In this two-part article we explore why translating academic texts is easier than proofreading them
As an author who has needed an article translated, maybe the following thought has occurred to you: “Wouldn’t it save time and make things easier for everyone if I ran my article through Google Translate first and then only need to have it proofread?”
On the face of it, there seem to be quite a few benefits:
- Proofreading is generally faster than translating, so if you’ve got a tight deadline it could give you an extra day or two to polish up your article.
- Proofreading is generally less expensive than translating, so if you’re on a tight budget then proofreading might sound more appealing than translating the whole article.
- With technology as advanced as it is today, why not make use of it?
- Everyone’s time is valuable, so why waste the translator’s time by making them translate an entire article that they could just proofread if you give them a text that has already been machine-translated?
These points are all valid to a certain extent, but in the following two-part post we’re going to explore why so-called “machine translation” (i.e. Google Translate & Co) makes absolutely no sense for academic translation and can actually end up being a lot more work for everyone involved. In the first part, we will show how Google Translate works since it is the most popular machine-translation tool and then look at one specific problem that arises for academic texts in regard to vocabulary. In the second part of the post, we will then discuss sentence structure and how machine translation can render a text so unclear as to make it nearly impossible to proofread accurately.
All Roads Lead Through English
First let’s discuss one term that sheds light onto the difference between proofreading a text translated by a human translator and one translated by a machine. In the translation business, if you revise a text that was translated by a machine, you are not “proofreading.” Instead, you are “post-editing.” This should be an indication to laypersons that the essence of what is done is different enough to deserve a different name. Proofreading often involves looking for typographical errors, consistency in terminology, or stylistic concerns. Post-editing, in contrast, involves making many inferences about what the source text might have intended when the text to be edited is ambiguous or completely unclear.
To understand why the text that results from a machine translation can be more ambiguous or unclear than a human translation, we will look at one key point of how at least the most common machine translation tool works. This will demonstrate why it may work wonderfully for a translation of that e-mail you need to send to a British or French colleague about what hotel to book for your next conference but fails miserably when it comes to your journal article.
When you put a text into Google Translate, it uses statistical machine translation to translate your text, meaning that it looks at all the translated documents available to it and finds the best matches. That sounds nice in theory, and in simple sentences it works fairly well, particularly when English is one of the languages. Let’s take a look at what happens when we take a fairly simple German e-mail such as described above and have it translated into English and French.
ich schaue gerade nach Hotels in Amsterdam. Hast du schon ein Zimmer reserviert? Das Konferenzhotel ist mir zu teuer – es wäre schön, wenn wir im selben Hotel wären, dann könnten wir zusammen hinfahren.
I just look for hotels in Amsterdam. Have you made a reservation? The conference hotel is too expensive – it would be nice if we were in the same hotel, we could go together.
Je regarde juste pour hôtels à Amsterdam. Avez-vous fait une réservation? Cet hôtel de conférence est trop cher – ce serait bien si nous étions dans le même hôtel, nous pourrions aller ensemble.
The English is pretty good. The first sentence uses the simple present instead of the present progressive or present perfect progressive (“I just look for…” should be either “I’m looking for…” or “I’ve been looking for…”), but the meaning is clear and the British colleague would understand the e-mail with no problems.
Now let’s look at the same German e-mail translated into French. The main problem is in the second sentence. The German reads: “Hast du schon ein Zimmer reserviert?” (“Have you [informal] already reserved a room?”) The French, though uses the formal “Avez-vous”, reserved for use between people who are not on a first-name basis. Why would Google Translate have done this? The reason is simple: when translating between two non-English languages, Google Translate first translates a text into English and then into the other foreign language (German -> English -> French). Because the English “you” is ambiguous and can be either “tu” or “vous,” Google needs to make a decision, and here it chose incorrectly. Imagine that on a more abstract scale. An unambiguous German word is changed into an ambiguous English one, and then that English one is translated into the word in your target language (in this case French). The potential for incorrect choices abound. Even when staying within a pair that includes English, however, decisions on sentence structure and vocabulary need to be made constantly by the machine. Now we will turn to one language pair, German to English, and look at the problems that arise when working with academic texts.
Problems Increase with Academic Texts: Vocabulary
A second point on how Google Translate works is that it bases its choices on statistical frequency, whereas many academic texts distinguish themselves by a particular jargon that is specific to the discipline but not common in regular speech or writing. Just taking one word as an example, let us look at the German “Gestalt” – in a casual sentence, “Gestalt” often means “figure” or “shape,” and Google Translate correctly translates the word in the sentence “Hast du die Gestalt gesehen?” as “figure.” But in the field of psychology, “Gestalt” is a specific term that is left untranslated in English (“gestalt psychology”, for example). When using Google Translate, the sentence:
“Es werden bevorzugt Gestalten wahrgenommen, die in einer einprägsamen (Prägnanztendenz) und einfachen Struktur (= „Gute Gestalt“) resultieren.“
“Preference is given to figures which result in a memorable (preeminence tendency) and simple structure (“good figure”).”
If a proofreader is given this English sentence to review, from the context it is possible they might wonder if “figure” is trying to refer to “Gestalt,” although that depends on how well they know the subject area and whether or not they know the German “Gestalt” is typically translated as “figure” in English. It may be, however, that they don’t notice that a specific term is missing or assume that the author intentionally used “figure” instead of “gestalt.” This could result in missed terms that might lead to a reviewer criticizing the researcher for apparently not being familiar enough with the field to use the correct terminology. If the researcher would have requested the original be translated by a professional translator experienced in psychology, however, there would be almost no chance of “Gestalt” being improperly translated. The problems only arise when a proofreader is given a text that was poorly translated in the first place, rendering the terminology unclear or unrecognizable.
This first part has hopefully given the reader an idea of how machine translation can work and why difficulties can arise especially when translating between two non-English languages. In addition, we described why machine translation that relies on statistical frequency of translated words can be particularly unsuitable for academic texts, where terms are often used with different meanings than in everyday speech. In the second part of the post, we will analyze sentence structures that come out of Google Translate to see how a post-edited text can differ in meaning from what was intended in the original.
 Gestaltpsychologie, Wikipedia article. Last accessed: 24 November 2016. https://de.wikipedia.org/wiki/Gestaltpsychologie
If you found this article useful, here’s how you can share it with your colleagues: