Posts from the ‘machine translation’ Category

My View on the Reliability of MT

I feel that I owe some explanations after my first article about machine translation (MT). In my previous post, I stated that machine translation can be a very useful tool, but it cannot and should not replace the human translator. I will very briefly try to explain why, after which I promise to drop the topic for good .

The first argument against machine translation (and I would include all engines here, not only Google and Bing) is creativity (or the lack thereof). Everyone agrees that humans are creative creatures, they can create stories, objects, works of arts, to name only a few. And believe it or not, language is probably the greatest creative achievement of the human mind. But not in the sense of a finite object, because language is not something ”made”, it’s something ”in the making” (an activity). This fact lies at the heart of Humboldt’s theory of language and is acknowledged by all modern paradigms of Linguistics. According to this principle every speech act is new even though it is indeed based on our historic linguistic experience. But the way in which we use our linguistic knowledge to create new meaning is completely different from the way a machine uses the corpora of text to ”create” new translations. Humans create new speech acts (in essence, a translation is a speech act) based on meaning and sense, they dress them up with words, which means that the same idea can be expressed in very many ways, using completely different words. A  computer, on the other hand, will always use the same sentence to express a certain idea. And this brings us to the second part of my post.

The second argument against MT is meaning. Every translation course is based on the assumption that the students/trainees attending the course are fluent in at least two languages – their mother tongue and a source language. Clearly, in order for anyone to translate, they have to speak and understand the two languages. In other words, when they read and attempt to translate a text, they will be required to understand it. Not only do they have to make out the meaning of individual words, but also grasp the sense of the whole text. Mind you, it has been proven that the best approach to translation is a text-linguistic approach. So I am going to ask you the following question: Is a computer capable of grasping meaning? Can it understand a text? I’m afraid the answer to this question is no. A translation engine only recognizes words and sentences and uses a statistical model to compare them with parallel structures in the target language corpus in order to come up with the best solution. Wait a minute… Did I say it recognizes words? WRONG! It recognizes signifiers (signifiants) which are only graphic representations. It doesn’t even bother to go beyond signifiers and discover the actual meaning of these representations simply because it can’t! A computer does not speak any language. A computer does not understand meaning. A computer does not read texts, let alone understand them.

It’s true, MT has evolved a lot in the last couple of years and the statistic approach it uses is to a certain extent functional. But the question is, is this really enough? I mean if MT is so great, why does it need post editing? Why does it need human intervention? Yes, I know, some will argue that MT helps increase productivity. But is this all that matters? Does everything come down to money and quantity?…

Maybe a day will come when a computer will be able to fully understand us and yes, to translate flawlessly, but that day, my friends, will be the day computers will fully replace humans in all their activities…

It’s Us Against the Machine?!

One of the hot topics at the FIT Europe Seminar in Bucharest this weekend was the so-called threat of automated translation to translation professionals and to the translation industry in general.

The conclusion of the FIT Europe Seminar panel discussion was that  machine translation, although a fast-growing technology, does not pose a real threat to the translation profession in terms of quality, in the sense that it will not make the translation profession extinct. And this is because it will never be able to replace human professionals and translations made by machines will never be as good as human translations. This does not mean to say that automated translation has no place in today’s society. On the contrary, it is (and should be) used for different purposes than human translations. Its main role is to translate very large amounts of text within very short periods of time making content ”accessible” to various target readers. Translations produced by the machine are not supposed to be 100% correct and are not meant for publication. They are useful whenever someone wants to grasp the general idea of a certain text (or corpus of texts). So yes, machine translation is very useful and necessary sometimes.

So far, there is nothing wrong with the machine. Human translators and translation engines can co-exist. Nonetheless, if attempts are made to shift today’s translation business model by including machine translation in the process, as a recent TAUS analysis suggested (see here), things can get nastier. First of all, according to the same analysis, translations will still require post-editing from humans since their quality is far from perfect, so there will be no such thing as ”automatic translations” – they will be at best ”semi-automatic”. You can see the threat, right? In such a scenario, Internet giants (such as Google, Microsoft) and organizations promoting machine translations (such as TAUS) will become the leaders of the industry. Mind you, they will still require translators (which they will elegantly label as post-editors) because, let’s face it, no-one will risk publishing a faulty translation produced by a machine. But these ”new” professionals will have a much reduced significance from the perspective of the automated translation provider and will be paid less…

Yes, you might think that the inclusion of machine translation in the process will make translation much speedier. But if you look at it objectively, machine translation will not fully replace human translators, they will still be required to intervene and it is not clear how much it will take ”post-editors” to correct the errors made by the machine. I’m guessing it will take pretty long, because on the one hand, reading (carefully!) such large texts takes a lot of time, and on the other hand, certain sentences or paragraphs will require retranslation. So not much is gained in terms of time. The only real gain here is money. If this business model is applied, the costs will be considerably lower, and this is because, as mentioned before, translation professionals will be paid a lot less for their editing and proofreading efforts (that is of course if the true translation professionals do agree to adopt this model…).

So what should translation professionals and translation organizations do? It was agreed during the panel discussion that it is important to explain to the business environment the differences between human translators and translation engines (when it is good to rely on human translators and when on machine translations).  The key word to this entire debate is transparency. We should always be transparent about various translation technologies. And, more importantly, we should ask machine translation providers to be transparent about their practices and beliefs.

I will not list the reasons why I think machine translation will not be able to translate properly because it would make my post way too long and there are a lot of articles on the Internet about that .

*Photo by sheelamohan