How to Create and Deliver Intelligent Information

Automatic. Seamless. Good. What do machine translation and chocolate gifts have in common?

Artificial intelligence has long been part of everyday life in the translation industry. That’s not surprising, as it promises time savings and an increase in efficiency. Providers advertise with automated processes, seamlessly integrated solutions and good results. However, our experience has shown that output is like a box of chocolates: You never know what you’re gonna get.

Using machine translation reasonably – in 3 steps

It’s always a good idea to consider in advance what is suitable for whom and for what purpose – not only when it comes to chocolate. The following preliminary considerations make sense if you want to make the most appropriate use of MT technology.

Step 1: Is my text suitable for MT?

Not every type of text is equally suitable for machine translation. In our experience, using generic systems, the following text types have generated the best output:

  • Contracts
  • Technical instructions
  • Software documentation
  • Press releases
  • Newsletters

Advertising texts, puns and texts with considerable length restriction, such as display texts, are completely unsuitable for machine translation systems. That kind of output could look like this:

Source text (German)Output (English)
Luftvert.-Taste hi re Luftvert.-Taste hi liAir key hi re Air key hi li
Der arme Baum, besser nicht umfahren sondern umfahren.The poor tree, better not to drive around it, but to drive around it.

Another important factor is the quality of your source text: it strongly affects the output. Spelling mistakes in the source text, for instance, may have fatal consequences:

Source text (German)Output (English)
FarbtemperatureinsatellungColour temperature insatellite

As you can see, it pays off to “pre-edit” the source text and, above all, correct errors that immediately affect the output. This is far less time-consuming and cost-intensive than subsequently post-editing your texts – possibly in several languages.

Step 2: Which engine is right for us?

It feels like MT engines are already a dime a dozen, and new ones are released almost every month. The large number of engines reflects the variety of their features:

  • Online services
  • Available online and free of charge
  • Holistic solutions for machine translation
  • Complex machine translation solutions tailored to the translation industry
  • Proprietary engines for machine translation
  • You, of course, develop your own engine and feed it specifically with your data.

So far, so good. But which solution is right for your company? To answer this question, it is wise to establish selection criteria such as cost, available languages, etc.

Of course, it is essential that your criteria meet your company’s needs.

Step 3: What should I consider during post-editing?

The output of machine translation is unpredictable, especially as machine translation systems are constantly learning. As a rule, you will receive a rather fluent, legible text which, at first glance, does not contain any errors in grammar or spelling. Incorrect references to content, on the other hand, are often only noticed in direct comparison with the source text, as you can see in the following example:

Source text (German) Output (English)
Dieser Bereich wird zur Verhütung von Straftaten durch die Polizei videoüberwacht.This area is video-monitored to prevent criminal offenses by the police.

Therefore, if you intend to publish your text, it is crucial to have a human translator post-edit it.

The icing on the cake: How do we integrate machine translation into our translation process?

Processes have become one of the most important elements in translation management. Even though machine translation is not yet equally suitable for all languages, its seamless and efficient integration into existing processes is certainly beneficial.

It is especially important to connect the engine to your CAT tool. This enables you to use existing entries in the TM for your translation instead of the machine output. Fuzzy matches, i.e. almost identical translations, can also be integrated. They are often more valuable to the post-editor than the raw output from the engine: Although the content may have to be adapted, fuzzy matches generally ensure that your translation is consistent.

Last but not least, it is up to you to determine whether to store your final product in a translation memory database, and if so, which one to choose: After full post-editing, the quality of your translation is comparable to that of a human translation. Thus, it makes sense to save it in the same TM as the entries originating from a human translation. In contrast, raw machine output or lightly post-edited machine translations should not be stored in your regular TM, as they do not meet the high-quality standard that is required.

Summary

Machine translation systems translate automatically. They can usually be seamlessly integrated into existing process landscapes but deliver unpredictable output. Turning it into good output requires the judgment of a human being who, through his or her expertise, makes the necessary adjustments in the source text and quickly and carefully corrects typical output errors in the target text. That way, you always know what you’re gonna get. And, as with all projects: Test before you go live.

Your appetite for this exciting topic has not yet been satisfied?

Find out more at tcworld conference 2019 by attending the German lecture “Effizienz 4.0: Maschinelle Übersetzung und Post-Editing” by Diana Winokur and Oksana Mikitisin! Satisfy your hunger for information on Tuesday, 12 November at 9:15 a.m. in room C5.2. Bon appetit!

The Authors

Diana Winokur

Consultant / Process Manager at Transline Deutschland GmbH

Due to her experience in advising industrial customers, she has specialized in process flows. She has successfully set up and implemented machine translation services both internally at Transline and for customers.

Oksana Mikitisin

Team Leader Translation & Proofreading at Transline Deutschland GmbH

As a translator, proofreader and post-editor of technical documentation, she knows exactly what it takes to optimize translation processes.

Lisa Mümmler

Onlineredakteurin & Bloggerin at tekom
Lisa Mümmler studied German language and literature and philosophy at the Ruprecht-Karls-University in Heidelberg. She is a passionate online editor, social media manager and blogger. Since October 2018 she has been supporting tekom's marketing team, including the iiBlog.

Like it? Share it! Spread the Word...

Sag uns jetzt deine Meinung per Kommentar!