AI-based DeepL is different

« previous post | next post »

So says DeepL CEO Jarek Kutylowski.

"DeepL translation targets Taiwan as next key Asian market:  CEO says AI-based model is aiming to refine nuances, politeness", Steven Borowiec, Nikkei staff writer (September 16, 2024)

DeepL Write is one thing, DeepL Translator is another.  We've examined both on Language Log and are aware that the former is already deeply entrenched as a tool for composition assistance, but are less familiar with the special features of the latter.

The article by Borowiec, based on his interview with CEO Jarek Kutylowski, begins with some not very enlightening remarks about the difference between simplified characters on the mainland and traditional characters on Taiwan, attesting to the truism that CEOs and CFOs often don't know as much about the nitty-gritty technicalities of the products they sell as do the scientists and specialists they hire to make them.

The article then focuses on the business aspects of  DeepL, where Kutylowski is on much firmer ground, when he tells us how many hundreds of millions of dollars investment DeepL's translation software has attracted and how many billions of dollars of valuation it has achieved.

When the conversation turns to more general concepts of different approaches to machine translation, I perked up and was all ears.

DeepL was founded in 2017 and touted itself as the first online translation platform to use neural networks and machine learning.
 
The model was fed countless examples of translated sentences in each language in order to teach it to recognize the natural structures of sentences, Kutylowski explained. He contrasted this with the more conventional approach, in which models rely on estimates of probability and try to "guess" which words are most likely to follow one another in a sentence.
Jarek, a native speaker of Polish, pointed to how his mother tongue and other languages have different forms of address depending on the level of familiarity between speakers.
 
He hopes to soon introduce improvements to DeepL that can improve the quality of translations in such areas. "What helps the AI to solve those problems is having a lot of context. We are actually working on some technology that is going to allow us to solve that by trying to gather that information and trying to gather that context from the user, when it's necessary. I expect that to be available pretty soon."

I have often exclaimed how remarkably good Google Translate is, and I'm absolutely astonished at how many different languages it can translate to and from, but DeepL is aspiring to give it a run for the money.

 

Selected readings

[Thanks to Don Keyser]



2 Comments »

  1. Chester Draws said,

    September 18, 2024 @ 4:03 pm

    I use Google Translate, DeepL and the Microsoft equivalent regularly out of Russian and Polish.

    DeepL is the best of the three, and MS the worst, in my opinion.

    What is interesting is how when DeepL gets in a tangle that Google Translate often doesn't and vice versa. Clearly they are using quite different methods.

    They are also clearly improving.

  2. AntC said,

    September 18, 2024 @ 9:20 pm

    I have often exclaimed how remarkably good Google Translate is, …

    You have; and I plain don't understand why. I agree with @Chester that DeepL translate is usually the better — when they differ.

    I'm not referring to translation of literary works and fancy flourishes. But quotodien tourist/traveller practicalities. Too often the alleged translation (to/from English/traditional script Chinese) gives blank incomprehension. I keep to plain declarative sentences/questions AFAP, no subclauses or heavy use of pronouns. I try to avoid words i know to be polysemous. (Also I have to constantly beware autocorrect: no, 'polygamous' would make no sense there, stupid.)

    They might be "improving". That doesn't yet make them fit for purpose.

RSS feed for comments on this post

Leave a Comment