Exceeding the floor at scale is fundamentally a production problem, and it has a historical precedent. Before the printing press, each book was copied by hand, one word at a time. The limiting factor was not the skill of the scribes but the physical capacity to produce copies. A single volume could take months, so texts remained concentrated among the few who possessed copies. The expansion of literacy that followed the press was not primarily a change in the demand for reading but a change in the cost of reproduction: the press separated the availability of knowledge from the speed of the copying hand.10
Sign language translation has remained in an analogous pre-print condition, referred to here as the Scribal Era. The constraint has never been the translators, who are highly skilled linguists performing rigorous work, but the absence of infrastructure capable of scaling that work.
Producing a single finished minute still requires hours of work surrounding the translation itself, regardless of the source material. This aggregate overhead is referred to here as the Studio Tax. Its visible component is equipment: the studio, lighting, camera, green screen, and editing workstation. Its larger component is the surrounding workflow, meaning the staff time required to move a project from intake to delivery, most of which is spent on tasks other than translation. This workflow routinely costs more than the translation work itself.
Scale here represents the ongoing capacity for continuous translation—rendering into every required delivery format across a dynamic, growing library of content—rather than a single massive project. In practice, this increases the completed minutes generated by each translator.
With this infrastructure, a translator records in an ordinary room, without a green screen or specialized lighting, and the platform performs the remaining steps: background removal, framing, rendering, timeline synchronization, and delivery. Work that once required a physical studio now operates in software.
Once translation and production are separated in this way, the principal bottleneck is removed. Output is no longer limited by production hours, and cost tracks the translation work rather than the studio infrastructure around it. Total output across the field can increase substantially.
The categories of content affected are familiar: workplace and HR materials, healthcare and public-service content, educational materials, and customer-facing and public information, among others. As production cost declines, such content becomes feasible for organizations to make accessible.
Historically, access to many of these materials has depended on engaging a qualified interpreter each time. This approach can be resource-intensive and difficult to scale. By contrast, translating content once and making it available on demand can significantly reduce ongoing costs while enabling skilled interpreters to focus on situations that require live, interactive communication. This is particularly valuable given the continuing demand for qualified sign language interpreters across many regions and service domains.
There is a commercial argument alongside the ethical one. In the United States, deaf workers are employed at a rate of approximately 58 percent, more than 15 points below the rate for hearing adults, and much of that gap reflects access rather than ability.11 When the information a role depends on, such as onboarding, training, policies, and routine communication, is available in a deaf employee’s primary language, that employee can learn the role, perform it, and advance on equal terms. Employees who are not required to work around inaccessible systems tend to be more productive and more likely to remain. Retaining a productive employee also avoids the cost of replacement, which is estimated at between half and twice annual salary.12 As the cost of signed translation falls, the question shifts from whether an organization can afford access to whether it can afford to forgo it. The following section addresses where AI belongs in this work and where it does not.