CHEN Qianxun | 2019 | 習字

習字 (Graphein) is a series of visual poetic experiments dedicated to an aesthetic computational study of visual and linguistic patterns in Chinese characters. It is a repetitive computational writing practice without the intervention of the human hand. Graphein's graphological and critical attention focuses on the inner structure of Chinese characters in an attempt to bring out visual poetic patterns and defamiliarize the components of a writing system and language itself. Graphein reads in-between the gaps and by way of transitions. It allows certain human readers to re-examine their ownership (or not) of what is supposed to be as familiar and intimate as their own language, or as strange as someone else's. Graphein is the dynamic algorithmic appreciation of an imperfect system that is used by some humans to read and by others to imagine reading.

The word “習字” means the practice of writing, with “字” being the unit of the Chinese writing system and “習” indicating both practice and learning with a hidden message in the necessary process of repetition. The project focuses on two different aspsects 1) the visual rhythm of Chinese characters through composition 2) visual similarities within Chinese characters.

倉頡之初作書蓋依類象形故謂之文。其後形聲相益即謂之字。
—— 許慎《說文解字·敘》

“When Cangjie first invented writing, it is presumably because he copied the forms according to their resemblances that they were called wén “patterns.” Then forms and pronunciations were added to each other, so they were called zì.”
Xu Shen, Shuo wen jie zi
1.1 The Copybook 習字帖
A collection of selected digitally mediated patterns based on Chinese characters (or radicals) is created by a customized web interface.
1.2 Radicle Patterns

Dynamic visual patterns of a short list of Chinese radicals (major components of Chinese characters).

1.3 The weaver

The weaver looks into the visual rhythm of Chinese characters in motion. The example shows the weaver with the first sentence from Tao Te Ching.

2. Chinese Visual Embeddings

In the second part of the project, I turned my focus from composition to the visual characteristics within each character. Chinese visual embeddings are generated using a Convolutional Neural Network (CNN), treating each Chinese character as an image and trying to extract inner visual features out of them. In the vector space of these Chinese visual embeddings, visually similar characters become neighbors.

The embeddings as well as related code can be found in the github repository for Chinese Visual Embeddings.

2.1 From Zi To

From Zi To is a digital textile created using Chinese visual embeddings. It starts from the character “字” and traverses the whole common character set (including both traditional and simplified Chinese) by selecting the one visually most similar to the current one as the next character from among all the candidates that have not been visited yet. It is a new computational way of reading through common Chinese characters, following a path with subtle visual changes. [Reading direction: from top to bottom, then right to left.]

2.2 Traverse

A dynamic gradient traversing experience based on Chinese visual embeddings. Starting with a randomly selected character from the vector space, the program gradually replaces each character with its nearest neighbor. The result is a constantly changing grid pattern created by three to four Chinese characters that interlace with each other.