Olympic Targets

by Jost Zetzsche

ight years and two Olympic Games ago, I wrote an article titled " Tapping Into the Olympic Spirit: The Pillars of Translation." Inspired by the classical columns of the Greek Olympic hosts, I tried to narrow down the four pillars of translation that give us the required balance in our everyday lives as translators.

Here's what I identified:

  • Profound knowledge of the grammatical rules that govern the source and target language(s)
  • Lexical knowledge of source and target language(s) and the complicated relationships between the two lexicons
  • Ongoing, practical interaction with the source and target language(s) and the ability to understand contextually (I called that "living knowledge of language" back then)
  • Knowledge and hands-on experience with the tools of the trade, which included the many computerized resources such as computer-based reference materials and computer-assisted translation tools.

Examining these four pillars, I concluded that what distinguishes us from machine translation systems is the third criterion—machine translation systems might know language rules, lexicons, and technology, but since they don't truly understand language in context, they often fail us.

(And before you get too upset, I also mentioned "other supporting pillars, such as marketing, client education, and the ability to work on a team.")

Now that the world's best Olympic athletes are assembling again, this time in London, has anything changed?

The short answer is "yes, to some degree." The "pillars" that I listed remain, but there is a different emphasis on the final one. And I also would draw a slightly different conclusion when it comes to machine translation.

The Changing Fourth Pillar: Our Knowledge of Tools of the Trade

First of all, I think that we can give ourselves a little pat on the back for having come a long way with translation technology in these last eight years. While technology for the professional translator did not develop as fast as it could have (too much effort had to be spent to convince us of its usefulness), our appreciation and application of it has. Without any hard numbers to prove it, I know that among professional technical, medical, legal, and other functional translators the actual employment of translation environment or computer-aided translation tools has increased significantly. Many of us might have owned a copy of Trados or one of its many competitors in 2004, but as far as really understanding how to use it efficiently for every job that came through our office doors or email inboxes—that was a different matter altogether. Today, while we are still far from full and perfect implementation, we've come a lot closer.

The same is true with online resources. Many of us not only use simple online dictionaries, we also access much more complex corpora. Some of us are even building our own corpora (or translation memories) to support our translation process.

One area that unfortunately has not changed when it comes to technology is the sparse use of our translation environment tool's terminology feature. Many of us still aren't willing to invest the time to learn how to use these tools adequately or build up our own terminology resources. Change here might come indirectly through the increased use of subsegment leveraging (I've written about this often, including right here).

Machine Translation on the Move

And why would I draw different conclusions in regard to machine translation today?

Just the week before this article was written, Yahoo! Babelfish was officially put out to pasture. That's the same Babelfish that put machine translation—for better or worse—at everyone's fingertips and paved the way for all other online machine translation engines. It's not that you can't access the same kind of results anymore. These were always provided courtesy of Systran, and you can find the same engine on their website. But aside from a certain historical relevance surrounding the retirement, what's most interesting is that it was replaced with Microsoft Bing Translator. This means that all large search engines now have machine translation features that are based on statistical machine translation (SMT) engines rather than rules-based systems (you can find definitions and links on the different approaches here). Google uses Google Translate, Bing and Yahoo! use Bing Translator, and the leading Chinese search engine Baidu and the Russian leader Yandex use their own proprietary SMT engines (Yandex only for its most important languages of Russian, Ukrainian, and English).

So, my 2004 conclusion that machine translation lacked a "living knowledge of language" could easily be disputed today: statistical machine translation is based on actual translated texts rather than mere rules like its rules-based sibling. Does that actually make it better? It depends. The large online systems might perform better "out of the box" than their rules-based cousins for some texts; on the other hand, the rules-based systems are generally superior when they're trained for a specific subject matter (see this article in the Translation Journal).

What's probably more relevant to us when it comes to machine translation is that more and more translators are actually using it for a first "dirty" translation pass if no translation memory hit is found. How many? Certainly more than in 2004! Look at the fact that machine translation has become a staple in virtually all translation environment tools, a state of affairs that was almost unheard of back then.

To switch our metaphor now to one of the Olympic sports, it's safe to say (and is becoming increasingly politically correct) that alongside the arrows of translation memory leveraging, terminology management, quality assurance, and file management, machine translation is a weapon that is increasingly finding its place in the professional translator's quiver.

It's also good to know, though, that some things never change: Just like eight years ago, none of those arrows in our quiver will hit the target without our guiding and steady hands.