Two languages, one record, zero tolerance for drift. Where a live transcript helps a certified court interpreter — and what no tool will ever carry.
A courtroom interpreter is the only person in the room working in two languages at once, in real time, under oath — on a record that can be appealed. A dropped digit in a case number is not a typo; it is an inconsistency opposing counsel can use. Court interpreting is the setting where "close enough" is a professional failure, which is exactly why a live transcript beside a certified interpreter earns its place. What follows is practical orientation for working interpreters, not legal advice.
When testimony passes through an interpreter, the interpretation *is* the testimony. The witness's actual words, in their actual language, usually never reach the record — only the interpreter's rendition does. That is why court interpreting has certification exams, codes of ethics, and an oath of its own: the rendition carries legal weight, and interpretation quality has been raised on appeal often enough that every certified interpreter knows the stakes.
The standard is verbatim in spirit as well as letter. Hedges, false starts, and self-corrections all stay in, because "I think he was there" and "he was there" are different pieces of evidence. Three things make this unlike any other interpreting work:
Court interpreters switch between modes all day, and each one fails differently:
Rendering the judge, counsel, and witnesses in real time so a defendant can follow their own trial. There is no pause button; anything missed is gone unless the interpreter stops the court to ask — which interpreters are trained to do, and which everyone prefers to keep rare.
Question and answer through the interpreter, with attorneys listening for daylight between what was said and what was rendered. Memory and note-taking carry the load — and numbers are the first thing working memory drops.
A document the interpreter may never have seen — a plea form, a lab report — rendered aloud on the spot. Legal register, dense syntax, no preparation. It is the mode that most resembles a stress test.
The modes demand different support, which is worth understanding before adding any tool to the workflow — the longer treatment is in our guide to simultaneous vs. consecutive interpreting.
Ask working court interpreters what actually goes wrong and the list is remarkably consistent. It is almost never grammar or general vocabulary. It is the high-entropy details:
Notice what these have in common: they are things text is good at and short-term memory is bad at. That is the opening for a transcript.
The right mental model is a verification layer, not a second interpreter. The interpreter renders by ear, as always; the transcript sits in peripheral vision holding the details still. Unicaption was built for exactly this copilot role:
Preloading terminology before a hearing is its own small craft — build the habit with our custom dictionary guide.
A transcript hears words. A court interpreter manages meaning under adversarial conditions, and that division of labor is not going to move:
| The interpreter owns | The transcript can carry |
|---|---|
| Register and tone — a witness's hedging, sarcasm, or distress, rendered faithfully | The literal words that were said, as reference |
| Legal equivalence — choosing renditions between legal systems that don't map one-to-one | Spellings and terms of art, held still for verification |
| Procedure — objections mid-sentence, sidebar boundaries, when to stop the court and correct the record | Digit strings and dates that memory drops first |
| The oath — personal, professional accountability for the rendition | A second look, before the rendition is spoken |
This is the same conclusion we reach in the broader argument about whether AI will replace interpreters: the tool is leverage for the professional, not a substitute for one — and nowhere is that clearer than in a courtroom.
Courts control what devices operate in their rooms, and policies vary by jurisdiction, courthouse, and judge. Captioning that stores nothing is not recording — but the court's rules govern, so ask the clerk or the judge before your first use, the same way you would about any other working aid.
No. Courts require qualified — in many places certified — human interpreters because renditions carry legal weight, demand judgment about register and legal equivalence, and are given under oath. AI transcription is useful in court settings as a verification layer beside the interpreter: Unicaption holds numbers, names, and terminology on screen so the human can confirm details without breaking flow. The rendition itself stays human.
It depends on the court. Judges control devices in their courtrooms, and policies differ by jurisdiction and even by courtroom. Ask the clerk or judge before using any tool. It helps that Unicaption stores nothing — audio is never stored and transcripts are deleted when the session ends — but a favorable privacy design does not override a court's device rules. When in doubt, ask first.
As peripheral verification, not as a script. The interpreter renders by ear as trained, and glances at the live transcript to confirm high-risk details — case numbers, dollar amounts, dates, and the spelling of names. Before the hearing, they load party names and expected terminology into Unicaption's custom dictionary so the transcript renders them consistently from the first mention.
Hold any tool to this standard: end-to-end encrypted sessions, audio never stored, transcripts deleted when the session ends, and session content never used to train models. Unicaption meets all four, and is end-to-end encrypted with nothing retained after the session. Nothing about the proceeding persists after the session — which is the property that makes a working aid compatible with privileged and sensitive matters.
Live captions with legal terminology built in — nothing stored, nothing retained. 30 free minutes every week, no credit card.
Start Free Trial →