In brief
Three things to keep straight
- Participation and authority are different.
- A benchmark cannot define culturally acceptable failure by itself.
- Sign First’s technical safeguards do not replace external Deaf-led evidence.
What current research says
A mixed-method study of Deaf and hard-of-hearing users found that text AI can support communication and information access while also presenting challenges involving ASL support and Deaf cultural understanding.1
A Sign-Speak pilot states that involving a Deaf researcher made key study insights possible and recommends designing with, rather than for, the Deaf and hard-of-hearing community. It also records limits to the pilot’s duration and deployment evidence.2
Google says Deaf perspectives shaped SL2T from conceptualization through data collection, user studies, impact assessment, and an advisory committee. That is a substantive governance claim; its exact authority and future operation should remain inspectable as the product expands.3
What governance must be able to change
Community leadership should be able to redefine success, reject a data source, narrow an intended use, identify harmful defaults, require an abstention, or stop a release. Consultation that cannot change the product is feedback, not authority.
Age, race, gender, region, disability, signing background, language history, camera access, and educational setting can all shape failures. “Deaf-vetted” is not a substitute for publishing who evaluated what, under which conditions, and with what decision rights.
The current Sign First boundary
Sign First has technical governance controls: purpose separation, revocation behavior, visible evidence boundaries, and rejection of unsupported recognition claims. Those are necessary and insufficient.4
The public site must continue to say that Deaf-led, age-specific external validation is open. The correct response to that gap is to preserve a testable platform and a precise protocol—not to relabel internal ACE checks as community approval.4
Questions answered
The short version
Is a Deaf advisory board enough?
Only if its members have clear scope, access to evidence, meaningful decision rights, and a continuing role. The label alone proves little.
Has Sign First completed Deaf-led student validation?
No. The platform is a working research system; student outcomes and age-specific usability remain unvalidated.4
Why publish before that study?
A bounded public research record can make the product, claims, methods, and open questions inspectable before asking people to participate. It cannot be marketed as completed validation.
Source register
What this article relies on
- “We do use it, but not how hearing people think”ACM CHI EA 2025 ↗
Mixed-method study of 80 Deaf and hard-of-hearing participants plus 11 interviews.
- Evaluating an AI Bi-Directional System for CommunicationACM CHI EA 2025 / Sign-Speak ↗
Pilot case study; short-term usability evidence, not a classroom outcome study.
- Putting sign language AI into users’ handsGoogle DeepMind ↗
Primary product announcement and stated limitations for SL2T on Pixel 11.
- Evidence registerSign First ↗
Public, bounded record of internal tests, failures, and current product limits.