Designed and built a breakthrough keyboard for the Arabic script. This new design compresses the the keyboard into a learnable shape-based layout, and represents the letters’ cursive nature. To power underlying artificial intelligence, built the highest-quality open-source Urdu text corpus available today.
For this project I manage the work of 3 engineering teams comprising 9 people, a design team encompassing interaction and brand, 4 lawyers, 3 PR professionals, 3 organizations donating Urdu text, 2 copy editors, 10+ advisors, dozens of interviews, business contacts and presentations, and 500+ beta test sign-ups prior to launch.
Urdu is spoken by a 100 million native speakers, and by even more as a second language. In Pakistan, where Urdu is langua-franca, it is typed on a keyboard phonetically mapped to QWERTY. This is illegible to millions in this region who don't speak English and will come online for the first time.
Worse, the context of Urdu software is decades of compromise. All Pakistani publishing relies on pirated 90s software, and Urdu fonts represent the language so poorly that many Urdu speakers would rather use the Latin script.
I looked for areas of intervention that would allow improvements to multiple technology layers. Better keyboards would require and produce better language models and dictionaries.
While users blamed themselves for not understanding the interface, technologists consider it a solved problem.
Given the difficulties in Urdu technology, a simple task such as running a survey goes through multiple layers of translation, losing the ability to capture data effectively.
Tens of millions will come online in this region for the first time, have no technological baggage, and powerful smartphones. Using this environment, we looked to the design of the Arabic script itself to inspire simpler interactions.
A small number of shapes make up all letters in the Arabic script. These shapes are used to teach writing in this script. We use this inherent layering to compress the number of keys required, and make the keyboard more legible.
Arabic script is also cursive. To manifest how letters change shape, the keyboard changes as users type.
Our intelligent shape-based keyboard requires an autocorrect system to put in the correct dots after users select shape. To do this we sourced 6.2 million tokens of quality Urdu text, transformed and cleaned this data. We open-sourced this dataset, called Makhzan.
Creating our text corpus required the design of a custom process combining human and machine intelligence. After identifying patterns of errors, we created new software to aid human reviewers. This was then converted into a multi-step process to efficiently and reliably fix tens of thousands of typos, many of which were promulgated by bad word processors.
Urdu culture is intimately tied with typography. The development of Matnsaz involved significant research into typographic history. The Nastaliq style of calligraphy used in Urdu requires pushing the boundaries of modern technology. I frequently lecture on Urdu typography, and work with typographers on Matnsaz. Here are some teaching materials I use in lectures about Urdu type.
I'm frequently invited to speak about Matnsaz at conferences, and design institutes. These include:
Full list of press and talks available here.