Writing Content For Machine Translation: Keep it Simple, Recycle, Repeat
By United Language Group
Translation errors can be embarrassing – sometimes extremely so.
Recently, Coke learned this the hard way when an advertisement on its vending machine inadvertently said "Hello, Death" in Maori.
Lucky for the soft drink giant, its reputation wouldn’t be foiled by one linguistic misstep. Other smaller companies might not be so lucky.
As Machine Translation (MT) continues to grow in popularity, companies should be aware of its margin for error. Documents that are intended for public use necessitate an editor’s eyes after going through an MT system to account for complicated phrasing a computer wouldn’t recognize.
Post Editing helps detect MT output errors after the fact, but pre-editing is a great way to create more successful MT outcomes. Writing content for MT that’s simple and to the point will mean less work after machine translations are completed. With MT, Post Editing will always be necessary, but a good pre-edit will make it easier.
Keep It Simple
Wherever possible, content that’s being prepped for MT should contain short, simple sentences. There are many robust MT systems out there, but even the best will have trouble identifying jumbled, rambling segments of text.
In the same vein, it’s a good idea to get rid of unnecessary words in your text. However, sentences that are too pithy may be hard for an MT system to understand, too. Finding a happy medium between too long and not long enough is ideal.
Long, “two-dollar” words won’t go over well in an MT environment. Short expressions that don’t require a dictionary to understand are going to make for a much easier Post Editing process. Try to avoid the use of acronyms, too, as an MT system will likely try to translate these three- or four-letter abbreviations as words themselves.
The more you can use the same phrase in MT content, the better. MT systems notice repeated phrases and have an easier time translating something if it’s already been detected in a text. Often writers try to say the same thing in different ways to spice up their copy; in the MT world, this usually makes things more difficult.
If you need to say something more than once, say it the same way. There’s no need for variety if it runs the risk of sacrificing accuracy.
Remember the Basics
MT systems don’t deal well with copy that contains grammatical or spelling errors. If a word is spelled incorrectly, a computer will definitely not be able to translate it accurately -- and excessive or misplaced punctuation marks in a text will jumble things up even more. For example, if a period is missing at the end of a sentence, an MT system won’t know that a complete thought has ended.
Improve Your Output
These tips will help you create improved MT output, but they don’t represent any sort of panacea. MT systems cannot create the same translations that experienced translators are capable of. But, that’s not to say that using the new technology is fruitless – MT creates quick translations for less money, and that’s a welcome attribute for anybody.
Leveraging the combined capabilities of linguists and strong MT systems will provide the best translation results.
ULG's insights delivered straight to your inbox.
Thoughtful editorials from industry experts delivered weekly in bite-sized pieces.
Google+ Facebook LinkedIn Twitter Instagram
United Language Group
Andrew is a staff writer at United Language Group. He is especially interested in digital marketing, translation technology, as well as cultural and linguistic studies.
Ready to learn more?