AI typeface shapeshifts for speedy reading

The appearance of text affects its readability. Dyslexia-friendly fonts, for instance, tend to be sans serif, widely spaced, and without underlining, italics, and other variation to avoid a crowded appearance. While some fonts are considered more readable than others, everyone has their individual preferences and needs regarding typeface.

The team of researchers at TU Darmstadt’s Centre for Cognitive Science began by investigating how the appearance of text affects its readability, with a view to creating fonts which can adapt to improve readability.

Once they had developed a technique for assessing reading speed, they needed to develop a method for synthesising new fonts to work alongside it. They achieved this using a machine-learning algorithm, which was taught the structure of fonts by analysing 25 popular and classic typefaces, including Helvetica, Times New Roman, Garamond, and Futura. The system was capable of using these typefaces to generate an infinite number of new fonts that can be any intermediate form of the 25 fonts, for instance, halfway between Garamond and Times New Roman.

They built a 3D representation of the original fonts; in this infinitely large space, all points represent a possible font which is a combination of the original fonts. The researchers applied Bayesian Optimisation to develop their adaptive font. This begins with an a priori uncertainty across a volume in the font space, and selects successive points in this volume (representing potential fonts), for which the user’s reading speed is evaluated through a task. Once the reading speed is known for that font, the algorithm reduces the uncertainty and edges closer to the ideal font. The researchers call their system 'AdaptiFont'.

Volunteers – all native speakers of German – were given text to read over an hour-long experiment, which used this algorithm to optimise reading speed through change in font. They read a total of 95 texts in different synthesised fonts from a German Wiki for children’s encyclopaedia texts, chosen for their simplicity. The researchers demonstrated that their algorithm did indeed generate new fonts to increase individuals’ reading speeds. While all the readers ended up with their own personalised fonts that they found most comfortable, this font may not be optimal in all situations.

“AdaptiFont therefore can be understood as a system which creates fonts for an individual dynamically and continuously while reading, which maximises the reading speed at the time of us,” said Professor Constantin Rothkopf, a cognitive science expert at TU Darmstadt. “This may depend on the content of the text, whether you are tired, or perhaps are using different display devices.”

The researchers recently presented AdaptiFont to the scientific community at the Conference on Human Factors in Computing Systems, and have since filed a patent application for the system.