Intento for Translators

What are the pros and cons of using a tool like Intento that allows you simultaneous access to dozens of machine translation (MT) engines? (When I say “a tool like,” I’m really not being particularly accurate because it’s actually the only tool that does what it does.) If you’ve used generic MT engines like those from Google, Microsoft, or DeepL via your translation environment, you know that it’s rather tedious to create an application programming interface (API) key on their respective websites and then enter it into the application you’re using. It might be okay if you just want to use one or two MT engines, but if you would like to access 10 or 20, this becomes a real headache (and honestly untenable—if only because you don’t even know many of these engines exist or how to jump the language barriers of the many East Asian specimens to connect to them).

Why exactly would you want to have access to so many engines? The answer to that depends on a number of variables. For instance, if you work in a language combination that’s not particularly well supported by MT engines, the performance between different texts you translate may differ widely. The same might be true for languages that are well supported, particularly when it comes to different kinds of text or subject matter. You might be well situated if you’re already using a customizable MT engine such as OPUS-MT or ModernMT. (And, yes, it’s also fine if you don’t use MT at all.) But for others, it might be rather attractive to have only an API-based agreement with one entity (in this case, Intento) to connect you to engines such as AISA, Alibaba, Amazon, Apptek, Baidu, CloudTranslation, DeepL, Elia, Fujitsu, Globalese, Google, GTCOM, IBM Watson, iFLYTEK, Kakao Developers, Kawamura International, Kingsoft, Lesan, LINDAT, LingvaNex, Microsoft, ModernMT, Naver, NTT Com, Pangeanic, Process 9, Prompsit, PROMT, Rozetta, RWS, SAP, SYSTRAN, Tencent, Tilde, XL8, Yandex, YarakuZen, Youdao, or Zinrai. (I know. I didn’t know many of these MT providers existed, either.)

Clearly not all of them will be relevant (Fujitsu’s Zinrai engine, for instance, supports only English<>Japanese), but there’s a good chance that some engines will be relevant for your language combination, and an even greater likelihood that there’ll be more than you think.

So, how does one work with Intento? First, you’ll need to work with either Lingotek, memoQ, Trados, Smartcat, Wordfast Anywhere, Wordbee, or XTM if you want to bring MT suggestions via Intento into your translation environment. For each of these tools, a free plugin is provided (either readily integrated or separately installable) that enables you to connect to Intento. Then you’ll need one of the plans that Intento offers to access its ready-made connectors to its MT partners. The newly unveiled plans that are relevant to individual translators include:

  • Localization Starter for $25/month: This will provide access to all non-customizable engines, including the option of selecting one of 16 domains and up to one million characters per month. (All fees that are payable to the original MT providers are paid by Intento.)
  • Localization Expert for $75/month: This will provide access to all non-customizable and customizable engines with your own credentials, including the option of selecting one of 16 domains and up to one million characters per month with the possibility of incremental payments for overage use. (All fees that are payable to the original MT providers are paid by Intento.)

You can find the different options and offerings listed at

While the second option is attractive, my sense is that the first will be the most popular with translators. Why? Because the main differentiator is access to customizable MT engines, including AutoML, DeepL (with its terminology training component), Google, Microsoft’s Custom Translator, ModernMT, and many others. I have a hard time imagining a single translator building and maintaining several different MT engines that they would like to access through a tool like Intento. Although, for a small or mid-sized language services provider, the latter option might very well be interesting.

By the way, before you apply your preferred MT engine to your text, you can run a test that compares several outputs in the Intento Console. Or you can just select the engine you might know is likely to produce the best results right in your translation environment.

The first time I wrote about Intento in my Tool Box Journal newsletter, I stated: “I very strongly encouraged Konstantin Savenkov [Intento’s chief executive officer] to look into developing ready-made access plugins for tools like Memsource, memoQ, and Trados.” Well, Konstantin has done that and much more, and I’m glad they did. I’m thankful every time an interesting piece of translation technology becomes available not only to enterprise customers or language services providers but to freelancers as well. And there’s no doubt that Intento’s technology is here to stay. Pavel Doronin, the product lead at Intento with whom I talked to for this article, told me that there are 57 (!) full-time employees working for Intento around the world at this point. Also, the reports on the state of MT that Intento publishes annually on its website are required reading if you’re interested in how the landscape of MT is morphing in front of our eyes.

Let me add one thing that I think is important to consider when using Intento, and then a few things I would like to see in the future.

First, it’s really important to know about the confidentiality settings of the different engines. I’ve mentioned many times before that Google, Microsoft, and DeepL all assure us that they treat our data confidentially if we use their API to access them (which Intento does), but my sense is that these are the exceptions rather than the rule. Depending on your clients and their privacy needs, this is very important to keep in mind. In fact, it would be a very helpful service for Intento to provide clarity about the privacy considerations of their various partners.

Two other options that I would welcome are:

  • The proposal of more than one engine per segment so the translator can mix and match the suggestions.
  • A “smart” routing to the best-suited engine on a segment and not just on a document level.

Since I was successful with my last suggestions (though it admittedly took three years), maybe I’ll be successful again.

Jost Zetzsche is chair of ATA’s Translation and Interpreting Resources Committee. He is the author of Characters with Character: 50 Ways to Rekindle Your Love Affair with Language.

This column has two goals: to inform the community about technological advances and encourage the use and appreciation of technology among translation professionals.

The ATA Chronicle © 2023 All rights reserved.