itinerant space

Defending Jingjie in Front of His Interpreters
Jingjie Zhang

Defending Jingjie in Front of His Interpreters can be summarised as a critical practice demonstrated through performance, exhibited at the RCA Research Biennial in February 2025. The project was presented as a continuous poetry recitation performance, beginning at 11 am on February 21st and ending at 6 pm, lasting 6 hours and 54 minutes. The project aims to construct an extremely pleonastic (redundant) poetic listening live. During this period, as a performer, I continuously recited approximately 40,000 Chinese characters of poetry generated by language models imitating my writing style. These poems were pre-printed on A4 paper, and after completing each page of recitation, I attached the finished A4 sheet to the wall. Simultaneously, a projector in the venue displayed scrolling English subtitles of these 40,000 characters. The performance stage faced audience benches, with another performer, Yuhui Huang, whom I invited, sitting on the left side of the bench. He played the role of a symbolically perfect listener, as he listened to the entire 7-hour recitation from beginning to end. During the final hour and a half of the performance, the microphone was connected to a series of resonators and effect processors, with effects gradually activated and amplified, ultimately transforming the recitation sound into an indistinct and unrecognisable noise.

The pre-generated poetry texts came from several widely used commercial language models currently on the market: ChatGPT, Claude, and Deepseek. All texts simulated my writing style—achieved through feeding large amounts of my earlier writings to these language models and attempted to use simple text mining methods to perform frequency statistics on nouns, adjectives, and verbs I commonly use, while considering my usage patterns of imagery and grammatical habits, to simulate and complete the generation of these 40,000 characters.

Through this long, uninterrupted recitation, relying on the liveness of the performance, it creates an extremely redundant noise practice model and a fluctuating social practice relationship (Fig.1). Due to the extremely long duration, the performance is no longer completely trustworthy and reliable content for the audience. The fluidity of the exhibition space means that almost no audience member can complete the listening from beginning to end, the audience relies on imagination to complete their presence. This implies a selection of pleonastic noise. It also completes the transformation of the human voice from linguistics and semantics to pure acoustic sound and noise. Therefore, this pleonasm is often recognised as part of the performer in the live setting, presenting a noise-like extension – a collision and fusion between the signifier of noise and the signifier of my identity — to hypothesise the extreme scarcity that may be produced by the excess and redundancy of symbols in the daily practice of AI technic objects in capitalist society. As Gustav Metzger is concerned about this machine participation in art, like other technologies (such as film), content that is produced too neutrally and generally might accelerate the process of being consumed by capital. Thus this practice relies on liveness to discover the liminal moments’ challenges of noise practice – entering a blank space of speechlessness through excess, entering the drift of thought and fatigue from information overload through pleonasm.

In my research, hallucination noise, as a noise phenomenon in the complex relationship of mutual influence and prompting between humans and technic objects, requires exploration and experimentation in its forms within cultural and artistic practice. Defending Jingjie in Front of His Interpreters specifically focuses on the human voice, treating it as a concrete form and practical method of hallucination noise research in practice. Making the human voice a core part of hallucination noise research methodology places human voice within a special category of sound—it simultaneously belongs to acoustics and linguistics, to both sound and meaning. The human voice overlaps between these two domains: the former can be summarised as the pure listening perceptual experience of acoustic waves, music, sonic elements, and noise; the latter can be summarised as the pure meaning comprehension experience of linguistics, semantics, symbols, text, and signifying chains.

Therefore, this practice intends to establish the important position and status of the human voice in the hallucination noise research, and to experiment with how the human voice will form a site of pleonasm and redundancy perception through artificial intelligence models, further exploring specific methods issues and challenges of hallucination noise in artistic practice.