An evening clinic
One evening's clinic list included a patient who stood out: a Spanish speaker from South America, in Taiwan on a business trip.
Before his visit he opened Brief on his phone and worked through the questions in Spanish. In the inbox, the doctor tapped the translate/edit chief-complaint button (labelled 翻/修主訴), and the Brief opened already translated into Chinese.
Once they were face to face, the two of them got by fine in English. But before the patient even sat down, the doctor had already read his chief complaint in the language he knows best — what was bothering him, when it started, and what he most wanted to ask.
It was the first time anyone had filled out an iRehab Brief in Spanish. We want to use it to explain how Brief deals with language.
When language gets in the way, the chief complaint goes first
A language barrier in clinic is rarely total. The usual case is people getting by: both sides working in a second language, the gist coming through, the details quietly falling away.
What falls away first is usually the chief complaint. Describing symptoms in a second language, patients reach for the words they have, not the words that fit. A stabbing pain becomes "pain." Pain that wakes them at night becomes "it hurts at night." Knee pain that only shows up on stairs becomes "knee pain." None of these simplifications is wrong, but the one clue the doctor needed is lost somewhere inside them.
So Brief was built on one principle from the start: what the patient actually says is clinical data. It has value in its own right, before anyone translates or summarises it.
How Brief handles language
Patients choose their own language
The patient side of Brief currently supports eight languages: Traditional Chinese, English, Japanese, Korean, Vietnamese, Indonesian, Spanish and Thai. Brief opens in the phone's language by default, and patients can switch by tapping a flag on screen.
Questions come from a human-reviewed multilingual question bank
No question a patient sees is machine-translated on the fly.
When doctors build their specialty's Brief template in the Brief Schema Composer, they can pick questions from each specialty's question bank. Those questions already exist in every supported language, and the translations were reviewed by people before going live. A patient who chooses Spanish sees that reviewed Spanish.
We deliberately avoid machine-translating clinical questions at run time. If a question meant to ask about "pain" comes out as "numbness", or "within a week" as "within a month", the patient's answer is wrong too, and the doctor has no way to tell.
The patient's own words, one tap away for the doctor
Whatever the patient types is stored exactly as written, and that original is what the doctor sees by default.
When the doctor wants help, tapping the translate/edit chief-complaint button in the inbox opens the Brief already translated into Chinese. Inside the Brief they can switch to English, or back to the original at any time for comparison. Translations are stored separately and never overwrite what the patient wrote.
Translation runs on the doctor's own Google Gemini key, which must belong to a billing-enabled, pay-as-you-go project rather than the free tier; doctors re-confirm this every 90 days, and translation stops if it lapses. The name, date-of-birth and visit-number fields are never sent for translation. If it fails, the screen simply shows the original and nothing gets stuck. If only part of it succeeds, any question that wasn't translated is marked "untranslated" and keeps the original text, so a partial translation is never passed off as a complete one.
Patients see this too. Just below the progress bar on the Brief form, a short line tells them they can answer in whichever language they're most comfortable with, that the doctor may use Google's AI translation tool when reading, so what they write (except their name and date of birth) may be processed outside Taiwan, and that what they wrote is kept in full. The line is available in eight languages and follows the interface language the patient has chosen.
What we deliberately don't do
- No automatic translation. Translation always starts with the doctor. Every translated line a doctor reads is one they chose to translate.
- No run-time machine translation of clinical questions. Every language version is prepared and reviewed before it goes live.
- No sending the name or date-of-birth fields for translation. Translation only needs to understand the complaint and symptoms, not who the patient is. If a patient writes their name or phone number inside another free-text answer, though, that text is still sent along with the rest.
- No translation replacing the original. Original and translation are stored separately, so the doctor can always go back to what the patient actually wrote.
What these have in common: translation is a tool that helps the doctor understand; it is never the record of what the patient said. That record is always the original.
Current limitations
- Questions a doctor writes from scratch are not translated automatically. If a template includes questions from outside the question bank, a Spanish-speaking patient sees the question text in whatever language the doctor typed it in (usually Chinese), and the answer options in English.
- Translation requires the doctor to set up their own AI key. Doctors who haven't done so see the patient's original text.
Closing
A patient on a business trip, one evening's clinic: a small moment. But it shows where Brief stands on language. Patients say it in the language they know best, doctors read it in theirs, and the translation step in between is the doctor's choice and never alters the patient's own words.
Related: First-Visit Brief: When the Patient Can't Name the Tissue · Why We Open-Sourced Our Questionnaire Format