What problem does it solve?
Public bodies and transport operators publish a constant stream of information as video and text: disruption updates, safety announcements, policy explainers, training content. Two recent projects show the gap this can leave. The Port Authority of New York and New Jersey found that customer facing information, such as how to use its Bus Terminal and where to find accessibility services at JFK Airport, "was not available in American Sign Language". Rail Delivery Group and Network Rail found that making their Visual Disruption Map videos accessible to Deaf passengers and passengers who experience hearing loss "presented a challenge". Signed video is usually produced by filming a human interpreter, editing that footage into the content, and filming again whenever the content changes.
A human interpreter remains the right choice for anything live, sensitive or contested, such as a court hearing, a benefits appeal or a safety critical announcement that needs a named, accountable signer. AI translation suits routine, repeatedly updated content instead, such as a disruption map that changes hourly, an internal training module or a webpage of standard information. Signapse, the vendor behind both deployments on this page, says the Rail Delivery Group trial showed BSL translations "could be delivered quickly and at scale".
How does it work?
- Capture the source content. Text, a script, or an existing video and its captions are prepared as the input, in the same way for every update.
- Generate the sign language video. An AI model renders a digital signer performing the content in the target sign language, either as a standalone video or as an overlay on existing video and web content.
- Review for sector language. A Deaf reviewer or sign language translator checks the output for technical terms and sector specific vocabulary before it is published, and corrects the glossary for next time.
- Publish across channels. The signed video is embedded on the website, shown on station or venue screens, or attached to the training platform, alongside the original text and audio.
- Keep a human path open. Where the law or the situation calls for a certified interpreter, for example a live safety announcement or a legal process, staff can still book one; the AI avatar does not replace that route.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Emerging
- Channels
- Web chat, Kiosk and branch, Internal tools
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
No public deployment has disclosed a measurable outcome yet.
Value drivers: Inclusion and access, Lower cost to serve, Customer experience.
Indicative value
A transport operator that publishes 50 passenger information videos a year
USD 13,000 to USD 34,000
Sign language video production cost avoided per year
How this is calculated
Formula: videos * (humanCost - aiCost). The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Information or training videos signed per year videos, videos per year | 50 | 50 | The reference operator. Editorial assumption, replace with your own volume. |
| Cost to film, edit and caption one video with a human sign language interpreter humanCost, USD per video | 300 | 800 | Editorial assumption, replace with your own production and interpreter day rate. |
| Cost to produce the same video with an AI sign language avatar service aiCost, USD per video | 40 | 120 | Editorial assumption, replace with your own vendor quote. |
What it leaves out: Covers per video production cost only. It leaves out the platform subscription, the review time of a Deaf translator, and the value of reaching Deaf customers who currently get no signed version at all. It also assumes the content is suitable for AI translation rather than a certified human interpreter.
Who already uses it?
2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Port Authority of New York and New Jersey
United States · Government and public sector · 2024
The Port Authority of New York and New Jersey found that customer facing information, such as how to use the Port Authority Bus Terminal and where to find accessibility services at John F. Kennedy International Airport, was not available in American Sign Language. After selecting Signapse through the 2023 Transit Tech Lab, it ran a proof of concept and a year long pilot of a website translation widget, built on Signapse's generative AI powered sign language software, that automatically translates page text into American Sign Language on request, starting with the JFK Airport Accessibility Services page and extending to Bus Terminal information.
No outcome disclosed.
Rail Delivery Group
United Kingdom · Logistics and transportation · 2024
Rail Delivery Group and Network Rail jointly run the Smarter Information, Smarter Journeys programme, which publishes short Visual Disruption Map videos explaining rail service changes and alternative routes. In late 2023 Rail Delivery Group trialled AI generated British Sign Language translation of these videos with Signapse, and by the end of 2024 had fully rolled out BSL translated Visual Disruption Maps across station screens, National Rail Enquiries, train operating company channels and social media, with staff requesting new translations weekly through Signapse's SignStudio platform.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Source text or video with accurate captions to translate from
- A glossary of technical or sector terms in the target sign language
- A policy on which content types require a certified human interpreter
Systems to integrate
- Content management system or website, for an embeddable translation widget
- Video or e learning platform, to attach signed video to training modules
- Digital signage or kiosk screens at stations, branches or venues
Complexity: Low
The AI translation itself is a vendor hosted service reached by API or a website widget; no core system integration is required. The work is in building a sign language glossary for sector terms, setting a review step with a Deaf translator, and deciding which content must stay with a certified human interpreter.
- 1
Start with high volume, low stakes content
Pick routine, frequently updated content, such as training modules or a disruption information page, rather than safety critical announcements.
- 2
Build the glossary before scaling
Load sector specific terms and place names so the avatar signs them consistently, and review the first batch closely with a Deaf translator.
- 3
Decide the human interpreter boundary
Write down which content always needs a certified human interpreter, such as live safety announcements or anything used in a legal or disciplinary process, and route it there.
- 4
Publish alongside the original
Keep the source text, audio or video next to the signed version, and label the video as AI generated, so a Deaf viewer can tell what they are watching.
- 5
Collect feedback from Deaf users
Ask the Deaf and hard of hearing people the content is for whether the signing is clear, and route unclear terms back into the glossary.
Guardrails
- Visible label that the signed video is AI generated
- A named boundary for content that requires a certified human interpreter
- A maintained glossary reviewed by a Deaf translator, not by the vendor alone
- The source text, audio or video kept alongside the signed version
KPIs to instrument
- Share of published content that also has a signed version, by content type
- Turnaround time from source content to published signed video
- Feedback from Deaf users on clarity, by content type
- Cost per video against the previous human interpreter process
Human in the loop
A Deaf reviewer or sign language translator checks new content types and glossary terms before publication. Staff can still book a certified human interpreter for any content, and must for content the organization has classified as requiring one.
Common failure modes
- Avatar signing seen as inferior to a human interpreter
- Some Deaf viewers find avatar signing less natural or harder to follow than a human interpreter, particularly for nuanced or emotionally charged content. Reserve AI translation for routine content and keep a human interpreter route open on request.
- Sector terms signed inconsistently or wrong
- Generic models miss place names, product names and technical terms specific to the organization. Maintain a reviewed glossary and re check it whenever new terminology appears.
- Treated as a replacement rather than an addition
- Rolling out an AI avatar can be read as withdrawing human interpreters. State explicitly which content keeps a human interpreter and keep that commitment visible.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
The translation service generates synthetic sign language video, so under Article 50(2) the provider must mark that output as artificially generated in a machine readable format. A deployer must disclose the video as AI generated under Article 50(4) only where the signer is photorealistic enough to count as a deep fake: content that resembles an existing person and would falsely appear authentic. There is no human review or editorial responsibility exemption for video; that Article 50(4) exemption applies only to generated text published in the public interest. Article 50(1), the duty to tell someone they are dealing with AI, applies only if the avatar interacts with a person directly, for example in a conversational digital human channel; a video played on a screen or on request does not by itself trigger that duty.
Rules that apply
Guidance
- European Accessibility Act, accessibility requirements for services (European Union, Europe). Sets accessibility requirements for transport and other services in scope from June 2025, which sign language content can help organizations meet without itself being a compliance guarantee.
- Article 50, transparency obligations for providers and deployers of certain AI systems (Future of Life Institute (EU AI Act explainer site), Europe). Requires providers to mark synthetic video output as artificially generated (Article 50(2)); requires deployers to disclose AI generated or manipulated video that counts as a deep fake (Article 50(4)); and requires that people who interact directly with an AI system are told they are doing so (Article 50(1)). The text only disclosure exemption for content under human review and editorial responsibility does not extend to video.
Controls to put in place
- Register entry for the translation tool with an accountable owner
- Written policy on which content requires a certified human interpreter
- Glossary maintained and signed off by a Deaf translator
- Visible AI generated label on every signed video
Frequently asked questions
- Can AI sign language avatars replace human interpreters?
- Not for everything. They suit routine, frequently updated content, such as a disruption information video or an internal training module. Live, sensitive or legally significant content, such as a safety announcement or a court hearing, should keep a certified human interpreter.
- Which organizations have deployed AI sign language avatars?
- Rail Delivery Group and Network Rail rolled out AI driven British Sign Language translations of passenger disruption information videos across station screens, National Rail Enquiries, train operator channels and social media, reaching full coverage by the end of 2024. The Port Authority of New York and New Jersey piloted a website widget, built on Signapse's generative AI translation technology, that renders American Sign Language on request for pages about JFK Airport accessibility services and its Bus Terminal.
- What does it cost compared with filming a human interpreter?
- Filming, editing and captioning a video with a human sign language interpreter requires a new shoot every time the content changes. We found no public source that compares the cost of that process with an AI translation service, so this page does not state which option costs less. The figures in indicativeValue above are editorial assumptions, not a benchmark from either case study: replace them with your own vendor quote and interpreter day rate before using them in a business case.
How to cite this page
Blits.ai AI Use Case Library, "AI sign language avatar translation for public information", last verified 29 September 2026, https://www.blits.ai/ai-use-cases/sign-language-avatar-translation. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 30 September 2026: First published