5 MISCONCEPTIONS ABOUT MACHINE TRANSLATION
by Diego Cresceri
Machine translation is one of the "hottest" topic in localization, and, I must admit, one of those fascinating me the most.
When it comes to machine translation, the first thought often goes to Google Translate: who has not tried it at least once in their life to translate a sentence from some unknown language or search for that word that just didn't want to leave the tip of the tongue?
Many are skeptical in relation to the professional use of this technology, and I often hear different opinions on this topic.
It is not easy to determine who’s right: rather, a case-by-case assessment is needed. There are first a few things to clarify though.
First and foremost, machine translation is not limited to Google Translate. As you’ll see, there are several machine translation programs out there (both on the cloud and not) and many IT giants have developed their own system for localizing their content.
Also, you can't talk about professional machine translation without considering integration with CAT (Computer Aided Translation) tools and pre- and post-editing, which necessarily require human work.
Not only Google Translate then, or Babelfish, but much more than that.
In this post I tried to collect some of the main misconceptions about machine translation, which I'll try to somehow disprove, without claiming to address the topic in full in a few lines.
So here are the 5 misconceptions about machine translation.
1. You cannot control quality of raw machine translation output in advance
Have you ever heard the saying "garbage in, garbage out"?
Applying this concept to machine translation, raw automated translation (raw output) depends on the information we are able to provide to the machine translation engine. There are several ways to control raw output quality in advance. Among them are certainly "controlled language" and "pre-editing".
No matter which machine translation engine is used, the more syntax of the source text is simple, the easiest it will be for the software to understand source text and translate it correctly. Applying "controlled language" implies following simple rules during source content drafting. This includes:
Pre-editing includes all operations that can be performed before the source text is "fed" into the machine translation engine.
Pre-editing can be used when the source text was not written with "controlled language " to simplify syntax, to eliminate the use of synonyms and pronouns, and to ensure source terminology consistency.
Tagging specific terms or software entries as not translatable in order to make it easy for the engine to recognize them can also be considered part of the pre-editing.
For example, you might tag your brand name and product names as untranslatable, so that the machine translation engine will recognize them as such and leave them unchanged.
Applying controlled language and pre-editing techniques will make the machine translation software’s life easier, and, above all, provide a better raw output.
This means that the automatically translated text will be more understandable and, if post-editing is foreseen, productivity will be much higher, which leads to cost and time savings.
2. Post-editing effort is the same for all languages
Different languages have different traits and different syntax.
For this reason machine translation works better in some language pairs and worse in others.
One of the main challenges for machine translation, for instance, is recognizing whether a term is masculine or feminine. From this point of view machine translation will work best when the target language does not have such a requirement.
Generally speaking, machine translation gives better results when source and target language have similar syntax and grammar rules.
3. Quality of machine translation is so poor that raw output is rarely useful
For years Google Translate was pointed at as a useless gadget, only able to generate translations that, at best are funny, in worst cases horrifying. Using Google Translator for an important multilingual newsletter without adding post editing might actually not be the best marketing strategy ever.
However, saying raw machine translation output is always too creepy to be useful is certainly an overstatement.
Let's have a look at where machine translation may work better than traditional (or human) translation, specifically in projects with tight deadlines and huge volumes of words.
4. Machine translation will be useful in the future, but we're not ready yet
Google, eBay and Microsoft are only some of the companies that have been investing large sums on research to develop proprietary machine translation in order to able to localize their growing content in shorter timeframes.
Moreover, the market is full of companies of all sizes that provide machine translation services both as off-the-shelf solutions (Systran) and on the cloud (SDL, Omniscien and Translated.net to mention a few).
In addition, all major CAT tools offer integration with different machine translation providers (including those mentioned above).
Finally, thousands of companies (including global brands such as Adobe, Amazon and John Deere) are successfully implementing machine translation (with and without post editing) to reduce timelines and costs of their localization projects.
This makes me think machine translation is far from being something futuristic, but it is actually very present and tangible.
5. Machine translation will replace human translation altogether
I chose to leave this question at the end because it’s often what worries my students the most and the topic that makes skeptical linguists laugh and professional translators get angry.
Once again, there is no single answer to back up the most catastrophic thinker or the supporters of technology at all costs.
Research on machine translation from big players goes on, and recent results (or at least claimed results) in relation to quality of machine translation leave no doubt: machine translation will continue to grow, with more and more content being translated automatically.
No need to worry though, this does not mean machine translation will replace human translators, at least in the medium term.
Like with many tech innovations, machine translation and human translators will continue to co-exist, without being mutually exclusive.
As we have seen, many companies use machine translation as part of their localization processes, but not exclusively.
For its intrinsic traits of creativity and originality, some content (marketing content in its different forms for instance) is not suitable for machine translation and should always be left to the inspiration of a translator (or transcreator) so that originality and creativity are preserved in the target language.
This post was originally created in Italian, my mothertongue. Given the topic, I decided to run an experiment, and have it translated automatically by Bing + post-edited by me (again, I am an Italian mother tongue, so please be tolerant of any odd sentence or term).
Scope: 1258 words
Post-editing without final reading: 39 minutes
Post-editing with a round of final reading: 59 minutes
This make 1935 words/hour for the post-editing without final reading & 1279 words/hour for full post-editing with final reading.
Do you still think machine translation is not good enough?
If you want to check whether machine translation is suitable for your content, contact us and we'd be happy to give you a free assessment.
Diego Cresceri - Founder and CEO of Creative Words, he does not deny his past but never looks back. An absolute lover of languages, he's an incurable optimist and cannot wait to se what the future holds.
"If you talk to a man in a language he understands, that goes to his head. If you talk to him in his language, that goes to his heart." Nelson Mandela