Some time ago, we published a post about Grammarly, a machine language proofreading program , and a second one devoted to alternatives to it. However, another serious player has recently appeared on the market of this type of application, for the development of which the authors have recently received a large subsidy. We are talking about the startup DeepL, which became famous for its translation tools based on artificial intelligence, and its new product is Write. However, will it work in proofreading scientific articles?
Another tool for machine correction or something more?
Write is a new tool whose task is to correct the text – it catches grammar and punctuation errors and provides suggestions for clarity and more creative formulation of sentences or even the tone of speech. Write runs on the same neural network that powers DeepL’s translator, and importantly, it marks the next step for the company in the way it uses artificial intelligence technology, especially in natural language processing.
The Write functionality may sound familiar, and it doesn’t seem to be anything new. This is because the new DeepL service competes with Grammarly, which is clearly emphasized by the creators themselves, stating that their modus operandi is to welcome competition with open arms and use it as motivation to do things even better. In its core product, the translator, the company has long competed with the two largest technology companies in the world, Microsoft and Google, which offer real-time machine translation for individuals and businesses. As the president and founder of the startup, Jarosław Kutylowski, said: “We are in the race all the time. We’re used to big opponents, and part of our work ethic is to cut through them.” Indeed, in many cases, the DeepL neural network translator performs better than others, capturing certain nuances and meanings that are overlooked by competing programs.
It seems that this approach is an indicator of how DeepL will try to deal with new challenges entering the market of machine proofreading products, the first step of which is Write.
In addition to catching basic grammar and punctuation errors, the emphasis was put on generating potential choices for users regarding style, tone, and word choice rather than rewriting everything that was entered into Write. As with basic translation tools, you can use Write free without registering (as with Grammarly).
Below we present a short test that shows that the program still has some shortcomings and room for improvement. Write, ironically, wasn’t very good at guessing what the author meant by misspelling “right” instead of “write”.
But the software corrected the misuse of “good” well enough.
Even this short test seems to show that this software is still not advanced enough to cope better than Grammarly with linguistic proofreading of scientific texts.
Who is the target user and further development direction of the program
“AI writing companion”, as Kutylowski describes his product, is aimed at both native speakers and people who write in a second language quite well, if a bit clumsily, and hope to give their words a native speaker touch. The next level of development of the program will be focusing on the elusive features of the tone of speech, “not so much the content, but the manner of expression, creative input” – said Kutylowski – but at the same time continuing to anchor this content in the author’s own words and thoughts, and not those generated by AI from scratch.
Kutylowski’s statement seems to confirm what was written above for the test, the program tends rather not in a scientific direction but rather in a domestic or business direction, so machine proofreading of scientific articles will cause him problems and will leave some rudimentary errors in the manuscript.
Man vs. Machine
As you can see, the DeepL solution is not perfect, so if you want to be sure that the text does not contain any errors or linguistic shortcomings and will be suitable for publication, it is best to use professional proofreading services, i.e., language corrections made by a native speaker. Despite advances in technology, this is still the best solution to ensure the highest linguistic quality of the manuscript.
An additional advantage of this solution in the case of scientific works is that by selecting the appropriate service, the text can be taken care of by an active scientist from an academic center in an English-speaking country. In addition, what this software cannot do, but what a proofreader can do is give your manuscript the fluency of a native speaker, as well as the insight of an expert who understands the research area and makes sure that specialized vocabulary is used.