4 Investment Challenges When Customizing MT

More and more R&D is being put into artificial intelligence—at least as it applies to machine translation—and an ever greater number of companies involved in the language services industry are beginning to take notice. Such actors must decide for themselves whether to take the leap and invest in customizing an MT engine.

Naturally, the prospect of taking on a new R&D venture may be daunting, and neural MT engines can be quite sophisticated in their own right—so it’s easy to get lost in the jargon and shy away from the chance to implement such new systems. However, all companies, no matter the sector, can add R&D into their business strategies Indeed, it’s hard to think of a smarter way to manage your enterprise.

So if you’re in the situation described above—especially if you’re a small- or mid-size Language Service Provider—and you are wondering if it’s time to invest in developing your own MT engine(s), bear in mind the following four main ROI drivers, and boost your chances of turning this solution into a profitable investment.

  1. Large-scale volumes. You’ve got to think big when it comes to properly training and customizing an MT engine. So whether you can conjure up one nice, big project of a few million words, or you have a particular loyal customer that you do business with frequently enough and who is looking for the most competitive offer—and could eventually generate the necessary large volume for you, that’s what it will take to get your engine to the quality level that you and your clients need.
  2. From light post-editing “gisting” to ready-to-publish output. If you’re using a customized engine, in the beginning there will always be some room for improvement in intelligibility and fluency. These are both relevant and, when done right, go a long way in justifying the original investment (i.e., provide a good ROI).
  3. Quantifying improvements. Let’s say your business is doing everything right in terms of incorporating MT into its strategy. What you’re inevitably going to have to ask yourself is: just how much have your engines improved through training? Thus we can appreciate the importance of having metrics on productivity, quality and usability—information that is crucial in determining whether or not a customized engine tends to yield “perfect segments,” rather than an output that must be “reworked.”
  4. Critical errors. These are the bane of any LSP—though critical errors, of course, have even greater weight in certain domains like medical and legal translations. Ranging from mistranslations that alter meaning to incorrect use of terminology, weeding them out and preventing them from being introduced into your engine should be an extremely high priority.

At present, promising new MT solutions are becoming available that can open the door to the winning combo of lower investment plus increased profit. The future of ROI on this particular matter is bright indeed.