Machine Translation

How to Collaborate with a Machine: a Translator’s Story

Post Editing

Translation as a profession is changing in the global information age. The development of language technologies (i.e., translation memory systems, content management systems, and machine translation systems) has led to cheaper and faster translations, making it possible for translators to produce more in less time. The development of tools like Google Translate have in a way “democratized” translation, making it possible for anyone to insert a segment of text in various source languages and (at the very least) “get the gist” of it.

Machines will not Replace Humans

The dilemma of the Internet age has involved the translation industry and many have been left wondering whether machines will replace human translators in the long run. Well, lucky for us translators, we have not been replaced by machines because we can do what machines cannot. Human translators not only translate words but they ensure that the message resonates with the target audience, and are able to recreate the same type of communicative effect as that intended in the source language. However, human translators have been required to collaborate with machines more and more to accomplish their translation tasks. Translators are often required to become machine post-editors, in an effort to save time and money.

The technical term, machine translation post-editing (MTPE), refers to the process of reviewing and editing the content that was previously translated by a machine. This means that, after an initial stage of automatic translation done by a machine, the human translator will add his subject matter expertise and knowledge of the target language’s cultural context to review the document for translation accuracy. But how accurate is a text translated by a machine, really? And how much editing is really required?

Full Post-editing and Light Post-editing

Language professionals distinguish between Full PE and Light PE, the former producing a text of publishable quality, the latter, one of merely understandable quality. While the goal of light post-editing is to make the text understandable and adhere to some specific requirements from the client (i.e., that some names are left untranslated, or always capitalized), the goal of full post-editing is to make the translation stylistically good and terminologically accurate.

Now, some types of text are a natural fit for MTPE. A machine is more likely to give a better translation of technical and scientific documents with limited vocabulary and few polysemic words (aka words that have more than one meaning) than literary works. Simply put, a machine is more likely to choose the right words in a scientific text (where there is only one way of expressing or naming a concept) than in a novel (where there are multiple variations of expressing the same things and various stylistic elements).

While not every piece of machine translation is inaccurate, how much time does a translator really put into doing MTPE? Is taking a machine translated text to start with really faster (and better) than translating from scratch?

MTPE does not Guarantee Less Work for the Translator

Unfortunately, for the translator working with a text that has been translated by a machine does not mean less work in producing the final output. Often times, the machine translation can contain very serious mistranslations. Google Translate, for example, very often omits words (especially negations) because it looks for similar sentence patterns in its translation memory rather than translating from scratch.

The job of a post-editor is, of course, to catch all the mistakes by the machine. But often times the job done by a post-editor involves so many changes and rewriting that it would have been less time-consuming (and cost effective in term of hourly rates) to translate the document from scratch.

This article is written by a professional writer, Ilaria Ghelardoni, associated with Ulatus.  

Share your thoughts