Characters are more than communication. They are related to culture, identity, history, and cultural heritage. So there are many reasons to get them. This could now be done in a very modern way – with the help of an algorithm.
It could be a revolution that researchers at the Ecole Polytechnique Fédérale de Lausanne (EPFL) have achieved. They worked together with the ECAL master’s program in Type Design to do this. It’s about art, but most of all, about the process of creating digital fonts. They have developed a machine-learning algorithm to generate Chinese characters (Hanzi) as images.
For example, to create an alphabet in the font design, each letter shape must be drawn in a uniform style. The so-called glyph is the basic symbol in every letter form. It is a legible sign for writing. For example, if you use the Latin script, at least 26 capital letters are required for the design. The Chinese Hanzi script has 6,763 commonly used characters and tens of thousands of characters for more complicated sentences. If you were to create a typeface conventionally, it could take years.
AI: Useful Support In Font Design
Shuhui Shi has been designing fonts for two years. “When I draw Latin scripts, a few dozen glyphs are usually enough for everyday use. However, when I tried to apply my knowledge of font design to the Chinese language, I found that it was almost impossible to complete a standard character set with 6,763 glyphs on my own,” explains the student. And she was not alone in this problem, as she found. Many other designers fared similarly.
With such a large number of glyphs, artificial intelligence might help. This is how the so-called AIZI project came about, in which the EPFL Computer Vision Laboratory played a key role. First, the researchers developed glyphs that they could use to train the AI. To do this, the student created a large database of characters that could be broken down into very simple components. Their idea: A new font only requires a few basic components to be designed. They could then be put together automatically to form all the fonts. “We then developed algorithms to do the second part, namely to create different characters based on these components.”
AI Could Make The Job Of Designers Easier
This algorithm can produce tens of thousands of hanzi as images. The result: Chinese characters can be created in this way. It takes less than 500 Hanzi entries for this. According to the researchers, the algorithm path represents a contemporary, transdisciplinary approach that stands at the interface between design and technology. “Although we’ve used the technology called General Adversarial Network or GAN many times, working on such a challenge was a first. GAN is a deep neural network that aims to generate images such as faces. Then, our idea was to use this type of technology to generate characters”.
Shuhui Shi has come to completely new perspectives through collaboration. An AI assistant can help designers create an entire font with fewer than 500 human-designed hanzi. “We currently have a database with more than 90,000 hanzi and tens of thousands of well-scaled, generated hanzi in the left-right composition”. However, the results would have to be revised and optimized again. As soon as the student is back in China, she wants to continue studying machine learning and font design. She hopes that this first step will make it easier for other designers to get started. The use of technology makes it possible to concentrate more on creative ideas without designing thousands of characters. Perhaps this way, more people would try their hand at their first Chinese font.
AI Could Also Preserve Cultural Heritage
For Matthieu Cortat-Roller, head of the master’s program in Type Design, this new approach has an even deeper meaning. He cites the example of the Syrian script, which is only used by very few people today, as a large part of the Syrian population stayed in other countries due to the war in their home country. “They worry about the future of their culture and identity. It’s not that easy to get a writing system when you live in the diaspora”. AI can help preserve cultural heritage.