10 Common Translation Mistakes (And How to Avoid Them)

A mistranslated slogan cost HSBC $10 million. A date format bug trashed $50,000 worth of factory equipment. A Facebook auto-translation got a man arrested. Translation mistakes aren’t just awkward — they break things. Here are 10 specific ones to watch for, with real examples and fixes that take seconds to apply.

1. Word-for-Word Translation of Idioms

Idioms rarely survive direct translation. The words mean one thing in context and something completely different when taken individually — yet this remains the most common translation mistake across every language pair.

❌ “It’s raining cats and dogs” → 「下猫下狗」(literal Chinese: “cats and dogs are falling”)

✅ 「倾盆大雨」(Chinese: “basin-pouring rain”) / “Il pleut des cordes” (French: “it’s raining ropes”) / “Es regnet in Strömen” (German: “it’s raining in streams”)

❌ “Break a leg” → 「رجل اكسر」(Arabic: literally “fracture a leg”)

✅ 「بالتوفيق」(Arabic: “good luck”) / «Ни пуха ни пера!» (Russian: “neither fur nor feather” — equivalent to “break a leg”)

❌ Coors beer “Turn it loose” → Spanish equivalent of “suffer from diarrhea”

✅ A localized slogan tested with native speakers in the target market

❌ “To have other cats to whip” (French: «avoir d’autres chats à fouetter») → literal English nonsense

✅ “To have other fish to fry” — the equivalent idiom in English

How to avoid: When you encounter an idiom, ask: “does this exact phrase exist in the target language with the same meaning?” If the answer is no — and it almost always is — find the local equivalent rather than translating the words.

2. Ignoring Formality and Register

English uses “you” for everyone. Most other languages don’t. Using the wrong form of address can range from mildly awkward to actively offensive — especially in business, legal, and customer-facing contexts.

❌ Japanese: あなた (anata) in customer service. Literally “you,” but in a service context it’s distant and dismissive.

✅ お客様 (okyakusama, “honored customer”) or the customer’s name + 様 (-sama)

❌ French: “Tu” instead of “Vous” in a business email. Using informal with someone you don’t know is presumptuous.

✅ “Vous” for anyone without an established informal relationship. Default to formal.

❌ Italian: “Tu” instead of “Lei” when addressing a client. Italian uses the third-person feminine “Lei” as the formal address.

✅ “Lei” + third-person verb forms for business and formal contexts

❌ Korean: Using 반말 (banmal, informal speech) in a business email instead of 존댓말 (jondaenmal, polite speech).

✅ Default to the highest politeness level (합니다 / 합니까 forms) unless the relationship is clearly casual

How to avoid: For any language you’re translating into, learn two things: the formal/informal split for address, and which register the target audience expects. When in doubt, go formal — it’s safer to be slightly too polite than accidentally rude. In Japanese, Korean, and Thai, formality systems are especially complex and machine translation alone won’t get them right.

3. Date and Number Format Mix-Ups

Only the United States and a handful of territories use MM/DD/YYYY. The rest of the world uses DD/MM/YYYY or YYYY-MM-DD. Getting this wrong doesn’t just look unprofessional — it corrupts data.

❌ 03/06/2026 in a contract for a European audience. Writer meant March 6 (DD/MM). Reader sees June 3 (MM/DD).

✅ “6 March 2026” or “March 6, 2026” — unambiguous across regions

❌ 1,500.75 in a German document. German swaps comma and decimal point: 1.500,75. Brazilian Portuguese does the same: 1.500,75. Swiss French and Italian use apostrophes: 1’500.75.

✅ 1.500,75 (German/Portuguese), 1 500,75 (French), or use locale-aware formatting

❌ Using the Gregorian calendar assumption for all audiences. Saudi Arabia uses the Hijri calendar for official documents. Thailand uses the Buddhist calendar (year +543). Japan uses imperial era years (Reiwa 8 = 2026).

✅ Confirm which calendar system your audience uses for dates

How to avoid: Store dates as ISO 8601 (YYYY-MM-DD) internally. Display them in the format the reader expects. For any translated document, explicitly confirm: “Dates in this document use DD/MM/YYYY format.” The 5 seconds it takes to add that sentence saves days of confusion.

4. False Friends: When Words Lie to You

A false friend is a word that looks identical or nearly identical across two languages but means something entirely different. They’re especially dangerous because they create an illusion of correctness — the translator sees a familiar word and doesn’t stop to check.

English Word	False Friend	Language	Actual Meaning
embarrassed	embarazada	Spanish	pregnant
actually	attualmente	Italian	currently
gift	Gift	German	poison
sensible	sensibile	Italian	sensitive
eventually	eventualmente	Portuguese	possibly / maybe
library	librairie	French	bookstore
demand	talep etmek	Turkish	to request (not “to demand”)
pain	pain	French	bread
magazine	магазин (magazin)	Russian	store / shop
brave	braaf	Dutch	well-behaved
location	location	French	rental
pretender	pretender	Portuguese	to intend

Parker Pens learned this the hard way: their English slogan promised the pen “won’t leak in your pocket and embarrass you.” The Spanish translation, confusing embarazar for “embarrass,” became “won’t leak in your pocket and impregnate you.” False friends are just one of many ways languages surprise us — our collection of surprising language facts covers more linguistic quirks that trip up translators.

How to avoid: Maintain a glossary of known false friends for each language pair you work with. Flag them with “Do Not Translate” warnings. When you see a familiar-looking word in the target language, pause and verify — the similarity is exactly what makes it dangerous.

5. Blind Trust in Machine Translation

Machine translation has improved dramatically — but it still makes category errors that a human wouldn’t. In July 2025, Meta’s auto-translation mistakenly announced that a living Indian state chief minister had “passed away,” based on a condolence post he wrote for someone else. His office called the tool “dangerous” and demanded it be suspended for Kannada.

Other real cases:

❌ Arabic → Hebrew (Facebook, 2017): “Good morning” posted by a Palestinian man was auto-translated as “attack them.” Israeli police arrested him before the error was caught.

✅ A human translator understands that صباح الخير means “good morning” and nothing else.

❌ Hindi → English (Uber, 2025): “Mother Dairy ke samne hun” (I’m in front of Mother Dairy, a well-known Indian brand) became “I am facing the threat of murder.”

✅ The translator needed to recognize “Mother Dairy” as a brand name, not translatable words.

❌ English → French (Montreal Transit, 2025): “Bishop Street” was AI-translated as “Beeshop” on bus line maps.

✅ Proper names are not translated. A human wouldn’t attempt it.

How to avoid: Machine translation is a draft, not a final product. For anything customer-facing, legal, medical, or safety-critical: always have a human review the output. Different tools also have different blind spots — our OpenL vs Google Translate comparison shows where specific tools tend to stumble. The rule is simple: if the consequence of a mistake is embarrassment, you can risk it. If the consequence is legal liability or physical harm, you can’t.

6. Text Expansion and Layout Breakage

Different languages take up different amounts of space. When you translate English into German, text expands by roughly 25–40%. Italian and Portuguese: 15–30%. French and Spanish: 15–25%. Even languages with compact scripts have surprises — Russian Cyrillic words average longer than their English equivalents, and Dutch compounds rival German in length. Short UI strings expand even more dramatically — a button label under 10 characters in English can triple in German.

❌ A “Submit” button at 80px wide. German “Absenden” overflows. Italian “Invia” fits but Russian “Отправить” doesn’t. Turkish “Gönder” breaks the layout.

✅ Design buttons for 30%+ expansion, or use flexible layouts that grow with content

❌ A PDF with tight text boxes — Dutch “Sollicitatieformulier” (17 chars) replacing English “Job Application” (15 chars), but the German “Bewerbungsformular” (19 chars) spills off the page.

✅ Check layout in all target languages before finalizing. Leave white space in the source design

How to avoid: For UI design, German is your stress-test language — if it fits in German, it probably fits in everything. For documents, always do a layout check pass after translation. If you’re translating a PDF, layout breakage is especially common — our guide to translating PDFs without losing formatting covers the tools and workflow for preserving design across languages. The W3C recommends budgeting for at least 30–40% expansion from English into European languages.

7. Inconsistent Terminology

Translating the same term three different ways within one document erodes trust and creates confusion — especially in legal, technical, and medical content.

❌ In an English legal contract: Page 1 says “Service Agreement,” Page 5 says “Services Contract,” Page 8 says “Terms of Engagement.” All refer to the same document. The reader wonders: are these three different things?

✅ Pick one term for each concept and stick to it throughout the entire document and its translations

❌ Japanese medical translation (NIH, 2025): 「患者」translated as “patient” in one paragraph and “case” in the next — creating ambiguity about whether the text refers to the same person

✅ Use a glossary for domain-specific terms and enforce it across the project

How to avoid: Before starting a translation project, create a mini-glossary of the 10–20 most important terms and their approved translations. Share it with everyone working on the document. Translation memory tools (included in most professional translation platforms like OpenL) automate this — they remember how you translated a term and reapply it consistently throughout the document.

8. Translating Names That Shouldn’t Be Translated

Proper names — people, brands, places, products — almost never get translated. But machine translation tools don’t know that, and even human translators sometimes over-correct.

❌ Spanish → English (Mexico tourism site): “Tulum” became “Jumpsuit,” “Acolman” became “I Blame,” “Progreso” became “Progress.”

✅ Place names are not translated. They stay in their original form.

❌ English → Chinese: “Palo Alto” → 「高木头」(“tall wood”). It’s a city, not a lumber yard.

✅ “Palo Alto（加州一城市）” — name stays + optional parenthetical

❌ Russian → English: “Василий” → “Cornflower” (василёк means cornflower). It’s a person’s name — Vasily.

✅ “Vasily” — transliterate, don’t translate

How to avoid: Add a rule to your translation checklist: “Proper names, brand names, product names, and place names are not translated.” If a machine translation tool touches names, immediately flag it. This is one area where machines consistently fail and humans don’t.

9. Cultural Blind Spots

Translation happens between cultures, not just between languages. A perfectly accurate translation can still fail if it ignores the cultural context of the audience.

❌ White as a wedding color targeting Chinese or Indian audiences. In many East Asian cultures, white = funerals; red = weddings. In parts of West Africa, red = mourning.

✅ Research the target culture’s color symbolism before finalizing visual + text combinations

❌ A gaming company’s Ramadan campaign for Middle Eastern players featured characters with food and beer. Ramadan is a fasting holiday.

✅ If referencing a cultural or religious event, consult someone from that culture

❌ Thumbs-up emoji in messages to Middle Eastern and West African audiences — the gesture is roughly equivalent to the middle finger in those regions.

✅ Text is safer than emoji across cultures. If using emoji, check meaning in the target culture first.

How to avoid: For any campaign or document entering a new market, ask one question: “Is there a local person on the review chain?” If the answer is no, add one. The cost of a 15-minute review by a native speaker is zero compared to pulling and redoing a failed campaign.

10. Character Encoding: When “Krüger” Becomes “Kr?ger”

It’s the least glamorous mistake on this list — and one of the most common. When accented letters (é, ü, ñ, ç, ø), CJK characters, or Cyrillic text are saved or transmitted with the wrong encoding, they turn into garbled nonsense. The technical term is mojibake.

❌ German: “Krüger” → “Kr?ger” (or “Kr�ger” — the replacement character glyph)

✅ “Krüger” — always save and transmit as UTF-8

❌ Japanese: 「翻訳」(“translation”) → ”�|��” in an email subject line

✅ 「翻訳」 — check encoding settings in email clients and CMS platforms

❌ Russian: “Россия” → “Ðîññèÿ” when Windows-1251 is read as ISO-8859-1

✅ Always use UTF-8. It handles every script: Latin, Cyrillic, CJK, Arabic, Thai — all in one encoding.

How to avoid: Always use UTF-8 — it’s the modern standard and handles every script in use today. Before sending or uploading translated text, do a quick visual check: do the special characters look right? When exporting from Excel, explicitly choose “CSV UTF-8” in the Save As dialog. When in doubt, open the file in a plain text editor — if the accents are broken there, they’re broken everywhere.

Quick Reference: The 10-Second Pre-Send Checklist

Before you send, publish, or print a translation, run through these four questions:

1. Idioms: Did any phrase get translated word-for-word? → Replace with local equivalent.
2. Formality: Is the register right for the audience and context? → Default to formal when unsure.
3. Dates: Are dates in the format the reader expects? → Label the format explicitly.
4. Names: Did any proper name get “translated”? → Restore the original.

Ten seconds. Four questions. It won’t catch everything, but it catches the mistakes that make headlines.

FAQ

Does ChatGPT / Claude fix these mistakes automatically?

Not reliably. LLMs are better than older machine translation at idiom handling and register, but they still make false-friend errors, still mistranslate names, and still lack cultural judgment. They don’t know that “Mother Dairy” is a brand name — they see two words and translate them. Treat LLM output the same as any other machine translation: a strong draft that needs human review.

What’s the difference between a free tool and a paid professional translator for avoiding these mistakes?

Free tools produce a raw translation. Professional translators (and professional-grade platforms like OpenL) add: terminology management to prevent inconsistency, format preservation to prevent layout breakage, and context awareness to catch idioms and false friends. The gap is widest on mistakes #1–5 (idioms, register, false friends, and cultural nuance). It’s narrower on #10 (encoding — good tools on both ends handle UTF-8 fine).

Which mistake costs the most money?

Mistake #5 (blind trust in machine output) consistently causes the most expensive failures. HSBC’s $10 million rebrand, the $50,000 factory equipment destruction, and the Meta politician-death incident all trace back to publishing machine translation without human review. The fix — having a human read the output before it goes live — costs minutes and saves millions.

Sources

• PMC/NIH: Errors in Japanese–English AI-assisted translation (2025) — taxonomy of AI translation failure types in medical texts
• STAR Translation: When Translation Goes Wrong — real-world mistranslation case studies including medical, branding, and signage
• POEditor: 6 Famous Mistranslations in Localization — HSBC, Pepsi, KFC, and other brand case studies
• Meta apologizes after auto-translation mistakenly announces Indian state chief minister’s death (CNN, July 2025) — Meta auto-translation falsely declaring a living politician dead
• Uber auto-translation: “Mother Dairy” becomes “threat of murder” (Livemint, 2025) — Hindi-to-English auto-translation blunder
• Montreal Transit: AI labels Bishop Street bus stop as “Beeshop” (CTV News, 2025) — AI signage translation failure
• Goethe-Institut: Oh Heavenly Berry — false friend stories from German learners — collection of false friend errors including “Himbeere” (raspberry) → “Himmelbeere” (heaven berry)
• W3C: Text Size in Translation — authoritative reference on text expansion rates across languages
• SandVox: Text Expansion in Localization — expansion rates by language with short-string amplification data
• Lingoda: False Friends in Languages — comprehensive false friend tables across major language pairs
• Ulatus: Lessons from Marketing Localization Failures — brand localization case studies and cultural blind spots
• Forbes: Why Global Brands Fail When They Communicate Locally (Feb 2026) — strategic perspective on localization failures