How many times have we received a pre-translated but text that was not good at all?
But where is the problem? The machine translation system? It's possible.
Or is it the source text? Very likely.
But why does the translation quality depend on the text?
It is very easy to understand! A machine translation system translates sentence by sentence, regardless of context.
So, what happens if we interrupt a sentence with Enter?
Or what happens if we include pronouns that refer to previously written sentences?
Or what about idioms? Can the system handle these idioms?
I do not think so.
Let's take a look at some examples translated from German to English by a NMT system:
- Du hast von Tuten und Blasen keine Ahnung.
- Du verstehst nur Bahnhof.
- You don't know how to do things like turkeys and bubbles.
- You only understand the station.
Or wouldn't it be better
- Du verstehst nichts davon.
- You do not understand anything about it.
Omission of articles and prepositions (HMI)
- Bewegung Ganghebel links Ruhestellung -> Betriebsstellung
- Movement gear lever left Sleep -> Operating position
Or wouldn't it be better
- Bewegung des linken Ganghebels aus der Ruhestellung in die Betriebsstellung
- Movement of the left gear lever from the rest position to the operating position
- Der Mann hat einen Hund namens Snoopy. Er hat Durst.
- The man has a dog called Snoopy. He's thirsty
Who is he? The man or the dog called Snoopy? So, wouldn't it be better to be more specific?
- Gestern Nachmittag habe ich
- mir ein Buch gekauft
- Yesterday afternoon I was
- bought a book
Properly segmented, the system knows how to handle the text:
- Gestern Nachmittag habe ich mir ein Buch gekauft.
- Yesterday afternoon I bought myself a book.
The correct writing method for a machine translation is therefore important. I have only given a few examples here, but when writing the documentation you should pay attention to both the reader and the writing style for machine translation. To do this, you should follow rules. Some companies that have an NMT system also have guidelines for writing documentation correctly. This increases the comprehensibility of the text and the translatability and reduces post-editing.
But what can a post-editor do if he notices that a text is not translatable?
He should make the customer aware of this. Otherwise, the post-editor will take longer than for a normal translation. And that is not how it should be.
- For companies: Clear writing rules
- For translation agencies: Look at the document before you translate it automatically.
- For post editors: Indicate when a text has not been well translated by an NMT system.