← Blog
5 min read

How to Keep Names, Numbers, and Technical Terms Accurate in Live Translation

The words most likely to matter in a business conversation — a budget figure, a client's name, a product term — are also the words general-purpose translation is most likely to get subtly wrong, which is why they deserve deliberate handling rather than being left to a model's best guess.

Why these three categories specifically

Most words in a sentence have some tolerance for imprecision — a slightly loose synonym rarely changes what a listener understands. Names, numbers, and technical terms don't have that tolerance: a name rendered inconsistently reads as two different people, a number off by a factor of ten is a different budget entirely, and a mistranslated technical term can flip the meaning of an entire sentence for someone relying on the term to understand the topic.

What actually helps

A glossary — a list of terms and their intended translation, supplied ahead of a session — is the single highest-leverage tool here, because it removes ambiguity the model would otherwise have to guess at. A company name, a product name, an acronym specific to an industry: once it's in a glossary, a well-built pipeline should render it the same way every time rather than translating it fresh (and inconsistently) each time it's said.

For numbers specifically, explicit unit and magnitude conversion between systems (like Chinese's 万-based grouping vs. English's comma-grouped thousands, discussed in more detail in our piece on Chinese-English interpretation) matters more than general translation quality — this is a narrow, well-defined problem that's worth solving directly rather than trusting a general-purpose model to get right by default.

For names, the most reliable systems track what's already been said earlier in the same session and stay consistent with it, rather than re-deciding the spelling or rendering of a name every time it comes up.

Setting expectations

No system — human or software — is error-free on this in real time. The realistic goal is a very high hit rate on names, numbers, and terms specifically, since those are the words a listener has no way to sanity-check themselves in the moment, unlike a merely awkward sentence they can usually parse around.

Be first to try real-time interpretation yourself.

In active testing — join the waitlist and we'll email you when it's ready.