The River Poem is a digital poem made by Maynooth University Lecturer Jeneen Naji, Independent Interactive Developer Pauric Freeman and Digital Exhibition Developer Mark Linnane in collaboration with Maynooth University’s Building City Dashboards research project, this poem projects snippets of text generated by the Generative Pretrained Transformer 2 machine learning algorithm onto a 3D model of Dublin City. The words themselves were generated by a GPT-2 machine learning algorithm that was trained on James Joyce’s Finnegan’s Wake.

Finnegan’s Wake is a narrative that is ideally suited to transposition to the digital medium because of its multilinear stream of consciousness style and its neologistic language. The 3D model of Dublin city was made by the Building City Dashboards research project that seeks to determine how to build more extensive and effective city dashboards by extending data visualisation tools and investigating novel forms of multimedia interaction, including the projection of data onto architectural models, and converting data into audio-textual forms. The River Poem intends to offer audiences an alternative approach to viewing the city by projecting words onto the Building City Dashboard’s 3D model. Snippets of phrases were removed from the algorithm’s output by the human authors and then placed in a spreadsheet. Python code in TouchDesigner software then randomly selects a new quote from the spreadsheet at specific intervals and places it into a queue. This queue forms a long string of text which is animated along a path that follows the route of Dublin’s River Liffey. When a quote reaches the end of the path, it is automatically swapped out for a new random quote. Using texture instancing in TouchDesigner kept processing power to a minimum and each letter is generated only once and instanced each time it is used in the long string of text. The path is also dynamic, as it can be edited in realtime to fit into different scenarios.

The final output creates a dynamic unique literary experience for each instance of play and explores notions of human versus machine agency and multimodal data communication within contemporary cityscapes. The original inspiration for The River Poem was Cento poetry which is a kind of patchwork poetry that is composed of lines from other poems, however the addition of the machine learning aspect has produced a contemporary literary artefact that calls into question the boundary between human literary expression and algorithmic agency.

We gratefully acknowledge funding from Science Foundation Ireland under the Investigator’s Award Program. Award number: 15/IA/3090