machine learning / surveillance / moving image / language

The age long quest for free speech has only got worsen during recent years in mainland China, by which everyone person who used internet could felt the palpable chokehold by the invisible hand. How many conversations are never delivered, or their tones altered for the purpose of propaganda? It felt like a losing game, until netizens weaponized themselves with intellectually aggravating and often time humorous ways to combat the censorship, which serves the sociopolitical foundation of this project. By using neural style transfer to interpolate across three East Asian languages - characters from Han Chinese, Korean, and Japanese, I have came to generate characters that do not belong to any culture in particular, yet bear a striking resemblance to the Chinese Han.

Through a painstaking process of filtering and selecting the more “look-alike” characters, a list of new glyphs are created as the replacement for censored keywords. The inspiration came from creative means of digital steganography that netizens used to bypass OCR-optical character recognition censorship. Transcripting is one of such, which refers to the linguistic practice of creating a script with elements from different writing systems, such as Chinese and English, or by mixing conventional language scripts with other symbols and signs including emoji.