Yep. The time has come to reap the savings offered by technology, but it’s still critical to have competent, professional human linguists involved after the robot generates its product. And that means in many circumstances that those savings just aren’t as dramatic as we might hope.
The internet guys have been telling us since the last century that, in the very near future, you can have a robot translate documents for you in a tiny fraction of the time, and at a tiny fraction of the cost, of traditional human translation. I’ve been beating my head against a wall for well over a decade, trying to convince bargain-hunting clients that, no, you can’t trust machine translation. That’s still true to a point, especially if the machine’s name starts with a G and rhymes with poodle.* See, machines can’t format documents. They can’t sufficiently intuit the sense of complex legal terminology. And they can’t exercise judgment when multiple correct answers are possible but some are superior.
Still, my translation providers have finally reached a level of confidence in their dictionaries** to offer machine translation, but…
- The big caveat is this: it’s absolutely critical to have a skilled human editor go through the machine translation to correct errors and refine complex text. Omitting professional human judgment from the equation is just as unwise (read: dumb) in a translation context as letting ChatGPT or Copilot write your pleadings without back-checking every cite. The standard is called Machine Translation Post-Editing (MTPE).
- The second caveat is that, especially if the documents accompany a Hague Service Convention Request or are to be submitted to a court or government agency, layout matters. Formatting is highly important, what with headers and stamps and seals and images to worry about. If the official reading the thing can’t find a word-for-word comparison, game over.
- The third caveat comes from the professional translation providers I work with: not all languages are ripe for MTPE. There must be a “sufficient corpus of training data,” as one of my providers terms it, for the dictionary to reach a certain level of sophistication. So you’re probably not going to get much depth in Icelandic– which I understand to be a monumentally hard language to learn– but there’ll be plenty of training data for Spanish or Chinese. Accordingly, the AI-generated product is easier to fix. And to boot, with languages like these, there are more skilled editors available to do the cleanup.
- Dictionaries are proprietary, and they’re not all alike. One provider may have a top-flight library of patent work that they’ve translated into Chinese, but they don’t have much in the way of vocabulary for contract disputes in Swedish. Yet another provider might have a data center in Stockholm that cranks out great work.
- Last caveat: MTPE puts a much heavier cognitive load on the editor. Sure, these are smart people, but it simply takes longer to review machine product than it does to review human product, and making changes means decision fatigue. That reduces otherwise expected savings– for a few hundred pages, cost reduction will be substantial. But for a ten-page job, MTPE might actually end up costing more than traditional work.
- There will be more caveats as we get into this.
We are finally at the dawn of the age we’ve hoped for for years, and we can really start seeing bills go down thanks to available technology. We just can’t get too confident about the whole idea just yet. But it’s coming.
Maybe this guy isn’t too far off.
* I use Google Translate all the time. Every single day, because it gives me a pretty good sense of what a foreign-language document says (thanks to Google Lens, I know exactly which gelato to ask for in Rome: Pistacchio, per favore), and it allows me to communicate somewhat competently with someone who doesn’t share a common language with me. I just used it last week to tell a housekeeper at a hotel in Mexico that, “yes, ma’am, we’re checking out, so you can turn the room around for the next guest.” It just can’t provide the level of quality necessary in a legal environment.
** I asked Gemini, Google’s AI platform, to define Dictionary in this sense: “In a Machine Translation Post-Editing (MTPE) context, a dictionary is defined not as a book of definitions, but as a structured digital database used to inject pre-approved, company-specific terminology directly into a Machine Translation (MT) engine.“