3 Reasons ‘Post-Editing Machine Translation’ Is Not the Choice for You

We talked about machine translation in a previous blog post, but a more hybrid form of computer-assisted translation that some in the industry have been using is Post-Editing Machine Translation, or PEMT.

PEMT involves a complex process of defining standards and variables, utilizing specific inputs and outputs, and a whole lot of editing to produce a translation of a piece of text. But in the most basic sense, PEMT means a human runs content through a machine translation program and edits the result for grammar, errors, style, and fluency.

PEMT programs aren’t actually a new phenomenon; the first machine translation editing project took place way back in the late 1950s and early 1960s at the RAND Corporation to translate technical documents from Russian to English. However, most significant developments in machine translation technology have occurred much more recently.

Although this process sounds like an ideal situation and a wonderful use of often flawed computer translation products, there are a few caveats that come along with relying on machine translations. Here are three reasons why PEMT isn’t the right choice:

1.    Editing can be unproductive

Proponents of PEMT often point to the speed and cost benefits that machine translations bring, balanced out, they say, by the supposedly quick skills of professional editors. This might effective in some cases, but some experts have pointed out that running text through a machine translation program and then editing it line-by-line is unproductive and can even produce lower-quality of translation than what a human translator could have delivered.

This is because machines cannot process idioms, emotions, voice, and style in the same way that a human translator can. And even with the best programs, errors will always be prevalent. Some companies think they’ve found a loophole by hiring monolingual editors in the place of bilingual translators to do the heavy lifting faster. But the problem with this is that referencing the source text is unavoidable when working with machine translations.

Going line-by-line through text after machine translation takes almost as much or even more time than plain human translation, rendering the service irrelevant in the face of talented, and professional human translators.

2.    Pre-editing is often necessary

Another drawback to PEMT comes with making text simple enough for a machine to properly translate. This entails going through content before translating it to comb out any complicated grammatical structures, cultural references, or stylish writing to make sure it will produce an accurate translation.

Although this technique could be helpful when looking to translate the overall meaning of content, it is unacceptable to remove or alter the original content in such a way for more individualized projects. A literary translation submitted to pre-editing machine translation loses its voice, its quirkiness, and its writer’s flair.


3.    Quality isn’t concrete

Some machine translation experts reference the process as one that produces the results you put into it. No one expects to find the best or most beautiful translation using Google Translate or another popular online translation service, so companies that offer these kinds of services generally are forced to negotiate the output’s level of quality.

This level of quality determines the amount of time an editing team will work with your translation. Projects with a decidedly high quality will require much more time and effort on the part of the editing team than low-quality projects. In the end, low quality levels will not be publishable content. If you decide that PEMT is a better option than human translation, you have to be ready to sacrifice the quality and integrity of your content.

So why are some people going crazy for PEMT? It represents a major step in the development of artificial intelligence and computer-learning software. As translators feed input into a machine and correct the mistakes in the translation, computers are essentially learning how best to do this kind of project. For the time being, however, no computer program or machine translator can match the skill, style, and quality of a human translator. And nothing suggests that this will change for many, many years.

About the Author:

Daniel is based out of Chicago and works as a writer, editor, and translator.

Leave a Reply

Your email address will not be published. Required fields are marked *