You teach a translation engine your church's vocabulary by feeding it context before the service: upload your sermon notes or type in a short list of proper names, place names, and ministry terms. This works because a glossary is a custom dictionary the engine uses to translate domain-specific terms and named entities consistently (Google Cloud Translation documentation). The result is fewer mangled names and cleaner translations in every listener's language.
Before your next multilingual service, here's the short checklist we'll walk through below:
- Gather the proper names in the message — the visiting pastor, people being prayed for, missionaries by name.
- List the place names your congregation knows — towns, neighborhoods, mission fields.
- Add your ministry-specific terms and any recurring phrases the sermon leans on.
- Paste your sermon notes or an outline so the engine has running context.
- Include unusual spellings or borrowed words you don't want translated at all.
Why do generic engines mangle church names and terms?
Here's the honest answer, and it isn't the engine being lazy. Proper names and specialized terms are low-frequency and domain-specific, so they're often simply absent from the bilingual dictionaries these systems lean on — which makes their translations inconsistent or flat-out wrong (Challenges in Context-Aware Neural Machine Translation, arXiv).
It goes deeper than a missing word, too. The attention mechanism at the heart of every neural machine translation system doesn't actually recognize terms or tags — it just processes words as numbers and generates output, and large language models inherit the same limitation even when you instruct them (AppTek, Dec. 18, 2024). So "Braga" becomes a common noun, a visiting pastor's surname becomes a verb, and everyone in the pews wonders what just happened.
How does uploading sermon notes fix this?
When you give the engine your notes ahead of time, you're doing the one thing the architecture can't do on its own: supplying context and a fixed reference for the rare words.
Context steadies the whole message
Paste the outline, and proper names and ministry terms improve both the transcription and the translation. The engine now knows the word coming out of the speaker's mouth is a name, not a mispronounced verb — and it treats it accordingly.
A glossary locks the tricky words down
A glossary is a custom dictionary the system uses to translate domain-specific terms and named entities correctly and consistently, and it can even prevent certain words — borrowed words, product names, a ministry's own coined terms — from being translated at all (Google Cloud Translation documentation). That's exactly what you want for a name that should read the same in all 83 languages.
What should go in your custom vocabulary list?
Keep it practical. You're not building a dictionary; you're flagging the handful of words a generic engine reliably trips over.
- Proper names: the preacher, the worship leader, anyone named in prayer or testimony, missionaries and their sending organizations.
- Place names: towns, neighborhoods, and mission fields your congregation recognizes — the "Ribeira" and "Braga" of your particular world.
- Ministry terms: words your church uses in a specific sense that a general engine would render generically.
- Borrowed and coined words: anything you'd rather see left untranslated.
There's plenty of room. On documented glossary platforms, a single term can run up to 1,024 UTF-8 bytes, and the whole file can hold up to 10.4 million bytes of terms combined (Google Cloud Translation Advanced documentation). Your Sunday list will not come close.
Why does this matter for a multilingual congregation?
Because the need is real and it's sitting in your pews. About 68 million U.S. residents speak a language other than English at home, and 29.6 million of them have limited English proficiency (U.S. Census Bureau/ACS, Oct. 2023). Among people who speak a non-English language at home, Spanish leads at 61.1%, with Chinese and Tagalog/Filipino next (U.S. Census Bureau, Dec. 7, 2023).
When a name gets mangled mid-sermon, an LEP listener doesn't just miss a word — they lose the thread of the whole point. Getting the vocabulary right is how the message lands intact, in the language of the heart.
Isn't this the same problem Bible translators face?
It's a cousin of it, and worth a moment of humility. In a study on Buddhist scripture translation, researchers found the model needs a substantial corpus, but the parallel text in that domain was so limited that training couldn't teach the decoder to translate proper nouns correctly (arXiv, 2022). Scripture and ministry terminology are hard for machines precisely because they're rare and specialized.
The stakes globally are enormous: there are still 1,663 languages with no Scripture and no work in progress (Wycliffe Bible Translators, ProgressBible Snapshot, August 2026). Real-time translation won't replace faithful Bible translation — but for a Sunday service, teaching the engine your words is a small, achievable act of hospitality.
Built in the pews, not the lab
That's the philosophy behind Jesus Speaks. Every feature came from a real service, not a whiteboard — including the choice that when playback falls behind, the system speeds up rather than skipping ahead, so the congregation hears every word of the message, never a summary. The vocabulary tool is cut from the same cloth: a fix for a problem we watched happen live, not a spec on a datasheet.
If you lead a Christian ministry and want to serve every language in your pews with the names and terms your church actually uses, you can subscribe to get set up. Bring your sermon notes — we'll help the engine learn them.