“The Seller shall compensate damages” - Google Translate rendered this as “The seller can compensate for damages” in German. One word - and a binding legal obligation turned into an optional goodwill gesture. The client submitted this translation to court, and the other party pointed at the text: “It says ‘can’. He’s not obligated.” This isn’t a made-up story - research shows machine translators get modal verbs wrong in legal texts roughly one in four times.
If you’ve ever thought about running your contract through Google Translate or DeepL to save time and money - this article explains why that’s a bad idea and what you’re risking. And if you’re a translator - you’ll find concrete examples worth showing clients who ask “why not just use Google?”.
What goes wrong: specific errors¶
Machine translation handles simple texts just fine - restaurant menus, social media posts, microwave instructions. But legal documents are a different planet where every word carries precise legal weight, and an “approximate” translation can flip the meaning of a document entirely.
Modal verbs: shall ≠ can¶
The most dangerous type of error. In legal texts, shall, may, and must aren’t stylistic choices - they’re three distinct levels of obligation. Shall means “is obligated to” (a strict contractual duty), may means “has the right to” (optional), must means “is required to” (a necessary condition).
Google Translate regularly confuses these constructions. A study of legal texts translated through Google Translate found 24 instances of incorrectly translated modal verbs in a single set of contracts. This isn’t a minor issue - it’s the difference between “the company is obligated to pay compensation” and “the company may pay compensation.”
In German, this problem is even sharper. “Muss” (obligated), “soll” (should), “kann” (may), “darf” (permitted) - each carries specific legal weight, and machine translation regularly mixes them up.
| Original | Machine translation | Correct translation | Consequence |
|---|---|---|---|
| shall compensate | can compensate | is obligated to compensate | Legal obligation disappears |
| may terminate | must terminate | has the right to terminate | A right becomes an obligation |
| must be notarized | should be notarized | must be notarized | Mandatory requirement weakened |
Terminology traps: when words don’t mean what they seem¶
Legal language is packed with terms that look like ordinary words but carry entirely different meanings in a legal context. Machine translators pick the most common meaning - and miss the mark.
Consideration - in everyday English, it means “thinking about something” or “being thoughtful.” But in contract law, consideration is what each party gives and receives in a deal - the mutual exchange that makes a contract binding. Without the correct translation, the entire section on party obligations becomes nonsensical.
Gesellschaft - in German, this means both “society” and “company.” Machine translation picks “society” (because it’s more common in general texts), but the contract is referring to a specific legal entity. One wrong translation - and the partner on the other side of the deal doesn’t understand who the contracting party actually is.
Default - in tech contexts, it means “standard setting.” But a default judgment is a court ruling issued when the defendant fails to appear. Machine translation produces something like “judgment by default setting” - and the lawyer on the other side has no idea what type of ruling is being referenced.
Legal systems don’t translate 1:1¶
A German GmbH isn’t a British Ltd or an American LLC, even though all are “limited liability companies.” Different minimum share capital requirements (EUR 25,000 for GmbH), different governance rules, different founder liability. Machine translation doesn’t know these nuances - it just plugs in a supposed “equivalent” from another legal system.
The same applies to “Arbeitsvertrag” (employment contract) - German and Ukrainian employment contracts are governed by fundamentally different legislation. Translating an employment contract requires understanding both legal systems, not just swapping words.
What courts say about machine translation¶
These aren’t theoretical risks - courts have already ruled on machine translation.
In 2021, a U.S. federal court found that Google Translate has “an alarming capacity for miscommunication and error” and isn’t reliable enough even for obtaining a person’s consent during a police stop. An officer used Google Translate to ask permission to search a vehicle - the court ruled the “consent” invalid because the translation was inaccurate.
In another case, a court rejected Google Translate renderings of Korean web pages, calling the results “nonsensical.” Courts refuse to accept documents translated by automated tools as evidence.
Immigration lawyers are also reporting a trend: applicants who submit machine-translated documents face rejections or processing delays. An inaccurate translation in a visa application can cost months of waiting.
The bottom line: a machine-translated legal document has no legal standing. No court, notary, or administrative body will accept it as an official document. If you need a translation for submission to a German embassy or court - only a sworn translator or notarized translation will do.
Confidentiality: where your documents end up¶
There’s another problem most people don’t think about. When you paste contract text into Google Translate or the free version of DeepL - that text lands on the company’s servers. Here’s what can happen next.
Free translation tools store and analyze the texts you submit. Google explicitly states in its terms of service that it may use your content to improve its services. This means confidential contract terms, amounts, party names, trade secrets - all of it enters a system where you don’t control who has access.
For legal documents, this is a double problem. First, you’re breaching your client’s confidentiality. Second, if the document contains personal data (names, addresses, identification numbers) - you’re violating GDPR, which carries fines of up to EUR 20 million or 4% of annual global turnover.
On a translation industry forum, one professional shared this story: “A freelancer at our agency pasted a confidential merger agreement into a free online translator. The client found out and terminated the contract with our agency. We lost a client who was paying EUR 50,000 per year.”
What about paid versions?¶
DeepL Pro doesn’t store texts and doesn’t use them for training. That’s better than the free version, but the accuracy problem doesn’t go away - you just get the same errors, only confidentially.
Research: how accurate is machine translation for legal texts¶
A study of DeepL’s accuracy on legal documents found: 53% of translations were accurate, 38% were partially accurate (needed editing), and 8.5% were inaccurate (changed the meaning). That means roughly one in ten translations alters the meaning of the legal document.
For comparison: on general texts (news, technical documentation), machine translation accuracy is significantly higher - 80-90%. The difference comes down to three characteristics of legal texts that machine translation can’t handle.
Why legal texts specifically are the problem¶
Self-reference. Legal texts constantly refer to themselves: “pursuant to Section 3.2 of this Agreement,” “as defined in Article 1.” Machine translation doesn’t “see” these connections and may scramble numbering or lose cross-references.
Context dependency. The word “address” in one paragraph means a mailing location, and in another means “to deal with an issue.” The word “will” can be an auxiliary verb (“I will send”) or a legal document (a last will and testament). Machine translators frequently pick the wrong meaning because they don’t understand legal context.
Consistency = accuracy. In legal texts, one term must be translated the same way from the first line to the last. If “Vertrag” is “agreement,” then it should be “agreement” all 47 times - not alternating with “contract” and “arrangement.” Machine translators often swap in synonyms for “better style” - but in a legal text, that can make it look like two different subjects are being referenced.
When AI translation can still be useful¶
It would be dishonest to say machine translation is completely useless for legal texts. There are situations where it helps - but with clear boundaries.
Where AI translation works¶
- Quick document review. You received a 40-page contract in German and want to quickly grasp what it’s about - general content, structure, key terms. Machine translation gives you enough understanding for that.
- Internal communication. You need to quickly explain the contents of a letter from a German partner to a colleague. For internal use, “80% accuracy” is acceptable.
- First draft for MTPE. An experienced translator can use machine translation as a starting point, then thoroughly edit the result. This is called MTPE (Machine Translation Post-Editing). The key word here is “experienced” - a junior translator might miss a critical error.
Where AI translation absolutely CANNOT be used¶
- Court documents - courts don’t accept them
- Contracts that will be signed - legal consequences of errors are unpredictable
- Visa or immigration documents - you risk getting rejected over an inaccurate translation
- Notarization - a notary won’t certify a machine translation
- Anything containing confidential information - data breach risk
Comparison: machine translation vs professional translator¶
| Criterion | Machine translation | Professional translator |
|---|---|---|
| Cost | Free or from $20/month | From EUR 30-60 per page |
| Speed | Seconds | 1-5 business days |
| Legal terminology accuracy | ~53% accurate translations | 95-99% with specialization |
| Legal standing | None | Yes (sworn translator) |
| Confidentiality | Breach risk | Protected by NDA |
| Liability for errors | Nobody | Translator/agency |
The price difference looks big - but compare EUR 50 per page of translation with potential losses from a mistranslated contract. One user on an expat forum wrote: “Saved EUR 200 on document translation for my visa, used Google Translate. Got rejected, had to reapply, lost 3 months, and paid for a proper translation anyway.”
What to do instead of machine translation¶
If you need to translate a legal document - here’s a step-by-step guide.
1. Figure out what type of translation you need. For court or a notary, you need a sworn translation. For internal use, a standard professional translation is enough. The difference between translation types is covered in a separate article.
2. Find a translator who specializes in legal texts. Not every translator works with legal documents. In Germany, you can find sworn translators in the justiz-dolmetscher.de database.
3. Provide context. Tell the translator what the translation is for and where it will be submitted. This helps them choose the right terminology.
4. Ask for a glossary. A good translator will compile a glossary of key terms before starting work - and confirm it with you.
And if you have a Word or PDF document and need a quick translation for review purposes (not for official submission) - ChatsControl will translate it in minutes while preserving formatting. For preliminary review and internal communication - it’s a perfect option. For official legal documents - use a certified translation service.
FAQ¶
Can I use Google Translate for translating a contract?¶
For getting a general idea of the content - yes, but only for yourself. For signing, court submission, or notarization - absolutely not. Courts have already found Google Translate translations unreliable, and a notary simply won’t accept a document without a qualified translator’s signature.
Why doesn’t DeepL work for legal documents either?¶
DeepL is more accurate than Google Translate for European languages, but research shows only 53% fully accurate translations of legal texts. Nearly one in ten translations changes the document’s meaning. Plus the free version stores your texts on its servers - that’s a confidentiality risk.
How much does professional legal translation cost?¶
In Germany, a sworn translation runs EUR 40-80 per page, with certification included in the price. Yes, it’s more expensive than regular translation - but cheaper than a lawsuit over a mistranslated contract.
Do courts and notaries accept AI-generated translations?¶
No. A machine-generated translation has no legal standing - no court or administrative body will accept it as an official document. You need a translation certified with the signature and seal of a qualified translator (in Germany - a sworn translator, vereidigter Übersetzer).
What is MTPE and does it work for legal texts?¶
MTPE (Machine Translation Post-Editing) is when machine translation is used as a draft, and then an experienced translator thoroughly edits the result. For legal texts, this approach only works if the editor is a legal translation specialist - otherwise critical errors can slip through unnoticed.