As a topic, machine translation is a lightning rod for controversy. Apparently, machine translation delights some people, especially LSPs (Language Service Providers)—and infuriates others, namely translators. Look at the comments on any translators’ forum and you’ll find this accusation: that translators who take on post-editing jobs are “unprofessional.” Here are my two cents on the subject.
Translators are Angry
I understand why many translators are angry. After all, translators fear that machines will replace them. In fact, most LSPs do indeed see machine translation and post-editing as the best way to maximize productivity and cut translation costs. To translators, though, it looks like LSPs don’t need them.
So, is it true? Are LSPs positioned to replace translators with machines?
Not exactly. Because most LSPs don’t necessarily have the capital to invest in professional-caliber machine translation—effective, customized machine translation, that is. These days, the vast majority of LSPs are settling for what’s called MTM—a combination of Machine translation and Translation Memory. But MTM, which is comparatively low-cost, is not a customized translation tool; at best, MTM offers mere suggestions—translation options only. And that’s assuming you have a clean translation memory and established terminology.
In worst-case scenarios, LSPs simply toss the source text into a free machine-translation system like Google Translate; or they use a commercial do-it-yourself solution. Then they send the output to a translator for so-called post-editing. Just like that. No terminology work, no pre-editing of the source text, no corpora preparation, no data cleaning—nothing. It’s no wonder that many linguists express rage against the machine. Who wouldn’t be infuriated by being asked to clean up a mess made of words?—and being paid next to nothing for their trouble?
Don’t Hate Machine Translation
Informed practitioners know that machine translation isn’t a 1-2-3 proposition. In the first place, usually machine translation only comes into play when huge volumes of data— typically hundreds of thousands of words —need to be translated quickly and repeatedly. And implementing customized machine-translation systems requires time as well as, most importantly, a serious financial investment. The choice of a machine translation system depends on many parameters, including the language pairs involved and volume of data to be translated, for example.
Machine translation is not about one-size-fits-all, but about a complex, costly process. LSPs that rely exclusively on off-the-shelf machine-translation engines for faster, cheaper translation might scrape by once or twice. But, in the long run, they’re setting themselves and their clients up for failure.
The Truth About Machine Translation
Likewise, post-editing a machine translation doesn’t simply mean a quick, low-cost look at a text for mistakes. Before post-editing can even begin, the output must be quality controlled. To assist post-editors, skilled LSPs create guidelines based on client expectations; the target audience; the purpose of the text; and many other factors.
Translators can benefit from an unbiased approach to machine translation. Those who take on post-editing assignments need to grasp what machine translation really is. They must differentiate between the various machine-translation systems and understand the errors that can occur. Mastery of the software is vital for translators, as is knowing how to clean up data with, for example, filters, macros and regex—before and after machine translation comes into play.
Fairness for Post-editors
Before the assignment even crosses the post-editor’s desk, the task needs to be defined: the parameters set, the client’s needs and expectations outlined, the intended use of the text—as perishable or publishable—explained, the machine translation output needs to be evaluated. And, clearly, translators who take on post-editing assignments must be paid a fair rate for the specialized work they perform.
Tweaking the Machine—Together
What’s needed is cooperation between direct clients, LSPs, computational linguists, and post-editors. The direct client and the LSP must set realistic goals, drawing on the expertise of computational linguists for the preparation of corpora, the development of a controlled language, and the configuration of the machine-translation engines. Only then can post-editors get on with the crucial work of correcting machine outputs and improving texts.