What producing in Kaohsiung taught me about supertitles, latency, and the most honest test any technology will ever face.
There is a particular seat in every theater that nobody envies: the producer’s. You don’t watch the show from it. You watch everything that could end the show.
On October 12, 2025, I was in that seat at Weiwuying — the National Kaohsiung Center for the Arts, and if you haven’t seen it, imagine if a banyan grove decided to become an opera house — for Kamata March Finale: Gin’s Departure (蒲田進行曲完結篇～銀之死), a co-production of IPAC and Ariel Imaginarium. The Kamata March story is so beloved in Japan it needs no introduction there. But here is the detail every theater person should savor: this finale — the Japanese stage play 『蒲田行進曲完結編 銀ちゃんが逝く』 — has been staged by many Japanese directors over the years and has never once been made into a film. It is a story that exists only on stage. Where else would we premiere its Taiwan production but a stage — a Japanese theatrical legend, in Kaohsiung, for an audience reading along in Mandarin.
It was not a perfect production. I say that as its producer, which is the only seat from which you’re allowed to say it. And precisely because it wasn’t perfect, I’ve been thinking ever since about the one department that fought hardest that night and got the least applause.
The supertitles.
The most honest technology in the house
Here is a thing civilians don’t know: every other technology in a theater can be rehearsed into submission. Light cues are deterministic. Automation runs on rails, literally. Sound has a console and a person who has mixed the show forty times.
Surtitles answer to none of that. They answer to the actor.
An actor takes a beat half a second early, and the supertitle spoils the line before it’s spoken. Half a second late, and the joke dies in the air between the stage and the balcony — you can actually hear where it lands, in the wrong-language half of the house, four seconds after everyone else. Any operator who has run titles for a comedy knows this particular heartbreak.
Now run that across two languages with opposite grammar. The audience at Weiwuying was hearing Japanese emotional rhythm and reading Mandarin sentence structure. Kansai-dialect wordplay, the kind that makes a Japanese house roar, arrives in Mandarin as a corpse unless someone who understands the scene — not just the sentence — rebuilt it. Translation is not the job. Interpretation is the job. Every director already knows this; we just rarely say it about the titles.
Film gets another take. Streaming has a pause button. The stage gives you fifty milliseconds and no apologies. I’ve come to put it this way: live performance is the hardest room any real-time technology will ever play. It started as a producer’s complaint. It has since become my business plan.
The confession portion of the column
I should come clean about my resume, because it confuses everyone at intermission.
I am a statistician by training. I spent sixteen years in the United States doing people analytics — measuring how organizations understand human beings, and how confidently they get it wrong. Then I came home to Taiwan and became a filmmaker and theater producer. Then I founded an edge-AI company.
Three careers. Everyone assumes there’s a rupture in there somewhere. There isn’t. All three are the same job: making sure a human being is correctly understood — by an institution, by a camera, by an audience, in real time.
Statistics taught me that data should be processed where it happens; every mile you move it costs you truth. Theater taught me that understanding isn’t translation, it’s cultural fluency. Put those two lessons in a rehearsal room together and they produce, almost inevitably, the thing my company builds.
The lawn where it actually began
But the honest origin story isn’t a rehearsal room, and it isn’t Weiwuying. It’s an amphitheater in California.
During my American years, I took a group of hearing-impaired students to see A Midsummer Night’s Dream at Stanford’s Frost Amphitheater, at a Shakespeare festival. We had studied the play together. They knew the plot cold — Puck, the lovers, the flower, the whole enchanted mess. They were, by any measure, prepared.
Afterward, one of my students gave me the sharpest production note I have ever received:
“I can understand the play. But I cannot feel it.”
Sit with that sentence the way I’ve had to. The captions had delivered the meaning. The theater had given the music — the timing, the tease, the ache under the verse — to everyone else in the house. My students got the plot. The rest of the audience got the play.
That is the night this product was born, years before a single line of code. Not as translation technology — as inclusion. Access to art means access to the feeling, or it isn’t access at all. It’s why, much later, our patents cover emotion annotation: the system doesn’t just carry what a line says, it carries how the line is said. Accessibility isn’t a feature we added to Supertitles. It’s the reason Supertitles exist.
The part where I tell you what it is
Storytelling Supertitles™ is on-device, AI-driven multilingual titling for live performance — patented in Taiwan and Japan (utility models), with an invention patent pending in Japan. We demonstrated it in live performance at Weiwuying. Taiwan’s public broadcaster PTS ran a story whose headline I’ll be quoting for the rest of my career: watching it live, they said, is “like reading a comic.”
The architecture is one sentence long: everything happens on the device. Audio is processed locally and discarded — nothing uploaded, nothing retained, no cloud in the loop. If the internet dies mid-act, the show does not. Each display serves one language, delivered to the audience that needs it — audience-targeted delivery, not simultaneous broadcast.
Theater people tend to like this architecture before I finish explaining it, and I’ve realized why. The theater is one of the last rooms in modern life where you are not being recorded, and audiences can feel it — that’s part of the covenant of presence, the reason we still gather in the dark with strangers. A titling system that phones home would break something it was hired to protect. Ours is built so there is nothing to phone home.
And let me be precise, because precision is a producer’s virtue: this system has been demonstrated, not deployed. The engineering is still maturing. Theater people, of all people, understand the difference — an opening night is not a sit-down run.
Nobody loses their job in this column
The first question, every time, in every country: who does this replace?
Nobody. And I mean that structurally, not diplomatically. The system trains on rehearsal data — it learns the production the way a titles operator does, by being in the room — and it follows the actors rather than making the actors follow it. The best supertitle work has always been the kind the audience never notices. Augmentation, not replacement, isn’t our slogan; it’s the design constraint everything else obeys.
What it adds is audience. Every language barrier a house removes is box office it was leaving on the table — tourists, immigrants, the Deaf and hard-of-hearing patrons the titles also serve, the international co-productions that suddenly become bookable. Accessibility keeps being filed under compliance when it belongs under growth.
Since Kamata March, IPAC has also produced The Velveteen Rabbit (絨毛兔之奇幻之旅). Each production is a rep run at the same idea: cross-language live performance as a thing our region does, and does well.
Why I’m telling Broadway this
Because Broadway is the biggest room, and the biggest rooms have the most to gain.
New York plays host to the world and performs mostly in English. Tokyo’s stages are having a remarkable decade and remain, for most visitors, beautifully sealed. Taipei and Kaohsiung sit between the two, producing across languages every season, out of necessity — which is why the necessity became an invention here.
My wager, from the producer’s seat: any technology that can survive eight shows a week — live actors, live laughs, fifty milliseconds, no second take — has passed the hardest test it will ever face. The stage is not a use case for this technology. The stage is where it had to be born.
The curtain goes up every night regardless. That is a powerful instrument. It deserves players who tune it by hand.
Sophia Huang is a statistician, data scientist, AI engineer, filmmaker, and theater producer. She is CEO of Ariel Innovations and founder & president of the International Performing Arts Consortium (IPAC). She writes from Taipei.