Tuesday, November 1, 2016

LSP Perspective - This terrible word “MT”

This is another in a series of posts about LSP perspectives on MT. I may not always share the opinions of the guest writers, but I am a proponent of sharing differing views, and letting readers decide for themselves what makes the most sense to them. I do not edit these opinions except sometimes for minor typos and basic reading flow related edits. I may sometimes highlight some statements in bold to highlight something I think is central to the view. I invite other LSPs who might want to use this forum to share a personal opinion as long as it meets some basic quality requirements. I have left the original voice intact and have not edited it anyway.

Unrelated to this post but worth a mention, I also saw an unfortunate and grossly ignorant, of how MT works, post by @dstaechelin which appears to have been taken down now. But it was interesting because it was representative of the most ignorant views on MT that we in the industry encounter with very superficial and facile criticism of how MT can be wrong. For those who have looked at MT for 15 minutes or more, it is clear that it is possible to trip MT up very easily, in the same way, that it is very easy to trip up a mentally handicapped human and make him look foolish and retarded. Possibly they took down the post since I challenged the quality of the observation, or maybe they realized how stupid the observations were especially for a company that called itself Language Intelligence. So now all we have left is the Google search residue shown below:



Image result for technolex translations

МТ (machine translation) in its today’s form is a rather mixed and controversial phenomenon. Numerous misunderstandings and disappointments related to it are explained by the fact that both translators and those who use their work results have adopted a wrong attitude towards MT.

Let’s look at what MT is for the translators who use it in their translations and for ordinary users who read such translations.


Machine, Carl!


“Ordinary” pedestrians do not see any difference between gasoline and diesel engines: the only thing they care about is that car with such engine is able to move. They are not interested why it happens. The same is true for ordinary users who do not see any difference between automatic and automated translation: they see only the text that can be read. Also, they are not interested in how exactly the text was obtained.

For ordinary users Trados and PROMT are as alike as two peas in a pod. Those two software applications are listed under the same section “Translation software” in the catalogs as their end result is a translated text. No attention is paid to the fact that text from Trados is translated by humans while the text in PROMT is translated by computer. That’s why MT is mostly perceived just as a specific type of usual text. Ordinary users do not notice the word “machine” in the phrase “machine translation”.

This raises misunderstandings: users get a wrong expectation that MT text will be just as any translated text. The software application is called “translation software” so it should produce translation. It does not matter how you call it — machine or computer-aided translation — the main thing is that words from foreign language are replaced with the ones from the native language, i.e. they are written in “human” language.

But the point is that MT is produced not by humans but by machine, which (as yet) operates with words, but not with sense. Even those who realize it are tempted to act according to the scheme “Now I’ll change grammar endings quickly and I’m done!”. However, disappointment follows: the text after MT is not “human” (and sometimes even inhuman) and often can’t be understood at all. Some funny machine translations from English into Russian (such as translation of MS mouse and mouse driver software help where computer mouse was translated as an animal and driver was translated as a herdsman) have long become Internet memes. There are many other examples of blunders in the manner of Google Translate.


No pasaran!


The professional translators who are “on the other side of the barricade” and especially those who translate into Cyrillic languages know very well: PROMPT work product cannot be considered a translation but rather a word-for-word translation. This is a text obtained by a literal replacement of words in one language with the words in another language. To think that such text is a translation is the same as to think that a stack of bricks is a house.

As a rule, translators meet with outright hostility the customers’ attempts to reduce the cost of TEP (translation, editing, proofreading) by replacing translation with MT. Not only because they perceive MT as their competitor (although this also is a case: “How come? I’ve put so much effort in mastering this highly skilled job and now you tell me that a dumb machine can do it!”). The main reason is that in practice MT, contrary to expectation, neither simplifies translation nor reduces its cost.

Strangely enough, getting the word-for-word translation “into shape”, i.e. turning it into a full-fledged, “classic” translation, takes not less, but often much more time than translation “from scratch”. No time is won by replacing the “foreign language —> translation” process with the “word-for-word translation —> translation” process.

Moreover, MT not only fails to speed up the translation process but also reduces the translation quality. Many translators complain that MT “hog-ties” them and “pulls the wool over their eyes”. The words replaced by computer automatically allow for no creating thinking, limit imagination, which is essential for high-quality translation, making the translation “heavy”. The text translated using MT becomes insipid and loses the “spark” that attracts the reader.

Therefore, machine translation is not a translation. That’s what caused all the fuss — this small terminology confusion.

As the result, the phrase “translated using MT” is perceived as oxymoron: either translation, or MT – there isn’t another choice.


MT, or not MT, that is the question!


However, there is another choice! And even a couple of other choices.

Yes, MT is not a translation and cannot be perceived as what we are used to call translation. But it does not claim this “title”.

First, in some translated language pairs — for example, in Russian-Ukrainian — machine translation in its today’s state is getting closer to “classic” translation quality. Surely, there are many funny blunders (such as “Robin Hood shot from onion”, as it is one of the translations for the Russian word “bow” in Ukrainian). However, in general, getting into shape the MT-translated text, especially a technical one, really does not take much time.

Second, MT allows saving much time while solving another task: when there are huge amounts of text impossible for human translation while the translation quality requirements are moderate, i.e. when word-for-word translation quality is enough. Computer does not sleep, never gets tired and does not go on vacation. When quantity is more important than quality, computer is second to none.
Third, MT makes it possible to get general idea of the text in a foreign language. Try to understand what is in the documents you are holding, if they are written, for example, in Swahili.

Five-ton truck is a great means for transporting cargoes that weigh 5 tons. To load it with 10 tons or, on the contrary, carry a sack of potatoes in it, means to use it for purposes other than that intended. The truck in itself is just a tool.

The same is true for MT: machine translation in itself is neither good nor bad. The question is not “MT, or not MT”, but to determine the tasks to apply it for and the tasks that are beyond its capacity.
Therefore, to use MT, or not to use MT, depends on what you are trying to achieve. Confusion arises when somebody attempts to use MT for solving tasks it is not intended for.


Excuse me for interrupting you…


… but MT advances along several fronts so translators will hardly be able to ignore it. But at the same time MT will hardly replace qualified translators and editors in the near future. Language is a very complicated thing.

Attempts will be made more often to apply MT in realms where machine has never been strong. This is normal and inevitable.

For us, translators, the conclusion is simple: don’t be afraid of MT. It can really squeeze some translators from the market but only those with poor quality of translation. But we need to go on with high-quality translations and adopt new techniques of work. What if a time will come when computers will translate words into sense?

The Author

Vladislav Demyanov
Project Manager, Technolex

Software Partner: Protemos LLC (

© Copyright by Technolex Translation Studio:


  1. Hi Kirti,
    when you refer to PROMPT, do you mean PROMT ( Or did you deliberately choose the wrong spelling?

    1. As is stated quite clearly above, this is a guest post written by Vladislav Demyanov and I suggest that you can contact him if you want to know the answer to this. But it is pretty clear that he meant the MT company in Russia

    2. PROMT was meant, of course, not PROMPT :)

  2. I respectfully disagree with most of the article, especially the idea that MT does a word-for-for translation (and that contradicts the statement that MT is not a "translation). MT has come a long way in the last couple of years. However, I do agree with the point made about MT not being a magic solution for every scenario. MT, as any other tool, process or product, has its own use cases.

    1. This is only a blatant manifest of a very narrow experience with MT. "Word-for-word translation" is exactly what people could draft from RbMT, especially at first sight.
      I agree with Kirti that these views deserve some room in spaces usually devoted to MT by MT advocates and experts, just as some room should be left in events and blogs that demonize MT to the views of MT specialists.
      Unfortunately, to paraprhase a comment in a blog notoriously adverse to MT, when you are angry and insicure, and, because of this, scared to death, censorship is usually the last resort of those with something to hide. Ignorance and fear, in this case.
      In this perspective, Mr. Demyanov's post is commendable.

  3. Nicole Chardenet

    Nicole Chardenet Good points, although I suspect HTs will be dealing with MT and post-editing a lot more in the future as the creation of content and the need for translation will grow exponentially. It'll just be too difficult for existing and even future HTs to keep up....with 2.5 *quintillion* bytes of data created every day. That's two and a half exabytes *every single day*.

    That's a lot of data, and a lot of coffee brewing for the HT team...