailed framework that checks:
- Whether Rae uses the correct signs and facial expressions
- If the grammar and structure match natural BSL
- Whether the message is understood easily by Deaf viewers.
These expert reviews are then used to calculate a percentage.
What’s Next: Our Roadmap to Tier 7
We're on a clear path to evolve our AI signer from Tier 2 to Tier 7, mapped against a robust Sign Language Production (SLP) model. That model includes decision points like noun/verb disambiguation, who is involved, and what emotions need to be expressed.
While our general AI signer Rae is currently at Tier 2, we’ve already reached Tier 6 in the transport sector, where we use Marcel’s appearance to generate accurate and expressive train messaging. This shows what’s possible and motivates our ongoing work to bring Rae to the same high standard for more general use.
Here’s how we’ll get there:
Stage 1: Foundational (Now – Tier 2)
- Signer looks natural and uses some non-manual features
- Relies on English grammar order
- Manual review still required
- Working toward improved grammar, placement, and consistency
Upcoming improvements:
- Grammar and sign order: expected within 6 months
- Space and direction handling: within 1–2 months
Stage 2: Transitional (August – Tier 3/4)
This is where our signer begins to feel genuinely fluent:
- 80% understandability
- 1-day turnaround
- Full use of BSL/ASL grammar structures
- Directional verbs and better contextual placement
- Improved emotional expression and fluidity
Goal: AI translation that doesn’t rely on spoken English grammar and instead speaks true BSL or ASL.
Stage 3: Advanced (December – Tier 5/6)
- 98% understandability
- Instant translation
- Deep understanding of context and paragraph-level meaning
- Expressive, nuanced signing with accurate role shifts
- Feels like a real human conversation
Goal: Reach near-human performance, sign language translation that requires minimal or no manual intervention.
Why This Matters
We’ve all seen sign language content online that misses the mark, confusing at best and misleading at worst. This roadmap and proficiency framework are our way of raising the standard.
It helps:
- Deaf communities clearly understand the quality and limitations of the technology
- Developers know what “good” really looks like
- Businesses and organisations decide when and where the tech is ready to use
Putting It Into Practice
Our digital signer Rae will be regularly assessed using this framework. The progress will be shared publicly as we push for Tier 6/7-level fluency across transport, healthcare, education and beyond.
We’re also exploring ways to:
- Reduce manual checks in low-risk settings (like marketing content)
- Involve Deaf communities in user testing
- Make continuous improvements based on real-world feedback
We’d Love Your Input
This journey isn’t one we can, or should, take alone. If you’re Deaf, Hard of Hearing, a sign language professional, or someone who works in accessibility, we’d love to hear from you.
Tell us:
- What’s working
- What still needs improvement
- What you’d like to see next
Together, we can build a future where sign language access is instant, seamless, and truly inclusive.
Wrapping It All Up
This framework is more than just a list of technical phases. It's a dedication to improving. A road map for the future. And a call to action for real accessibility, created by and for the Deaf community.
Subtitles won't do for us. We're creating something better. And we get there like this.
Want to see our full research on AI translation tiers? → Read the full document here
