On the heels of raising a big round of funding at a $1 billion valuation last week, DeepL is taking the wraps off a new language product, the first extension for a startup that made its name from its popular AI-based translation tools.
Write is a new tool that fixes your writing — catching grammar and punctuation mistakes, offering suggestions for clarity and more creative phrasing and (soon) giving you the option to change your tone. Write is based on the same neural network that powers DeepL’s translator, and significantly, it is another step ahead in how artificial intelligence technologies, specifically those in natural language processing, are being used to alter how humans are communicating with each other, a big theme at the moment.
The functionality of Write may sound a little familiar to you. That’s because DeepL’s new service is going head to head with a popular product already on the market: Grammarly is currently leaned on by more than 30 million daily active users and some 50,000 businesses and teams.
If you don’t use Grammarly, chances are that you at least know about it — its YouTube ads are (in)famously ubiquitous. And at last count, despite much nay-saying from VCs about the pitfalls of features versus platforms, Grammarly, as of 2021, saw a huge spike in its valuation, from $1 billion to a whopping $13 billion.
DeepL’s mantra is to embrace competition and use it as a motivator to do things better. In its primary (and before today, sole) product, the company has long competed against two of the biggest tech companies in the world, Microsoft and Google, which both offer real-time translators for individuals and as a service used by third parties. (And that’s before considering the myriad other options out there for real-time translation.)
“We are always in race mode,” CEO and founder Jarek Kutylowski said of its flagship translation service. “We are accustomed to big adversaries, and part of our culture is to push forward through that.” Indeed, for many, DeepL’s neural network-based translator works better than these others, capturing certain nuances and meanings that rivals have missed.
That approach, it seems, is now the template for how DeepL will tackle new product frontiers, starting first with Write.
Launching initially in English and German and as a monolinguistic tool (you put in English writing and get English results), the plan is to see how Write is used across these two languages first, both to improve them and to figure out how to develop Write, whether that means new features or new languages. As with its basic translation tools, you can use Write free without needing to register (as you do with Grammarly to use its free tools).
Options, it seems, are at the crux of the service: In addition to snagging basic grammatical and punctuation errors, the focus will be on generating choices for users covering style, tone, phrasing and diction, rather than re-writing everything that’s entered into Write. In doing that, you might ask if Write is showing its limitations, or if its creators are making a conscious choice of where AI can help people do their best work. That’s a debate that definitely has opened up with the release of OpenAI’s GPT service, which takes a basic brief and writes everything for you based on that.
My initial test-drive of Write was a mixed bag, and reveals that there is still room for learning and improvement. Write, ironcially, wasn’t very good (yet?) at guessing what I meant below when I wrote “right” instead of “write.”
Image Credits: DeepL (opens in a new window) under a license.
But it did a good enough job of fixing my misuse of good.
Image Credits: DeepL (opens in a new window)(opens in a new window) under a license.
Of all the directions that an AI-based translation startup could choose to go for its next product, Kutylowski told me that DeepL decided to work on Write because of patterns that it started to notice in how its translation product was being used (or misused, as the case is here).
“People were misusing the translator to write texts in first in a different language from English, and then plugging them back into English assuming that they cold get the AI to write a better original version,” he said. The team decided to build Write to essentially improve that and cut out the translation middle step, which was skewing the results anyway.
The “AI writing companion,” as Kutylowski describes it, is aimed both at native speakers but perhaps even more at people who write passably if slightly awkwardly in a second language and are hoping to give their words that extra native shine. “The idea is we could help a student improve a grade by one,” in a manner of speaking, he added.
The next level of development, he said, will be to focus on the elusive qualities of tone, “not content but phrasing, creative input,” he said — but critically doing so while continuing to anchor that content in a person’s own words and ideas, not those generated by the AI from scratch.