Menu translation: Japanese to English
The dish name is usually not the thing to translate
Japanese menus lean on names that function as proper nouns. Translating 親子丼 to “parent-and-child rice bowl” is literally correct and commercially useless — it tells a diner nothing and reads as a curiosity. The working pattern is to romanise the name, then add a short descriptor on first mention: “Oyakodon — chicken and egg simmered over rice.” The guest learns the word, can order it out loud, and knows what arrives.
The same applies to preparation terms that carry no English equivalent. Nikomi, agemono and yakimono describe technique families rather than single methods; flattening all three to “cooked” loses the distinction a menu exists to draw.
Where the allergen risk actually sits
The hazard in Japanese menus is rarely the headline ingredient — it is the seasoning. Standard shoyu is brewed with wheat, so a dish with no visible bread is still a gluten exposure. Dashi carries fish even in dishes a Western reader would assume are vegetarian. Miso ranges across soy, barley and rice bases depending on the maker.
None of this is visible in a dish name, in either language. It has to be recorded on the item as structured allergen data, not left to the description, because a guest filtering for gluten will never read the paragraph that mentions soy sauce.
Counting, portions and price
Japanese uses counter words rather than plurals, so a machine translation frequently produces “2 piece” or “2 pieces of gyoza” where a native English menu says “Gyoza (6)”. Portion conventions differ too: an English reader expects to know whether a price is per piece or per plate, and Japanese menus often leave that to context.
Prices need the same care. Yen carries no decimals, so a naïve currency conversion produces prices like “€8.43” that look computed rather than chosen. Set a deliberate price per locale.
How Intermenu handles Japanese → English
Intermenu treats every menu entity as a structured record — name, description, ingredients, allergen tags, diet tags — so translation happens per field rather than across a wall of text. Allergen and diet data are tags, not sentences, which is why they keep working when the menu renders in English.
Machine translation produces the first draft; a human reviews before a locale is published, and the review state is recorded per translation.