Launch of AIZI – a unique database of Chinese characters
08.12.2021
www.aizi.ch

www.aizi.ch

Despite being used by some 900 million native speakers, the Chinese writing system currently relies on a small number of digital typefaces, in print or on screen. This is partly due to the quantity of characters. There is no official figure, but some dictionaries reach up to 106,230 glyphs, with the Unicode standard featuring “only” 20,902 glyphs. A Chinese scholar knows over 13,000 and being able to read a mere quarter of this figure is nothing to be ashamed of in contemporary China. For a designer, creating a Chinese typeface can easily take more than a year and represents a far greater investment, both in time and money, than a Latin (or Greek, or Cyrillic) one. These practical difficulties also limit foreigners’ interest in Chinese type design, as the task seems insurmountable.

Could Artificial Intelligence (AI) help Chinese type design overcome its current limitations? Is it possible to teach a Machine Learning programme about the rules of Chinese composition and design in order to enable it to create the thousands of glyphs required for a typeface? “The initial idea behind the AIZI research project was to define a reduced set of basic “seed” characters that could be used as training data for an AI system, with the ultimate goals of democratising the design of Chinese typefaces and access to script for beginners and foreigners and to expand the stylistic range for a writing system that is largely dominated by traditional brush-based calligraphic shapes. Eventually, this could lead to greater quality in the production of fonts for Chinese script”, explains Shuhui Shi, who launched the project as part of the ECAL Master Type Design.

Rooted in traditional structures, the system that has been developed analyses and rationalises the construction of glyphs and has resulted in a database – available in open access on www.aizi.ch – that could be used to train AIZI algorithms as well as any other future AI tools. The database contains 93,876 entries and is geared towards type designers, providing them with the reference needed to draw a set of Hanzi, even if they do not speak or read Chinese.

AIZI was developed by a team composed of both designers (ECAL) and engineers (EPFL). As part of the ECAL Master Type Design programme, Shuhui Shi, initiator of the project, was in charge of the design part in collaboration with Kai Bernau, under the supervision of Matthieu Cortat. The Machine Learning process and algorithms were developed by Wei Wang, under the supervision of Mathieu Salzmann from the EPFL Computer Vision Laboratory.

Information
Shuhui Shi – Head of project, ECAL: shuhui.shi.ss@gmail.com
Matthieu Cortat, Head of Master Type Design, ECAL: matthieu.cortat@ecal.ch – www.ecal.ch
Mathieu Salzmann – Senior Researcher, Computer Vision Laboratory, EPFL: mathieu.salzmann@epfl.ch – www.epfl.ch/labs/cvlab/