One source, many rooms
Why the right answer to multilingual subtitling is N independent displays, not one stacked display — and how the architecture stays symmetric as you add more languages.
The lazy answer to multilingual subtitling is to stack the languages: one display, three columns of text, one column per language, scaled to whatever fits. It is the lazy answer because it solves the wrong problem.
The audience problem in a multilingual room is not "show me all of the languages at once." It is "show me the language I read, in the same beat as the speaker, big enough to read from where I'm sitting, without anything else competing for my attention." Three languages stacked on one display means two-thirds of every audience member's visual budget is going to text that's not theirs — and the unfamiliar script in their peripheral vision is cognitive load against the language they actually need. Every audience class — the coalition officer, the foreign minister, the shelter survivor, the village elder — deserves better than that.
So we don't stack. We fan out.
The architecture
One speaker, one laptop. The laptop runs the recognition once, runs the translation in parallel into every target language, and emits one subtitle stream per language to its own physical display. Japanese on display 1, Mandarin on display 2, Tagalog on display 3 — each at the size and pace and typographic register native to that language. The operator console runs on the laptop's built-in screen; the speaker's tele-prompt runs on a tablet they can glance at while presenting. Three audiences, three surfaces, three trust boundaries. Nothing leaves the laptop.
Each per-language display is driven independently. The Japanese renderer has no idea what's on the Mandarin display, and vice versa. A renderer crash on one display takes down only that one display; its peers continue to run. A reconnect re-binds the failed renderer to the latest subtitle in its own language without disturbing the rest of the room. The independence is deliberate — failure of one surface should not propagate into others.
Why this shape stays symmetric
The number of languages is a deployment choice, not an architectural choice. Adding a fourth audience language — Tagalog joining Japanese, Mandarin, and Korean — is the same shape. The recognition runs once; the translation pipeline forks into one additional branch; one additional display gets plugged in. The architectural lift is zero. The hardware lift is "another HDMI cable and another monitor."
This matters operationally. A coalition exercise that goes from three partner nations to five does not need a new system; it needs two more displays and a recompiled content package. A disaster shelter that adds Burmese and Tetum to its primary languages does not need a new system; it needs two more displays and the lexicon entries. The system grows the way the room grows — one display at a time, each one identical in shape to the others.
Dignity parity
The phrase we keep coming back to is dignity parity. Every audience class gets a translation surface that treats them as the primary audience of their own screen. The Japanese-reading audience member is not looking at a third of a stacked display next to two languages they don't read; they are looking at a full screen at the size, pace, and weight that fits Japanese. The same for every other language. Nobody is downgraded. Nobody is asked to share their visual budget with somebody else's language.
The same principle drives a less-visible choice: when the speaker steps outside the prepared material, every audience display fades to a single soft blinking dot at the moment the curated translation goes stale. The dot is universal — no language is favored over another in the failure mode. We cover the design of the dot in the next chapter; for now the relevant point is that the architectural symmetry across languages is preserved even when content isn't being shown. Each viewer's experience of the room is consistent with every other viewer's experience of the room, even though they are reading different languages.
One operator. One room. As many languages as displays will fit
The hardware ceiling on a single laptop is the number of independent displays the GPU can drive simultaneously. With modern professional laptops that's three or four external displays plus the built-in. For larger rooms or larger language sets, the architectural answer is "another laptop" — and the fallback continues to fit in one or two bags.
The next chapter covers what happens when the speaker steps outside the prepared material.