Rob Vandenburg recently wrote a post over at Wired Cloudline on the challenge facing online business we’ve previously discussed. That is that much of the web is English, while many web users are not. If business wish to reach and effectively market to these new consumers then they must speak to them in their own language.
Vandenburg’s post, is well worth reading, as it effectively breaks down this issue, which is much larger than many people realize. The solution he offers is one which the industry has already adopted to a smaller extent. He says that the problem with machine translation now is that the translation memories(TM’s) used are so small, and cannot cover the broad spectrum of language they have to translate. Therefore as the size of TM’s available to machine translators grow, the quality will also grow. But, this will never fully replace the human translator as an editor of the machine translated materials, as humans will have to look out for colloquialisms and idioms.
This is generally how the industry operates now, though on a smaller scale. Translators use their own TM’s which are usually client specific, and this accelerates the translation process. TM’s do automatically translate some of the materials, but Vandenburg overestimates how much they can do. The heavy lifting is still done by the human, and even as the TM grows, the translator experiences diminishing marginal returns as context determines word choice. The context of the translated material is far more important than any idiomatic words, and takes far more time to translate. Not to mention the fact that each automatically translated phrase must be approved by a human.
The key thing that Vandenburg misses is that as a TM grows in the broad sense that Vandenburg is advocating, the quality of the automatic translation that it provides actually decreases. This is due to issues with synonyms and technical language that must be tailored for each industry or even client. If we were to throw all of these technical TM’s into on great TM we would receive more confused translations as the language would jump around between different disciplines.
While Vandenburg is right that increasing TM size for machine translators would improve the quality of the translations produced, he overestimates the extent to which this can replace human translators. Improvements in machine translation will at best augment the human translation process, but not to the degree Vandenburg says, where humans are at most minor editors.