The Character Thinks Ahead

The Character Thinks Ahead (version 2) by Hazel Smith and Roger Dean is focused on the computerized generation of creative writing using deep learning neural nets. It knits together visual, sonic, linguistic and literary elements that all interact with each other. Of the three dynamically rolling columns of text in the upper part of the screen, the middle presents three pre-composed poetic texts that suggest ideas, feelings and contexts to do with war, hierarchy and competition respectively. The two columns on either side display text generation using deep learning nets: in the left column the text is generated by character, in the other it is generated by word. In the bottom part of the screen there are also three distinct elements to the display. An animated word cloud in the middle highlights features of the ongoing texts. To the left of it is a dynamic spectral visualisation of a (pre-recorded) rendering of the live speech: this is live-transformed to provide a sonic output visualized spectrally on the right. Besides the visual elements, the live speech and the live sonic transformation, there is also pre-formed sound — composed and improvised by Roger Dean, Sandy Evans (saxophone) and Greg White (computer). The spoken text (performed by author Hazel Smith) plays on different senses of the word character: character as part of a word, as a register of behaviour or as a fictional being. It also relates to “thinking ahead”, central to the predictive aspects of deep learning (or even “thinking a head”). Ideas of competition between word and character, which are a feature of the spoken text, are also a feature of the process of text generation. These ideas are further explored in the screened pre-composed poetic texts (middle text panel). 

Description by the authors adapted from Dean, R. T., & Smith, H. (2018). "The character thinks ahead : creative writing with deep learning nets and its stylistic assessment". Leonardo, 51(5), 503-504. https://doi.org/10.1162/leon_a_01601