Start Free Trial
← All posts
Names and numbers first

Mandarin-English meetings: where the numbers go wrong

Sentences mostly survive translation. Names and numbers are what fail — and the reasons are built into the language. A guide to the traps, and to captions as the verification layer in a deal.

万→亿 the 10,000 gap 6 min read Published July 2026

Watch enough Mandarin-English meetings and a pattern emerges: the arguments translate fine, and the mistakes cluster in two places — what things are called, and how much of them there are. That is not an accident of attention. The properties that make names and numbers hard in Mandarin-English work are structural — tones, homophone density, and a number system built on 10,000 instead of 1,000 — and they trip humans and machines alike. Which is exactly why a written, checkable record earns its place at the table.

Why names are the hard part: tones and homophones

Mandarin builds all its words from a small inventory of syllables — a few hundred, ignoring tone; still comparatively few with tone included. English has vastly more. The result is extreme homophony: the syllable *shì* alone corresponds to dozens of common characters. Everyday speech survives this because context disambiguates. Names are precisely the words context cannot rescue.

  • Personal names carry no redundancy. Whether the new counterpart is 李 Lǐ or 黎 Lí, 张 Zhāng or 章 Zhāng, nothing else in the sentence tells you — you either caught the tone and the character, or you guessed.
  • Company and product names are coined, often deliberately homophonous with auspicious words, and frequently have separate English brand names that share nothing with the Chinese one.
  • English names get re-cut into Mandarin phonology in the other direction, so "Schwartz" and "Swartz" may surface identically — a problem when both are on the cap table.
  • ASR feels the same pain. Speech recognition resolves ambiguity by context, and names have none. This is the single strongest argument for loading every attendee, entity, and product into a custom dictionary before the meeting.

万 and 亿: where million/billion errors come from

English groups large numbers in threes: thousand, million, billion. Chinese groups them in fours: 万 (wàn, 10,000) and 亿 (yì, 100,000,000). Every large figure in a bilingual meeting is therefore an arithmetic conversion performed live, in someone's head, mid-sentence.

ChineseLiterallyEnglish
一万 (yī wàn)one wan10,000
十万 (shí wàn)ten wan100,000
一百万 (yìbǎi wàn)a hundred wan1 million
一千万 (yìqiān wàn)a thousand wan10 million
一亿 (yí yì)one yi100 million
十亿 (shí yì)ten yi1 billion

So "3.5亿" is 350 million, "两千万" is 20 million, and one slipped step is a tenfold error in a term sheet. Even fully bilingual professionals do this conversion consciously rather than automatically — ask one. Interpreters drill it; everyone still double-checks the big figures, because the cost of being wrong is not embarrassment, it is the deal.

House rule for negotiations: every material number gets restated in digits — spoken ("three-five-zero million") or written in the transcript — before anyone moves on. Digits are the one notation both number systems share.

What "we'll study it" means

Chinese business register prizes indirection where American English prizes explicitness. A flat "no" is rare in formal settings; refusal arrives dressed as deferral. 我们研究研究 — "we'll look into it" — is often a soft no. 不太方便, "not very convenient," usually means "not possible." Praise is deflected, disagreement with a senior person is routed through questions, and silence can be substantive rather than empty.

Be precise about what captions do here: a transcript shows you exactly what was said, which is genuinely useful — you can revisit the phrasing later and notice it was 研究研究 and not a commitment. What it cannot do is tell you what the phrasing *meant* in the room. That reading is human work — an experienced interpreter or bicultural colleague — and it is a large part of why AI is not replacing interpreters in negotiation settings.

Simplified or traditional captions?

Spoken Mandarin is one language with two standard scripts. Mainland China and Singapore read simplified characters; Taiwan and Hong Kong read traditional, as do many overseas communities. The audio is identical — the captions are not, and displaying the script your audience does not read is a small constant tax on every reader. Match the script to the room, and when the room is mixed, English translation alongside the Chinese often serves as the neutral channel.

One adjacent caution: Cantonese is not Mandarin with a different script — it is a different spoken language. Mandarin captioning does not cover a Cantonese speaker; that is a separate language selection, not a settings toggle.

Code-switching is the norm, not the exception

Professionals in tech, finance, and trade rarely speak a pure stream of one language. English product terms, metric names, and titles ride inside Mandarin sentences — "这个 quarter 的 KPI" is unremarkable speech in a Shanghai office. Human listeners barely notice. Transcription systems historically stumbled exactly here, tagging the audio as one language and mangling the embedded other. Automatic language detection has made this dramatically better, but it remains a real differentiator between tools — test with your own team's actual speech, not a demo reel.

Frequently asked

Why do numbers get mistranslated between Chinese and English?

Because the two languages group large numbers differently. Chinese counts in units of 10,000 (万 wàn) and 100 million (亿 yì); English counts in thousands, millions, and billions. Converting between the systems is live arithmetic — 3.5亿 is 350 million, 十亿 is one billion — and a single slipped step is a tenfold error. The reliable fix is restating every material figure in digits, which both systems share, ideally in a live transcript both sides can see.

Should meeting captions be in simplified or traditional Chinese?

Match your audience's reading habit: simplified for mainland China and Singapore, traditional for Taiwan and Hong Kong and many overseas communities. The spoken Mandarin is identical either way — the choice is purely about what the readers process fluently. For mixed rooms, showing English translation alongside the Chinese is often the practical neutral channel.

Can AI captions handle a speaker who mixes Chinese and English?

Increasingly well, but it varies by tool. Code-switching — English terms embedded mid-Mandarin-sentence — is normal professional speech and historically a weak point for transcription systems locked to one language. Unicaption handles it with automatic language detection across 60+ languages, and a custom dictionary for the product names and acronyms that ride between languages. Test any tool on your own team's real speech before a meeting that matters.

Do live captions replace an interpreter in Chinese business negotiations?

No — they change the interpreter's job rather than eliminate it. Captions pin down the verifiable layer: exact figures, names, and phrasing, visible to both parties in real time. Reading intent through politeness and indirection — knowing that "we'll study it" is often a soft no — remains human work. The strongest setup in a serious negotiation is an experienced interpreter with a live transcript beside them, each covering what the other cannot.

Put the numbers where both sides can see them

Live Mandarin-English captions and translation beside any meeting platform. 30 free minutes every week — no credit card.

Start Free Trial →