No questions in this category yet.
Q01
Interpreting
Open for replies
Can ASL-to-text AI ever be accurate enough to replace human interpreters in high-stakes settings — medical, legal, employment?
Automatic sign language recognition keeps hitting benchmarks on controlled datasets, but real clinics, courtrooms, and hiring panels aren't controlled. Where's the line between "useful assist" and "dangerous substitute"? What has to be true for you to trust it — ever?
Q02
Data & Consent
Open for replies
Who owns the data when Deaf signers train AI models — and how do we stop our language from being extracted without consent?
Every major sign language model needs thousands of hours of signing video. A lot of it is being scraped from YouTube, TikTok, and Vimeo right now. What should informed consent actually look like for a community whose language IS visual identity?
Q03
Signing Avatars
Open for replies
If AI generates a signing avatar from a text prompt, does it still count as ASL — or is it something else entirely?
Facial grammar, body lean, rhythm, regional variation, Deaf cultural register. Most synthetic signers are missing 70% of what ASL actually is. Is a "good enough" avatar a bridge — or a replacement that erases native signers?
Q04
Pricing & Access
Open for replies
If AI interpreting and avatar tools end up costing more than human interpreters, who can actually afford them — and what happened to the promise of better access?
The original pitch for AI in the Deaf space was clear: better access, lower cost, real scalability. But early pricing signals across several vendors suggest AI-powered interpreting may land significantly above human interpreter rates — putting it out of reach for small Deaf-owned businesses, nonprofits, schools, and community orgs. Pricing also tends to be absent from demos and marketing, which makes it hard for the community to plan, compare, or push back. What does accessible pricing actually look like — and how do we hold the whole space accountable to it?
Q05
Ownership & Leadership
Open for replies
Most sign language AI companies are hearing-led. What does it actually look like to be Deaf-led — not just "Deaf-consulted" — when it comes to decision-making power, pay, equity, and who owns the upside?
A lot of AI-for-Deaf companies have a Deaf advisor slide, a focus group photo, or a community listening session. Few have Deaf founders, Deaf board members, Deaf executives on payroll, or equity pathways for the Deaf signers whose data and language built the product. The community has been invited into the room as users and consultants — rarely as stakeholders. What changes when Deaf people have actual ownership and authority, not just a seat at the table?
Q06
History & Adoption
Open for replies
VRI promised 24/7 access and got rushed into hospitals — then Deaf patients paid the price. If we're about to do the same thing with AI avatars, what's the lesson we refuse to learn again?
Video Remote Interpreting was adopted fast, marketed hard, and billed as the future of access. A decade later, the Deaf community is still fighting hospitals and agencies over VRI quality, appropriateness, and consent — frozen screens, wrong language pairs, interpreters subbed in mid-emergency. Now AI avatars are rolling out on the same curve: rapid demos, rapid deployment, with the Deaf community often brought in late or as a marketing photo. What specifically went wrong with VRI that we have to call out now, before AI follows the same path?
Q07
Maturity & Readiness
Open for replies
Where is sign language AI actually usable today — and where is it still a hard "not yet"? Who gets to draw that line clearly enough that vendors can't blur it in a demo?
The gap between a polished demo and real-world deployment is where Deaf users get hurt. Sign language AI may be ready for low-stakes captioning, fun consumer apps, or internal drafting — and nowhere near ready for medical consent, legal proceedings, or classroom instruction. But there's no shared standard for where the line is, and vendors have every incentive to blur it. What does a clear, community-backed readiness framework look like?
Q08
Data & Validation
Open for replies
If AI systems trained on ASL gloss are shipping "confidently wrong" translations, what does good training data actually look like — and who validates it before it ships?
Gloss isn't ASL. Strip away facial grammar, classifiers, spatial reference, role shift, and cultural register, and you get a model that's fluent in a language that doesn't exist — and confident while it's wrong. "Good data" is more than hours of video. It's validated, diverse, context-rich, and reviewed by Deaf ASL experts with authority to reject bad output. What should that pipeline actually require — and who holds the pen?
Q09
Culture & Craft
Open for replies
Avatars can produce signs — but ASL is art. Rhythm, face, timing, cultural register. In what contexts is a synthetic signer appropriate, and where does treating ASL as "output" cross a line?
A generated avatar can technically produce a string of signs. What it usually can't do is carry the artistry — the poetry, the humor, the cultural weight, the signer's voice. Treating ASL as "output" flattens a living language into a product feature. Where are avatars a reasonable tool, where are they inappropriate by default, and what guardrails should be non-negotiable so ASL expression stays art, not just output?
Q10
Policy & Procurement
Open for replies
When a hospital, school district, or federal agency buys an AI accessibility tool to satisfy ADA (or AODA, or EAA) — is that real access, or just a checkbox that lets them stop hiring interpreters? Who's supposed to catch the difference?
Accessibility law tells institutions what to provide but rarely how to verify it works. That gap is where AI gets dangerous: a shiny demo, a purchase order, a compliance checkbox — and the Deaf user on the receiving end is the one who discovers the tool doesn't actually work. Procurement officers aren't ASL experts. Regulators move slowly. Vendors market aggressively. What does accountable procurement for AI accessibility look like — and whose job is it to define the floor?
Q11
Labels & Taxonomy
Open for replies
"Avatar" is being stretched to cover everything from learned-data signing generation to 3D models to video-to-video skin overlays on a human signer. These are fundamentally different technologies. Who gets to define the terms — and should conferences require presenters to label their tech honestly?
In the sign language AI space, "avatar" now covers at least three very different things: true signing generation from learned data, 3D rigged models, and video-to-video tools that take a real human performance and swap the skin or face. Lumping them under one word makes it impossible for the community, researchers, procurement officers, or even funders to compare what's actually being built. Without shared definitions — and without venues enforcing them — misleading labels become the norm and trust erodes. What should the taxonomy look like, and whose job is it to hold the line?
Q12
Representation & Disclosure
Open for replies
If a company uses a single signer's footage and overlays different skin tones or faces to show "diverse avatars" without disclosure, is that inclusion — or tokenism dressed up as progress?
Diverse representation matters. But when "diversity" in signing avatars is produced by swapping skin tones or faces on top of a single performer's body and signing, the result can look inclusive while hiding a homogeneous source. Without disclosure, audiences can't tell the difference — and the Deaf BIPOC signers who would actually bring that representation are left out of the frame. What should disclosure look like, and where is the line between stylistic choice and misrepresentation?