Automated Translation (part IV and last)

I was fortunate enough to attend a demonstration of Portage, the statistical automatic translation project led by the National Research Council of Canada, and I was genuinely surprised by what I saw. Even if the translations weren’t directly publishable, they were a long way from the pitiful results arrived at a few years ago.

However, this initiative is not quite representative of what an automatic translation program can do, since it benefitted from highly favourable conditions.

In the previous post, we examined the criteria for producing a high-quality automated translation. Let’s now look at how things work in less ideal situations.

Volume and matching of available references (corpus)
To achieve its outstanding results, the Portage development team had access to over 100 million words from debates (English-French) of the Parliament of Canada. To give you an idea of the volume of text that this represents, imagine a 1,600-metre high (six storey) pile of paperbacks! This much text would require 200 years of work by an experienced translator, without a single vacation…

It is practically impossible to have this kind of volume in a specialized field. Even the Government of Canada, the leading user of translations in Canada, rarely has such a high quantity of text in a specific sector.

Quality of the corpus
Parliamentary debates in Canada are translated by hand-picked translators who go through an exceptional revision process. Their translations are submitted to very thorough and meticulous review. Unfortunately, in other situations, it’s a whole other story.

Putting together such a gigantic corpus requires the use of translations from a variety of sources, leading to problems with uniformity and suitability for the documents to translate. In addition, texts are all too often inputted directly into the automated translation program without undergoing quality control or revision.

Quality of the text to translate
Given how important their texts are to their credibility and effectiveness, organizations should call upon professional writers. Unfortunately, this is not always the case. As a result, organizational texts are sometimes rough and ambiguous, and they fail to reflect precisely what the writer was trying to say. These are problems that automated translation systems struggle with—unlike professional human translators.

Some may find it tempting to use automatic translation to eliminate a human translator’s ‘weaknesses.’ After all, a machine is faster, never stops, and appears to be more economical. Its children never get sick, and it never gets the winter blues. But as we have seen, in spite of its qualities, it is still far from being able to replace a man or woman.

Of course, automatic translation might reach this goal someday. But it’s important to keep in mind that translation is communication. Do we really want communication between humans, with all its nuance and its importance for mutual understanding, to be handled by machines?

Automatic translation can certainly be a very good tool, but only in the expert hands of translators.

(Translated by Joachim Lépine, C. Tr)


Laisser un commentaire

Entrez vos coordonnées ci-dessous ou cliquez sur une icône pour vous connecter:


Vous commentez à l'aide de votre compte Déconnexion /  Changer )

Photo Google+

Vous commentez à l'aide de votre compte Google+. Déconnexion /  Changer )

Image Twitter

Vous commentez à l'aide de votre compte Twitter. Déconnexion /  Changer )

Photo Facebook

Vous commentez à l'aide de votre compte Facebook. Déconnexion /  Changer )

Connexion à %s