Billions of people use keyboards daily, all across the globe. But the keys traditionally used were designed for English; special characters in other languages, like letters with accent marks, are difficult to access or missing altogether.
Now researchers have harnessed the power of the algorithm to create a new keyboard standard for French typists—and they say it’s easily adaptable to design new keyboards for all kinds of European languages.
The new AZERTY standard developed for France’s Ministry of Culture by AFNOR, the benchmark French body for voluntary standards, used a predictive algorithm to design a keyboard that is more intuitive and ergonomic for French speakers than the current AZERTY keyboard. It includes common French characters like œ and É, as well as 60 other new characters not included on the existing keyboard. Symbols such as @ and #, which have come into greater use in the age of Twitter and Facebook, have been moved to more accessible locations.
The project began in 2015, when the French government decided it wanted a new keyboard standard. A team at Finland’s Aalto University read about it in the news and, after consulting with a postdoc on their team from France, they seized the opportunity to apply their work in computational methods to the design.
“We saw it as a big chance to bring our research to life and make use of it in a public project,” says Anna Maria Feit, the lead researcher. “It was a perfect job for an algorithm.”
The challenge sparked work from an international collaboration led by Feit’s team and included linguists, economists, keyboard manufacturers and more stakeholders.
The problem, France explained to Feit and her team when they began in 2016, is inherent in the way young French people type. It’s easier to write an E or an A rather than a multi-key command to create É or Á.
“Especially younger people don’t type French in the proper way, to the point where many of them think the accents are no longer needed,” Feit says. “They wanted to counteract that trend.”
Easy enough, Feit and her team thought. The French government had given them four key parameters: The letter characters had to be kept as they were on the current AZERTY keyboard, to avoid too big a change for users. That’s also why the researchers built a “cost function” to quantify how far away new placement of characters would be compared with the standard keyboard. Performance was key, so they wanted similar symbols grouped together, and finally, the keyboard needed to be ergonomic and easy to use.
Feit and her team already had years of research on using algorithms for optimization, and it was relatively simple to collect datasets of billions of characters taken from not only French newspapers and legal texts, but also social media posts and Wikipedia entries. They also put together a large crowdsourcing study in which they asked 900 people to type letter combinations to show them how people physically type.
“We created this optimization model process, essentially listened to what the algorithm said was the best design, and said ‘Ta-da! Here is the keyboard!” Feit says. “[The French government] said, ‘Oh, well, that looks nice! But we were thinking this should go here and what is that over there and what about this character…’”
And so the algorithm’s recommendations were combined with human input, which also included the French government publishing the proposed layout for public comment. The project received more than 3,000 comments, which was a record for government projects posted for comment.
After submitting about 30 to 40 iterations, by Feit’s estimate, the team had created a tool for the French committee that can design ideal keyboard layouts based on parameters they enter. The design they adopted was formally introduced on 2 April, and keyboard maker CHERRY is slated to release the first keyboards this summer. In the meantime, an interactive AZERTY keyboard is available to get typists used to the new setup.
Feit says the tool can also be adapted to design keyboards ideal for different languages in the future.
“We are now in the process of making our algorithms openly available, and we purposefully made them general for adaptability,” Feit says. “With new data from other languages it would be very easy to design for the next one, which is really exciting to us.