top of page

p.s. Did you know we have a Slack community for UX localization? Join right here – and don't forget to join the relevant language-specific channels, too. Can't wait to see you there!

Pushing for change in big orgs with Gilad Almosnino

4 June 2024

To localize or not to localize: Making the right decision

15 May 2024

Localizing success messages: Creating local magic moments

7 April 2024

More to read

Great localization content in your inbox. Unsubscribe anytime.

You're in! Talk soon :)

Let's talk about machine translation. Smartling announced last week that they're launching patent-pending tech (woah, big words) meant to completely change machine translation as we know it. They're talking about using advanced prompting to tweak MT output, which basically means it'll give you the option to give your MT instructions on style preferences, voice, gender requirements, and other factors that impact translation results. Your MT will then proceed to take those instructions and use them very literally. Ever tried telling chatGPT to do something? It feels like the dad trying to make a PB sandwich in this very hilarious video.

But while their very ambitious claim to fame may be a bit of a stretch, I love the approach they're taking to this. Done are the days when all companies use one generic MT for their every need. In this age, we expect to have custom everything, including the MT engine we're using. The ability to customize our MT output is a long time coming. Despite LSP marketers' love for the term "human-led MT" (fancy code word for "we try let humans look at our MT output"), the industry is moving towards using MT in every loc task and for every language. Translated just announced they're now serving 200 languages with their Adaptive MT offering. That's a massive number of markets about to be served machine-translated copy in various levels of quality. If this is our new normal, it's about time we start thinking about how to make that MT ouptut readable to people, not just machines. And for localization, this means a user-centered approach to loc quality, even if it is powered by machines. I'm guessing Smartling's initial results won't be as impressive as promised. But it's an important first step towards getting better copy in all languages. It's a critical understanding – that copy is never a one-size-fits-all solution, and international users deserve to have tailored copy results, too.

Do you think Smartling's custom MT prompting is a game-changer or a false promise? Michal

Can bots make a sandwich? (30/4/23 newsletter)

New developments in MT let you provide custom instructions with your input

Michal Kessel Shitrit



bottom of page