Mac Live Captions Miss Names and Technical Terms: What to Try

Find out which part of the sentence is wrong

Keep the original caption visible while diagnosing a translation problem. Compare it with the audio and, when available, the agenda, slide, or written name.

• What you see: Much of the sentence is wrong; What to check first: Spoken language, audio quality, overlapping voices, and recognition setup
• What you see: Most words are right but one name is repeatedly wrong; What to check first: Expected vocabulary and a narrowly scoped correction
• What you see: The source term is right but the translation is wrong; What to check first: Translation context and the correct meaning of the term
• What you see: A caption contains speech that was never said; What to check first: The audio source, background sound, and recognition errors; do not merely replace a word

A correct name inside an incorrect sentence can make the caption look reliable. Read the whole statement, especially the verb, number, and condition around the name.

If the meeting platform already provides an approved glossary, corrected transcript, or useful vocabulary feature, start with it. A shared spelling in the agenda also helps everyone confirm the same term, regardless of which caption tool they use.

Fix the listening conditions before adding vocabulary

Choose the language actually being spoken. The app's interface language and the language of a translated caption do not establish the recognition language.

Then check the source. For a remote presentation, use the Mac's playback audio rather than a microphone pointed toward the speakers. For an in-person speaker, microphone distance and room noise matter. When using both playback and microphone captions, wear headphones to reduce the chance that the microphone picks up the playback again.

Listen to one clear sentence. If it is difficult for you to hear, adding a list of technical words is unlikely to solve the underlying problem. If everyday speech is already readable and the same unusual term keeps failing, vocabulary settings become a more useful next step.

Use hotwords to suggest terms before recognition

A hotword is a term you expect to hear. In Alison Workspace, open Hotwords in settings, or use Manage Hotwords… where that control appears. Add the correct spelling of a small number of relevant names and terms.

Start with the vocabulary for the actual meeting: the people attending, the product under discussion, and the few abbreviations that would be hard to infer. A long list of loosely related words gives you more to maintain and makes it harder to see which addition helped.

Hotwords influence recognition through the features available in the selected engine. They are not commands to insert a word, a pronunciation-training session, or a guarantee of correct spelling. Support differs between recognition paths; a setting that helps live recognition should not be assumed to work identically for file transcription or every language mode.

After editing the list, start a fresh short caption session and repeat the same clear phrase. Check the result in context. Do not count a correct name as a success if the rest of the sentence became less accurate.

Use a dictionary rule for a specific recurring text error

A dictionary rule says, in effect, “when this text appears, replace it with that text.” It acts on recognized words rather than teaching the model a new voice or pronunciation.

Here is an invented example, not a recognition test: a fictional product is named NeraFlow, but a caption repeatedly writes Nera Flow. If the recording and meeting materials confirm the intended product, a specific Nera Flow → NeraFlow rule may be useful.

Apply a narrow rule and check for side effects

In Alison Workspace:

1. Open Custom Dictionary, or Manage Dictionary… where available.
2. Put the recurring recognized text in Original.
3. Put the verified spelling in Replace With.
4. Add the rule, then check a new caption containing the phrase.
5. Check an ordinary sentence where a similar phrase could occur legitimately.

Avoid rules such as replacing every occurrence of a common short word with a person's name. A broad replacement can change unrelated words or make a normal phrase incorrect. Matching can also depend on the actual capitalization and punctuation produced by recognition; test the observed error rather than assuming the rule understands all variants.

Keep rules that solve a clear recurring problem. Remove one when it starts changing unrelated speech.

Keep spelling corrections separate from translation terminology

A source-language dictionary is not a bilingual glossary. Correcting a product's spelling does not tell the translator how every specialist concept should be expressed in another language.

For example, a research term may be transcribed correctly but translated with its everyday meaning. Replacing it with a different source word just to force a preferred translation can corrupt the transcript. Keep the source accurate, consult the field's terminology, and check the translated sentence separately.

When discussing a name, identifier, or abbreviation, ask for the written form in the meeting chat if needed. When a decision depends on a technical phrase, confirm the meaning with the speaker. Captions are useful for following the discussion; the actual spelling and intended meaning remain separate things to establish.

Check changes with a repeatable sample

Use a short recording you are allowed to process, or a brief agreed test before a call. Keep the audio source, spoken language, and recognition setup the same while comparing settings.

First listen without the new vocabulary entry. Note the exact error. Add only the relevant hotword or rule, then check the same phrase again in a fresh session. Finally, check a different sentence containing similar sounds.

This is a practical check of your setup, not an accuracy benchmark. A successful result on one sentence does not establish that every future speaker, accent, or technical talk will work equally well.

If a rule improves the name but damages ordinary text, narrow it or remove it. If the name never becomes recognizable, return to the audio, ask for its written form, or use a reviewed transcript instead of adding increasingly aggressive replacements.

Build a small meeting vocabulary you can maintain

Before a technical call, gather names and terms from the invitation, agenda, slides, or project documents. Add only what you can verify. For an unfamiliar acronym, keep its expansion in your own meeting notes so you understand what it means as well as how it is spelled.

After the call, review recurring errors rather than turning every one-off mistake into a permanent rule. Vocabulary changes should make future captions easier to trust, not hide uncertainty behind a growing replacement list.

Alison Workspace requires macOS 15 or later and an Apple-silicon Mac. Its supported recognition and translation run locally after the required resources are installed. Follow your organization's rules for meeting tools and for retaining transcript text; a local vocabulary feature does not change those rules.

For your next meeting, try a handful of verified terms and one carefully checked correction. Keep the original caption visible, and use written confirmation for any name, number, or commitment that matters.

The Mac App Store listing describes custom vocabulary features. The distinctions between recognition hints, text replacement, and engine-specific behavior were checked against the project source on October 24, 2026. No numerical accuracy improvement is claimed.