Sector-side day-two coverage is in "Canton Fair Phase 3 Day Two". This article describes the second live order our Shenzhen translation company handled today — and how MTPE actually performs on Southeast Asian small languages.
The order: Dongguan food exporter needs Shopee/Lazada five-language localisation
The client is Weiziyuan, a Dongguan Fenggang food exporter of nuts, dried fruit and meat jerky, exhibiting in Hall 8's food zone. Their main market has been the US and EU via Amazon — three years in, growth is flattening. Their explicit goal for Phase 3: ride the APEC Shenzhen tailwind into Southeast Asia.
Five Southeast Asian buyers visited their booth this morning — two Vietnamese, two Indonesian, one Thai. One Vietnamese buyer, a category manager from Shopee Vietnam, was interested in 12 nut products and confirmed intentions on the spot, with one condition: Shopee-Vietnam-ready product pages (title, five-point description, detail page, keyword list, Instagram and Facebook promotional posts) within two weeks. The other four buyers covered Shopee Indonesia, Shopee Thailand, Lazada Malaysia and Lazada Singapore with similar requirements.
The client had a draft English translation ready, which the Vietnamese buyer rejected outright — "Shopee-Vietnam buyers convert three times better on Vietnamese-language pages than English". Several agencies quoted high prices (five languages × 12 products × 8 fields = at least 480 translation units) and said two weeks was too tight.
Our process: MTPE batch draft + language-matched translators + platform-rule checks
Three layers. First: MTPE batch draft. We ran the Chinese product manual and English reference through MTPE engines tuned for Southeast Asia, producing all five language drafts simultaneously. Here's where the gap showed up: Vietnamese and Indonesian engines were acceptable at around 45% revision ratio; Thai and Malay engines were unstable at over 60%; Singapore English was best (being English) at under 25%.
Second: language-matched translators. Vietnamese draft went to our senior Vietnamese translator — she has delivered Shopee-Vietnam food localisation before, knows the platform rules, and controlled revision to under 40%. Indonesian went to our Indonesian translator who covers both Shopee and Tokopedia. Thai and Malay drafts were poor enough that we had translators start from scratch — faster than fixing bad MTPE output. Singapore English went to our English translator, focusing on keyword ordering and local phrasing.
Third: platform-rule verification. After each language translator delivered, our cross-border e-commerce operations specialist checked character limits (Shopee Vietnam = 60 chars max title, Shopee Indonesia = 80 chars, Lazada Malaysia requires at least three core keywords), keyword search heat and description-format conventions — all things MTPE cannot handle.
The process is on track. The first batch (three products across all five languages) was delivered this afternoon and the client is satisfied. MTPE's real value here was not cost savings — it was time savings. Without MTPE batch drafting, five-language translation would take at least three weeks; now we are on pace for two.
MTPE reality across small languages
This order confirms something we have observed before: MTPE quality correlates directly with training-data volume. English, Japanese and Korean have massive public datasets and stable engines; Spanish and Arabic have sufficient data; Southeast Asian small languages have far less — especially food-category e-commerce data — so engine quality lags. For these languages, MTPE is good at "setting up the basic translation skeleton" but human translators still determine final quality.
Compare with yesterday's Spanish Zalando order: the Spanish MTPE revision ratio was only 20–35%, focused on keyword ordering and local phrasing. This Southeast Asian small-language order runs 45–70%, with many sections requiring complete rewrites. Pricing differs accordingly — Spanish was MTPE post-editing rates, Southeast Asian was human-translation rates. But MTPE still cut total hours by at least 40%.
For cross-border e-commerce localisation in Southeast Asian small languages or other platforms, contact us with your language pair, platform and target market.