Tue 22 October 2024 (verändert am Thu 24 October 2024) Translation, ai translation, machine translation, mtpe AI Translation, Human Translation, Machine Translation, MTPE, Professional Translator, Quality Control, Quality Translation, Technical Translation, Translation Quality
“We would greatly appreciate your expertise in conducting a thorough review. [...] Although the content has already been proofread by another translator, we have some quality concerns”
These were the words of a new client who approached me to review a highly technical translation in the field of high-performance metal drills. Their quality control tools had flagged numerous formal errors, prompting them to seek out another specialized translator for a comprehensive evaluation. The goal was to ensure the highest quality for their final client.
I suggested an initial two-hour proofreading session to assess how much the translation would need to be revised.
It didn't take long to confirm the client’s concerns. The translation appeared to be the product of machine or AI translation, given the recurring, tell-tale errors:
During the first two hours, I meticulously annotated and explained each of my changes to demonstrate that my corrections were necessary, not merely ‘preferential’. After reviewing my initial feedback, the client gave me the green light to revise the entire document. It ultimately took a bit over 10 hours to thoroughly edit the 72-page document.
That wasn't the first time I'd encountered a translation that bore the clear marks of machine translation. In one particularly extreme case, the translation seemed to be nothing more than raw machine output pasted into a CAT tool without a single check. The errors were so numerous and glaring that, after just an hour of proofreading, I informed the project manager of the situation. She promptly decided to have the translation redone by a more conscientious translator.
I’ve had this discussion several times in translators' forums: Some professional translators are convinced that machine translation can already produce flawless results and might even replace human translators. This belief has led some, whether unscrupulous or naive, to think that they can easily earn money by relying on machine translation while investing minimal own effort.
It’s true that well-trained engines can produce excellent translations —but only when they’ve been trained on billions of similar texts. I experienced personally only one case where reviewing MT output felt like proofreading the work of an excellent human translator, with hardly any corrections needed. The text, in that case, was a standard set of terms and conditions for a website —a type of text that has been translated countless times, indeed.
However, most small to medium-sized companies are unlikely to produce enough material to train a specialized translation engine for their niche content anytime soon.
The case described above reinforces my perspective: Machine translation still falls short of perfection. Its grammatically correct and fluid output can be deceptively convincing, even for experienced translators, leading to a cursory review rather than a critical one.
While machine translation can be a valuable tool for improving productivity and assisting in quality control, it still requires expert oversight. Only a skilled translator can identify the subtle yet critical errors that appear plausible at first glance. In my experience, the careful scrutiny required significantly tempers the productivity gains promised by MT —at least for those of us who are committed to delivering accurate and reliable translations that meet the client's needs.
In this case, overestimating the capabilities of MT and failing to perform a critical review had costly consequences for my client. I estimate, based of my exchanges with the project manager, that more than 30 additional hours of work, including my involvement, were needed to bring the target document up to an acceptable standard. Not to mention the loss of credibility and inconvenience to the end client, due to the delivery delay.