Thread by @oulasvirta

Thread: How HCI research helped improve the French keyboard🇫🇷

It came out yesterday: The new AZERTY, the first keyboard standard created with the help of HCI and algorithms.

What a ride! Optimization + Models + Participatory Design

aalto.fi/news/changing-…

Scroll on👇 (1/14)

The NEW AZERTY includes commonly used characters, such as œ, « », É, -- and 60 other new characters. Arranged into intuitive groups. The layout is designed to improve ergonomics and performance.

Shout out to @AnnaFeit Mathieu Nancel, Andreas Karrenbauer, Maximilian John
2/

It all started in 2015 when the Ministry of Culture wrote in Le Monde that Azerty does not support grammatically correct French. (Ironically, it had better support for English characters.)

Here's the OLD Azerty:

Try typing « À l’évidence, l’œnologie est plus qu’un ‘hobby’ 3/

We emailed AFNOR, the standardization committee, and offered help with optimization problem related to the assignment of characters.

The new design is a result of 3 years of collaboration between experts, researchers, and the public.

Here's Mathieu Nancel with the new kbd - 4/

Collaboration was hard! Optimization methods and design are like oil and water.

Optimization requires point-precise inputs and at first there was no consensus even on what the goals are.

That's why we focused on tools for PARTICIPATORY OPTIMIZATION -- 5/14

We created tools that helped the committee explore designs, test hypotheses, and study the consequences of design decisions.

You can try one tool like that yourself here:
norme-azerty.fr - 6/14

We could change the layout by hand and the tool would tell us how this impacted typing speed or ergonomics.

We then adapted the underlying computer model to also take into account, for example, cultural aspects and comments from the French public

7/14

On the way we had to solve some SERIOUS algorithmic and modeling challenges.

We built statistical models of character use in modern French, drawing on newspaper articles, French Wikipedia, legal texts, emails, social media, and code.

8/14

We gathered the keypresses of over 900 people in a large-scale crowdsourcing study to learn what counts as an ‘easy’ key press.

We defined four HCI objectives for efficient solution: performance, ergonomics, familiarity, learnability.

9/14

And we formulated an algorithmic problem (called the QAP) to solve all of this.

The size of the problem in the end was whopping -- in the order or 10^200.

We solved it efficiently using integer programming 🧐 10/14

I'm super proud of Anna Feit, who won the @sigchi Dissertation Award 2019 for her thesis on this topic:
sigchi.org/awards/sigchi-…

TIP: She's giving her award talk at CHI in Glasgow #chi2019

Here she is with me and her opponent @ShuminZhai in June 2018 - 11/14

Now the norm is out. But will the keyboard be used? Time will tell!

At the moment its fate is in the hands of keyboard manufacturers.

Meanwhile, we will work to offer drivers for DIY-minded users who want to try it already now.

Mathieu in the release event in Paris: 12/14

Takeaway for me?

HCI needs to develop its OWN computational methods:
global.oup.com/academic/produ…

The vision of artificial general intelligence is incompatible with the requirements of design and interaction. And it should not offer a comfy pretext to avoid theorizing. 13/14

There are literally hundreds of languages lacking a proper -- or any -- standard. You can only guess how devastating this is for the development of a language.

Our method can be applied to ISO keyboards using the Latin alphabet.

It is being open sourced. -- 14/14

Thank you for reading! 🙏  
#EU-funded #ERC @ERC_research #FCAI @FCAI_fi #Aalto @AaltoResearch #ClavierFrançais #CHI2019 #computationalinteraction

This is work with INRIA, Max Planck Institute for Informatics, ETH Zurich, Aalto University 🇪🇺