Top Ten Pre-Editing Tips (for Even Better Machine Translations)

Top Ten Pre-Editing TipsShare your news

Do you have news for your fellow language professionals in Europe?

The Language Sector website makes your article, opinion or other content (written in any European language) as accessible as possible to as many language professionals as possible in as many languages as possible.

With a limited number of relatively small interventions, you can substantially improve the accessibility of the machine-translated information in your post, article, opinion, or other content.

Here are the top ten tips for optimal pre-ediing from our home expert Valentin Descamps.


Post-editing is good 

With the optimal accessibility of your information as a goal, we naturally make use of the technologies that the language sector itself produces. As you know, the quality of the translation machines improves quickly. But where necessary, you can always fine-tune every article on The Language Sector website to the desired level of quality in the post-editing environment integrated into our website


Pre-editing is even better

With good post-editing, you can significantly improve the accessibility of your information. That's right. But even more time- and cost-effective than a good post-editing is a good pre-editing of your source text in order to achieve optimal translationability by the translation machines. With a limited number of relatively small interventions, you can substantially improve the accessibility of the machine-translated information in your post, article, opinion, or other content.


Top Ten Pre-Editing Tips

Here are the Top Ten Pre-Editing Tips from our home expert on the matter, Valentin Descamps. Valentin obtained a master's degree in translation from the University of Mons in Belgium last year. He specializes in translation from Russian and Spanish into French.

For The Language Sector website, he presents you ten simple but effective techniques that make your text more translation machine-friendly and therefore more accessible in translation.

In summary, this is his advice: with 1 well-aimed pre-edit you will save the time and cost of 20 post-edits on The Language Sector website.
So don't hesitate!

 

Ghent, 18 December 2020, Valentin Descamps - The aim of pre-editing is to generate the most understandable machine translation possible in as many different languages as possible. To the extent that it is needed, the post editor can then "idiomatize" the machine translation.

The tips I present do not improve machine translations to the same extent in all languages. I tested them on French, Russian and Spanish machine translations from Dutch. The tips make 9 out of 10 French and Spanish translations more understandable, and 6 out of 10 Russian machine translations. It may happen that a pre-edited tip that makes the French and Spanish machine translation more understandable just has the opposite effect in the Russian translation.

To evaluate the effect of the tips, I used Google Translate. In practice, other translation machines can show a slightly different effect. If
for you the Russian translation is more important for any reason than the Spanish or the French, then you can consider a pre-editing that focuses more specifically on the Slavic languages.


After this I present my Top Ten Pre-Editing Tips.

Would you like to respond, do you have a question or a suggestion? Please contact me viavalentin.descamps@detaalsector.be. I look forward to all possible feedback.

You can also download the extended version of this article inPDF. The pdf illustrates each tip with different examples (NL > FR / ES / RU).

 

Tip 1: avoid abbreviations

Avoid abbreviations. Abbreviations are often difficult for translation machines to translate. Write abbreviations in full.

 

Tip 2: Avoid sentences without a verb

Sentences without a verb are difficult for the translation machines. Add a verb, otherwise a verb risks missing from the translation.

 

Tip 3: Clarify the gender of own names

Personal names can be specific to a language or country.

For a reader with no background knowledge about the country and for a translation machine, it is not always clear whether a name refers to a man or a woman.

Tip: Use male or female referring words as often as possible in the immediate vicinity of personal names, so that the machine may be able to deduce the person's gender from context.

 

Tip 4: Avoid too long sentences

Although neural translation machines can process much longer sentences than the previous generation (statistical-based) machines, it is still better to split up really long sentences.

 

Tip 5: Play with the structure

Keep the structure of your sentences as clear as possible. Don't try to cram two ideas into one sentence. If two sentences have a logical connection to each other, use connecting words to indicate the relationship.

 

Tip 6: Play with the context

Modern translation machines look at every word in the context of (at least) the whole sentence. If you introduce a difficult word, introduce it in a context that contains a lot of related information that the translation machines can easily understand. This increases the chances that they will translate the sentence correctly.

 

Tip 7: Play with synonyms

If a very specific word proves really difficult for the translation machines, set yourself up a bit smoothly and be prepared to consider a synonym. Think of the optimal accessibility of your information in as many languages as possible.
Is that one super precise word with that one super precise meaning really so important that it is justified that you create twenty translation problems with it and make twenty post-edits necessary.

 

Tip 8: Avoid language-specific idioms

In Dutch, people send their cat when they don't show up for an appointment. Most translation machines then translate such idioms literally ("send his cat"), and that is often not understandable to the reader in translation. It is better to write in concrete, explicit language.

 

Tip 9: Correct language errors 

Set the bar high. Be strict about the quality of your own text. Language errors in the source text can lead to language errors in the target text. So, check your grammar, spelling, punctuation and the coherence of your source text. There are tools that can help you do that.

 

Tip 10: write in one language 

Is your text in Dutch? Write as much Dutch as possible. That makes sense, but some people use a lot of English words and expressions in their Dutch texts. No problem for the English translation machine, but often all the more so for the other machines, which then perform less well (in French and in the other Romance languages for example).

 

Download the full version of this article (with examples) inpdf.

 


Powered by CrossLang

Author: Valentin Descamps

Machine translation: SDL Machine Translation (previously SDL BeGlobal)

Post-editing: No post-editing

Source language: Nederlands (nl)


Additional information