Machine Translation Guide and Tips
(Originally published April 2012)
Improvements in fully automated translation quality, particularly in the last 5 years, have attracted attention, and led to some remarkable successes in commercial use. In this article, we look at how to get off on the right foot.
First, we need to introduce some terminology. Fully automated translation is called Machine Translation, or MT for short. This abbreviation is easy to confuse with TM for Translation Memory. Both MT and TM are software tools for translation. Translation memory is a database technology that saves previously translated sentences for reuse later when the same sentence, or a very similar sentence, needs to be translated again. Machine translation technology employs statistical or linguistic algorithms to translate new sentences from scratch. Machine translation developers have hoped for 50 years that their software would be used by translators, but until recently, translation quality was almost never good enough to revise into a publishable product cost effectively. Translation memory, which simply lets a translator be more efficient and reuse their own work, has an immediate payoff and became an accepted norm in the translation industry. Syntes has used TM since the early 1990s for most customer projects!
Can MT now be used to produce high quality translations?
Machine translation will not produce well-written translations by itself. However, it can be an efficiency tool under the right circumstances. For a translation to be readable and appropriate for the end user, it must be revised or post-edited by a qualified linguist. Materials that reflect an organization’s brand identity, text that requires the utmost accuracy and good style, and publication-quality materials, in general, are likely to need several revision steps. Machine translation offers productivity by saving a lot of typing and by retrieving terminology for the translator. But, the MT software doesn’t understand the meaning and nuances of the text, so it cannot detect its own errors in translation. To a machine translation system, the sentence, “Time flies like an arrow” might bring up a reference to a particular kind of insect, “time flies” that have an affinity for arrows. To a non-living machine, this is just as plausible as the intended meaning about the fleeting nature of time.
When people try one of the free online MT systems like: Google Translate, Freetranslation.com, Microsoft Translator, or others, they may be impressed that the system produces some understandable output. But the reasonable-sounding “translations” will contain mistakes that make the output unsuitable for use as is. In addition, the free online MT systems are “general purpose” systems. That means that they do not contain the industry or proprietary terminology that is key to customizing machine translation to make it cost effective.
How can MT be used effectively?
1. Determine whether a text is suitable for machine translation. Texts that will yield real productivity gains with MT:
• Contain objective information that is clearly written. Examples would include text of a “how to” nature such as user manuals or on-line knowledge bases in which the goal is to explain how to solve a specific problem or complete a particular text as opposed to needing to be linguistically perfect.
• Contain terminology that is clearly and consistently used throughout the text.
2. On the other hand, machine translation does not produce highly useful translations of texts that are intended to sell or persuade the reader, text where the style and nuance of the language are especially important, or texts that rely on humor or cultural references for their impact.
3. Invest in customization of the MT system up front. Much of the potential benefit from using MT is in automatic retrieval of terminology.
4. Use qualified translators who are trained in post-editing machine translation to revise the text.
How can I get started?
1. Assess the material that you need to translate, and identify texts or classes of publications that are suitable. User manuals are often good candidates because they are written in a clear, succinct style. Avoid sales and marketing materials whose impact will be lost in a literal translation.
2. If you have not already identified your key terminology and its standard translation, it is important to do so for MT customization. Organizations may have between a few hundred and tens of thousands of terms.
3. Let us help!
We can help you to identify the key terminology that needs to be incorporated into the MT system, and help with developing standard translations for those terms.
Syntes has machine translation software that can be used for your project, and we can customize it for you. The investment in customization will pay off on all future jobs.
Whether you use our machine translation service, or your own customized MT solution, we have qualified post-editors who are happy to revise the output to the level you require. Contact us today for information or questions.