For many years, the quality of a translation was assessed by linguistic aspects only. Large enterprises maintained internalk language service departments which established themselves as guardians of language. Translations received were dissected linguistically. Sophisticated evaluation methods, such as LQA, were developed to categorize and rate all errors. If you ever performed such an evaluation together with your colleagues under lab conditions, you certainly realized how subjective these categorizations are. This frequently caused the style of the language to be weighted higher than its technical correctness in turn leading to many discussions.
Example 1: Style guides
Language service departments often provided style guides – collections of explanations, instructions, and style examples – to their providers. One of our largest clients provided a style guide with more than 120 pages. It was used as the basis for the evaluation of the providers‘ translation quality. If the average quality dropped by a certain threshold, further orders were jeopardized. we did not have this problem, since our average was much higher. But we felt that many of our „errors“ were not justified as such. Hence we requested a meeting with the quality representatives. We discussed a number of errors reported as wrongly categorized by my colleague. After 5 minutes, I sat back and relaxed., although I usually am quite active in meetings. The reason: the two quality representatives started an intense discussion between themselves. In the end, they had to admit their style guide was not clear and included contradictions at several points. Finally, none (!) of our „errors“ held up.
Language is not logical. Thus, any attempt to force it into a corset is doomed to fail.
Example 2: Technical language
One of our customers appointed us to translate documentation and help for a software application (that was at a time when there still were user manuals). For this purpose, the US company had hired a German with linguistic diploma and put him in charge of defining the terminology for this application. We got terminology and style guide and followed them. Our translation was well received.
Later, we were asked to translate version 2. We were highly surprised that we could not reuse any part of our first translation, since the whole terminology had changed. The first version was delivered in perfect German, but had nothing to do with the technical language used in its subject matter area.
Instead of a linguistic expert, a subject matter expert woud have been the better choice.
Example 3: The time factor
Time? That is a different topic!
Time, however, may become a quality aspect of its own. What good is a perfect and sophisticated language if the text is not available in time?
For another customer, we translated the documentation of ink-jet printers into many target languages. The challenge: after development, time was taken to write the documentation. And after that, the translation needed more time. And only then the device could be sold. If you realize that the shelf life of such a product is less than a year and that the period, in which you can achieve a good price for the product, is even shorter, than you realize that each week by which you can shorten the go-to-market period is worth a lot of money.
In co-operation with the customer, we were able to establish a process accelerating the market start. And this was deemed an important quality argument by the customer.
Example 4: Completeness
Any survey will tell you that there is way more content created in the world than can be translated by all translators combined. Support websites form a prominent example. If enterprises sell their products globally, they need to provide support pages in all of the target languages. In most cases, that is much too costly. Thus, many enterprises focus their efforts on „important languages“ and important texts“. Important languages simply are the languages of the countries with the highest market values. On the other hand, important texts are determined by developers or established by means of factors valid in the homeland of the enterprise. This, however, does not mean that these texts are also important in other countries.
But this is the real issue: if a French person visits a French site and is directed by the first link to an English page, he or she will consider the quality substandard to put it nicely.
Obviously, you can push everything through MT. Unfortunately, its quality proves to be insufficient in many cases. What should you do?
By combining the two approaches, you gain a self-improving system. If you put aside a small portion of your budget for the bulk MT translation, you can use the rest for the human translation of important texts.
In this way, you obtain a complete translation (granted with different qualities). And you can improve the overall quality by utilizing usage statistics to direct the human translation to really important texts.
Now what is the quality of a translation?
Of course, the linguistic quality is of importance, but it needs to match the intended purpose. You need a different type of language for high-gloss marketing articles than for chat sequences.
And then there is the specific technical language. It not only consists of terminology, but also of wordings established in the expert communities. By using other wordings, you quickly prove yourself to be an outsider.
As we have seen, the quality of a translation may include time aspects (see example 3) and cost aspects (see example 4) casting doubt on the whole localization triangle.
None of the three corner points can be assessed on its own. They do not only depend on each other, but sometimes disguise themselves as one of the others.