Documents and AI
Documents in another language
A certificate in German, an invoice in Chinese: what can be done and what should be translated.
As soon as you buy or sell abroad, documents start arriving in languages nobody on the team reads. The typical reaction — translate everything — is expensive and unnecessary; the opposite — file it without understanding it — is worse.
Three different things that get confused
| Need | What it solves | When it is needed |
|---|---|---|
| Understanding what it says | Knowing whether it serves and what data it holds | Almost always |
| Being able to find it later | Locating it by its content | Almost always |
| Sworn translation | Official standing before an authority | Only when someone specific requires it |
Important
The first two are solved by reading the document on upload: data is extracted the same way, and search works on what it says, whatever the language. The third is a legal formality that costs money: request it only when someone requires it, in writing.
What to do on receipt
- 1
Upload it as it is, without waiting for translation
The original is the document; the translation is an addition.
- 2
Extract the values you care about
Dates, amounts, identifiers. They rarely depend on language.
- 3
Note in your language what it is
One line is enough: "conformity certificate from supplier X, valid until…".
- 4
And if a translation is needed, link it to the original
Both together; a loose translation loses its reference.
Watch out
Watch dates and numbers. Formats vary by country: the same document can read as 3 June or 6 March, and thousand and decimal separators swap. It is the costliest and quietest error.
When a sworn translation really is needed
Otherwise, understanding the document and being able to find it is all you need, and that requires translating nothing.
Worth knowing
If you send requests to suppliers abroad, sending them in their language changes response rates considerably. And what comes back will still be in their language: that is normal and does not stop you working with it.
›Are non-Latin scripts read?
Yes, though quality depends heavily on the scan.
›Is machine translation enough to decide?
To orient yourself yes; for a decision with consequences, verify.
›What if nobody on the team understands it?
Note in your language what it is and who it is from: that alone prevents 90% of problems.
A real case
The situation
A company piles up Asian supplier certificates, untranslated and unuploaded.
What you do
- Uploads them as they are and extracts dates and identifiers
- Notes in English what each one is
What you get
It can find them and know when they expire without paying for a single translation.
The situation
A document arrives in a language nobody reads.
What you do
- Extracts the fields that are needed
What you get
The document is usable without translating it all.
The situation
It is fully translated when three fields would have done.
What you do
- Translates only what needs understanding
What you get
The cost matches the need.
The situation
A client asks for the document in their language.
What you do
- Asks whether an official translation is needed
What you get
You translate what is appropriate.
The situation
The supplier always sends in their own language.
What you do
- Agrees the language on the order
What you get
It arrives usable from the first shipment.
The situation
It is filed without knowing what it says.
What you do
- Notes what the document is when filing it
What you get
It is located later without opening it.
This article answers
- document in another language what do i do
- reading a foreign certificate
- when a sworn translation is needed
- extracting data from a foreign-language invoice