Although I like machine translation for many reasons, I am concerned about the impact it might have on translation in general. Obviously, it is changing the landscape of the translation industry, and we translators would prefer not to have such a strong competitor. But I think there are serious drawbacks to overreliance on MT, which lead me to believe it will not just hit us translators, but will permanently cripple the industry. It is not just a matter of feelings, but common sense.
Note that this article is about overreliance on MT; that is, making it the cornerstone of the translation process, with human post-editors. Using it as a tool to increase productivity and minimize routine translation work is another story.
MT relies on humans to revise its output. Like any other automated technology, it requires human control. The problem is that whereas most automated technology works perfectly and humans simply oversee it without intervention, MT produces suboptimal results by definition and requires heavy post-editing. Do translators like doing this intervention? Some do, but many do not. People generally do not want to revise bad translations, whether from a human or machine. Cleaning up a poor translation takes a lot of time, but is often not compensated well. Instead of feeling the surge of creativity typical of doing one’s own translations, the editors often feel frustrated at having to deal with challenges that the translators should have not left there in the first place. I have yet to meet an editor who did not sometimes wish that he could have done the translation himself.
Prediction: Good translators will not edit machine translation output, because they have better things to do. Only less-qualified translators will post-edit machine translation. And those translations edited by less-qualified people will then be used to teach MT engines. Over time, this vicious cycle will destroy what’s left of translation quality.
Agencies will get hungrier.
We are already seeing translation agencies attempting to slash translators’ rates by 50%, just because they have inserted MT output into the bilingual files—all while expecting the same quality as when the translators were paid 100%. Are you willing to accept that? I definitely am not. So who is going to work for those agencies? Right, less-qualified translators. More bad translators on the market means lower quality in the long run. In fact, it is already happening. Some translation agencies, even major ones, that used to be focused on quality, are now sending out mass emails with rock-bottom rates, hoping that among hundreds of recipients, there will be someone who is desperate for money at the moment.
MT produces robot-like style, and this is what end users will have to put up with in the years to come. First of all, post-editors will not be editing style much, since this is not their job—they already have enough work finding critical errors. Second, the more editors work with MT, the more they get used to that robotic style. The editor will eventually think it is okay for a text to read like this. Among the three predictions I make in this article, this is the one that I am most sure about, because we are starting to see evidence of almost daily. It is a rare post-edited machine-translated text that is even close to the readability of a translation by a good translator.
Perhaps, things are going to change, and the next generation of translators will be more compatible with MT. But I doubt it, and think that larger agencies, translation buyers, and other organizations who want to make MT the cornerstone of the translation process are actually pushing us all into a death spiral. MT is a good servant, another useful tool in a translator’s toolkit, but it is not the Holy Grail. Automation is only good when what it produces is equal or better than what humans produce, which is not the case with translation at the moment.
Read more about why translators are not fond of MT in the article “Machine Translation Is a Lazy Partner.”