The Translation Playbook
Anyone can paste text into a translator. Almost no one builds the workflow around it. These are the walkthroughs we use for the work where a single wrong word carries real consequences — legal, medical, and education.
A translation is a decision, not a button.
Free engines like DeepL and quick transcription tools like Rev are excellent at one narrow job: turning words in one language into plausible words in another. That is roughly 30% of high-stakes language work. The other 70% is everything around it — classifying the document, locking terminology, preserving structure, certifying accuracy, protecting confidentiality, and getting a qualified human to sign off.
This playbook documents that 70%. Each play is a repeatable sequence: the same steps, in the same order, every time — so quality stops being a matter of luck and starts being a matter of process. LANGUU is built to run these plays end to end, combining translation, transcription, text-to-speech, and closed captioning in one auditable workflow.
Identify the document and jurisdiction
Is it a contract, a court filing, a patent claim, or an immigration record? Which jurisdiction governs it? This determines whether you need a standard translation or a certified / sworn one, and which formatting rules apply.
Build the glossary before you translate
Fix defined terms, party names, and terms of art up front — "Indemnify," "Force Majeure," "without prejudice." Decide once, apply everywhere. Inconsistency is what opposing counsel exploits.
LANGUU termbase & translation memoryKeep source and target aligned
Work segment by segment so every translated clause maps back to its original. A reviewer should be able to audit any line in seconds, not re-read the whole document.
Mirror clauses, numbering, and exhibits
Clause 4.2(b) must remain Clause 4.2(b). Schedules, exhibits, and cross-references have to line up exactly — structure carries legal meaning, not just text.
Attach a certificate of accuracy
For filings, contracts, and immigration, append a signed statement of accuracy and arrange notarization where the jurisdiction requires it. Uncertified output is often simply not accepted.
Secure the data, then get human review
Privileged text stays encrypted and is never used for training. A qualified reviewer signs off on high-liability documents before release — the machine drafts, the human decides.
Encrypted, NDA-grade handlingWhat a raw translator won't catch
- No certificate of accuracy and no notarization — rejected at filing.
- No audit trail linking each translated clause to its source.
- Your privileged, confidential text retained or used to train models.
- "False friends" in legalese translated literally, changing the obligation.
Sort content into risk tiers
Patient-facing materials (consent forms, discharge instructions, dosing) carry the highest risk. Clinical and regulatory documents (IFUs, adverse-event reports, study protocols) carry the highest scrutiny. Treat them differently.
Use standardized medical vocabularies
Map terms to recognized standards — ICD, MedDRA, SNOMED — and use approved drug names and units. "Medical English" is not the same as the language a clinician will accept in another market.
Domain glossaries & terminology controlReconcile high-risk content
For consent, dosing, and safety text, translate the target back into the source language and compare. Differences surface ambiguity and errors before a patient ever sees them.
Match health-literacy level
A consent form a patient can't understand isn't informed consent. Translate into plain language at an appropriate reading level — not a literal rendering of dense clinical prose.
Transcribe dictation and telehealth, then translate
Clinical audio — physician dictation, telehealth visits — becomes an accurate transcript first, then a translation. The chain of accuracy starts at the recording, not the document.
LANGUU transcriptionAdd accessibility, protect PHI
Caption patient-education video and offer multilingual text-to-speech for low-literacy populations. Handle protected health information with HIPAA-aligned controls and full audit logs throughout.
Captioning · TTS · audit loggingWhat a raw translator won't catch
- Dosage, unit, and decimal errors with no back-translation to catch them.
- No mapping to MedDRA / ICD / SNOMED — terms drift between markets.
- Protected health information leaking into a public model.
- Consent text translated literally and unreadable to the patient.
Separate records from learning material
Transcripts, diplomas, and enrollment forms need certified, exact translation. Course material, lectures, and family comms need localization. Mixing the two standards is where quality breaks.
Translate transcripts with credential notes
Academic records translate word-for-word, with equivalence notes explaining grading scales and credit systems so a receiving institution can read them correctly.
Transcribe and caption every lecture
Accurate closed captions are the foundation of accessible learning and a legal requirement. Get the captions right in the source language before translating them into others.
LANGUU transcription & closed captioningAdapt examples, idioms, and units
A worked example that assumes local currency, names, or customs confuses learners elsewhere. Localization adapts the lesson — not only the words — so the concept lands.
Add multilingual text-to-speech
Read-aloud in the target language supports emerging readers, language learners, and visually impaired students — and turns written material into accessible audio at scale.
LANGUU text-to-speechKeep terminology consistent across everything
Newsletters, IEPs, and enrollment forms should use the same glossary as the classroom content. One shared termbase keeps every message to families clear and consistent across dozens of languages.
Shared termbase across all documentsWhat a raw translator won't catch
- Auto-captions full of errors — failing accessibility and the students who rely on them.
- Terminology that drifts between the transcript, the lecture, and the newsletter.
- Literal translation of culturally specific examples that confuse learners.
- No certification on academic records — rejected by receiving institutions.
A tool gives you words. A system gives you outcomes.
DeepL, Rev, and free translators are good at the middle step. The plays above need everything around it — and a single workflow instead of five disconnected apps.
Stop translating. Start running the workflow.
Bring your legal, medical, or education content into one system built to handle the 70% that free tools leave to chance.