← Blog
6 min read

Chinese–English Interpretation: Why Simultaneous Translation Is Uniquely Hard

Chinese and English put information in different places in a sentence — Chinese often saves the key verb or qualifier for later than English would — which means a simultaneous system sometimes has to wait for more of a Chinese sentence before it can commit to an English translation, or vice versa.

Word order is the biggest issue

English sentences tend to front-load structure: subject, verb, object, with modifying clauses trailing off the end. Mandarin frequently does the reverse for time, place, and condition — "next Wednesday, at nine, in the main room, we'll meet" is a perfectly natural Chinese order that an English translation has to substantially reorder, not just word-for-word substitute. For text translation this is a solved problem; for simultaneous translation it means the system sometimes can't commit to the start of an English sentence until it has heard more of the Chinese one than you might expect.

Topic-comment structure compounds this. Chinese frequently states a topic and then comments on it in a way that doesn't map onto English subject-verb-object directly, so a naive word-by-word approach produces grammatically broken English even when every individual word is translated correctly.

Numbers, measure words, and names

Chinese numbers group by 10,000 (万) rather than by 1,000 the way English does, so a figure like 850,000 is naturally spoken as 85万 — a direct, careless translation risks an order-of-magnitude error if the conversion isn't handled explicitly rather than left to a general-purpose model's judgment. Measure words (个, 位, 家, and dozens more) don't exist in English at all and carry no independent meaning to translate — they need to be dropped, not translated literally.

Names are their own challenge in both directions: Chinese names transliterated into English have no single canonical spelling unless the person has already established one, and English names spoken in a Chinese-language conversation are often approximated phonetically in ways a speech recognizer can mishear entirely.

What this means practically

The upshot is that a system built for Chinese-English simultaneous interpretation benefits from language-pair-specific handling — not just a general translation model pointed at Mandarin — particularly around numbers, units, and a short mechanism for looking up how a name or technical term has already been rendered earlier in the same conversation, so it stays consistent rather than shifting mid-meeting.

Be first to try real-time interpretation yourself.

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