Choosing languages from evidence
The usual approach is to add the languages that feel obvious for the region. It is also how venues end up maintaining a German menu nobody opens while their actual second-largest guest group reads none of the four options.
You almost certainly already hold the evidence:
- Reservation platform data, which usually records the country the booking came from.
- Card acquirer reports, which show issuing country by transaction. This is the least biased signal most venues have and the least used.
- Hotel concierge relationships, if you take referrals -- ask which nationalities they send you and when.
- Seasonality. A coastal venue may need different languages in July than in February, and a menu that switches languages by season is entirely reasonable.
Once a multilingual menu is live it produces the cleanest signal of all: which language guests actually switch to. That is the number to reassess against after a season, and it frequently contradicts the assumption that started the exercise.
What to translate, and what to leave alone
A menu item is three different kinds of text and each wants different treatment.
The dish name: usually leave it
Named dishes are proper nouns as much as descriptions. Translating them literally destroys the thing the guest is trying to order and often produces something unappetising: "baked yoghurt lamb" is not what anyone wants to see where "Tavë kosi" was. Worse, it breaks the moment at the table where a guest points at a word and a server recognises it.
Keep the original name and put the explanation underneath. This is also how the guest learns the word, which is a small part of why they travelled.
The description: translate it properly
This is where translation earns its money. Ingredients, cooking method, heat level, portion, what it comes with. A guest deciding between two dishes is reading this text, and it is the text most likely to be machine-translated badly and never checked.
Allergen and dietary words: translate them, then verify them
A mistranslated allergen term is not an awkward menu, it is a safety issue. "Nuts" and "peanuts" are the same word in some languages and different in others; "gluten-free" is a regulated claim in some markets. Have a native speaker check every allergen and dietary term in every language you publish, once, and keep the list stable afterwards.
Show both, do not swap
The tempting design is one language at a time: the guest picks a language and the menu becomes that language. It reads cleanly and it removes the one thing the guest needs at the moment of ordering.
A guest who has read the English description still has to say the dish name to a server who may not speak English. If the original has been swapped out, they are pointing at a phone screen and hoping. Keeping the original name visible alongside the translated description solves this with no extra work.
The same argument applies to prices and to anything the kitchen or the till needs to match on.
Arabic, and what right-to-left actually requires
Adding Arabic is often filed as "add a language". It is a layout change.
- The whole reading direction reverses, so the visual order of a dish name and its price flips with it.
- Alignment, padding, and any icon that implies direction -- a chevron, an arrow -- have to mirror.
- Numerals may be Western or Eastern Arabic depending on the market, and mixing them inconsistently within one menu looks careless.
- Line lengths change substantially, so a layout that was tuned for a short Latin string may wrap awkwardly.
None of this is difficult, but it is not free, and a menu that renders Arabic text inside a left-to-right layout is immediately recognisable as an afterthought to the people it was added for.
myQRMenu ships English, Turkish, German, French, Spanish, Italian, Arabic with right-to-left support, Russian, and Greek as menu content languages. That set covers the Western European and Mediterranean corridors and the Gulf, which is a deliberate scope rather than a complete one — a venue whose second-largest guest group reads none of the nine is better served by asking than by guessing.
Where machine translation stops working
Modern machine translation is genuinely good at menu descriptions and genuinely unreliable at menus as a whole. The failure modes are consistent enough to list.
- Regional ingredients that have no target-language equivalent get replaced with the nearest common thing, which is often a different ingredient.
- Cuts of meat are named differently in every butchery tradition, and a literal translation frequently names a cut that does not exist in the target market.
- Cooking methods collapse. Braised, stewed, and slow-cooked come out as one word.
- Brand and regional protected names get translated when they should not be. A protected designation is a legal name, not a description.
- Anything ironic, dialectal, or jokey in the original arrives as nonsense.
The rule that follows: machine translation drafts, a person approves. Restaurant menu translation covers the review workflow in detail.
Switching languages at the table
Two small decisions make more difference than the translation quality does.
First, the language switch has to be visible without scrolling. A guest who cannot read the menu also cannot read a label telling them where the language switch is, so it needs to be recognisable as a control by shape and position, and it needs to be near the top.
Second, switching should not reload the page and lose the guest's place. Someone halfway down a long menu who switches to Greek and lands back at the top has been punished for not reading English.
A guest-facing language switch should only appear when the restaurant has actually published more than one language. A switcher offering a language that turns out to be half-translated is worse than no switcher at all.
Knowing whether it worked
The honest measurement is narrow, and worth stating alongside what it cannot show.
You can see how many guests switched language, which languages they chose, and how that mix moves through the season. That answers "was this language worth publishing" and "which language should be next" -- the two questions that actually drive the decision.
You cannot see whether the translated description changed what they ordered, because the menu does not know what was ordered. Claims that a menu platform can attribute revenue to a translation are, absent a POS integration, describing something the data cannot support. What QR menu analytics can and cannot tell you draws the line more fully.
One further consequence worth knowing: a multilingual menu published as real web pages is also readable by search engines, which matters for a venue that tourists find before they arrive rather than after they sit down. Restaurant menu SEO covers that side.