Artificial intelligence has reshaped nearly every corner of the film industry. From streamlining production to digitally resurrecting actors, the technology is now capable of altering an actor’s lips and voice so they appear to speak a different language—no subtitles required.
But for deaf and hard of hearing audiences, this kind of innovation misses the point. Subtitling for the deaf and hard of hearing (SDH) isn’t simply about converting speech into text. It is a complex, interpretive art form, one that industry professionals say AI is still unable to master.
“SDH is an art, and people in the industry have no idea,” said Max Deryagin, chair of Subtle, a nonprofit association of freelance subtitlers. “They think it’s just a transcription.”
That perception, subtitlers argue, is precisely why machine learning systems fall short. Unlike standard captions, SDH subtitles must convey more than spoken dialogue. They include sounds, tone, emotion and narrative cues that deepen the audience’s experience.
According to Deryagin and others in the field, this requires human judgment—when to signal that music is swelling ominously, when to capture the pause before a confession or admission of guilt, or how to render the cacophony of overlapping dialogue in a way that feels natural. It’s much more nuanced than copying verbatim, he says.
The nonprofit Subtle, founded in the United Kingdom in 2006, was created to promote quality subtitling and fair pay in an industry that often undervalues its professionals. Despite subtitles being integral to accessibility, many subtitlers are underpaid and bound by punishing deadlines.
A statement from the association emphasizes that subtitling is not just about inclusion for those with hearing loss—it also helps with language learning, literacy, cultural exchange and even day-to-day convenience. “Subtitles connect us by giving us a way to communicate with each other and be understood,” the group wrote.
Yet subtitlers say their craft is poorly understood. Nota Mouzaki, a professional subtitler and Subtle member, explained the hidden hurdles of the work in a blog post. “When actors fire 80 WPM, and you have your usual 259 parameters when subtitling, you never know if your truncation will come back three episodes later to haunt you,” she wrote. “Our translations are important. A perfectly made film or series can be ruined by a bad translation.”
Mouzaki argued that subtitlers juggle hundreds of decisions at once—timing, tone, readability—while striving for what she called “invisibility.” The goal, she wrote, is to make the viewer forget they are even reading.
Other subtitlers describe their work as an intricate puzzle that requires imagination. Simone Taylor, a Brazilian audiovisual translator, recalled her experience creating subtitles for Full Steam Punks, a steampunk series later adapted into a film.
“One aspect of subtitling that immediately drew me in, in terms of translation, was its puzzle aspect,” Taylor wrote. “You need to translate not just words but images; you need to fit them in a tight space, and I absolutely love a challenge of this type.”
The project demanded not just translation but transcreation: inventing new gadget terms, adapting rhymed verses, renaming characters and reworking wordplay so it resonated in English. The work, she said, was “riddled with difficulties from beginning to end.”
For Taylor, the project was not simply about making dialogue legible—it was about preserving meaning, style and atmosphere across cultural and linguistic boundaries.
While machine learning systems can generate captions, they tend to flatten these subtleties into literal transcriptions. They struggle with overlapping dialogue, regional slang, irony or wordplay. Even more critically, they rarely capture the nonverbal dimensions—like the difference between a door creaking ominously and a door slamming in anger.
Despite all the talk about automation, human subtitlers remain indispensable. Maybe one day AI will eliminate the SDH industry, but today the humans element is essential. AI is a powerful tool for certain tasks—syncing text to video, creating draft translations—but professionals insist that it cannot replace the interpretive labor at the heart of subtitling. For millions of deaf and hard of hearing viewers, quality subtitles determine whether a story is truly accessible.
As Mouzaki stated, film can succeed or fail on the strength of its subtitles, a reality AI has yet to grasp.
