Poetrify

Installation

Just as with music, digital anthologies should ideally adapt to reader preferences. But how can we best capture tastes without resorting to long questionnaires or to painstaking observations over the course of the readings? The exploration of new forms of anthology aims to improve and better understand how wider audiences connect with poetic heritage. The principles behind this installation are similar to musical platforms such as Spotify or Apple Music that allow users to create "radios" reflecting their musical tastes. Rather than following a selection already composed by a poetry expert, this anthology enables interaction and encourages discovery. The project is developed from 1,500 tagged texts that are then entered into the database. The poem “playlists”, or sequences, are adapted to the tastes of the visitor who can then immediately share these by reading them aloud to someone else. Here, readers also become creators of logical suites, thereby discovering 19th century poetry. The anthology is in French and English and obeys two distinct principles: the first according to similarities in qualities (tagging), and the second through Artificial Intelligence. To further enhance joint discovery, the anthology allows for reading aloud as well as the recognition of the criteria determining the poem sequences.

Technique

This first presentation of the bilingual anthology, “Poetrify”, uses 1,003 texts in French and 500 poems in English – that is, over 20,000 verses, couplets, or lines of prose. The first phase of the project consisted in verifying the edition of the texts, ensuring their quality and tagging them (“tags”) according to the author’s gender, poetic form, themes and the various criteria selected. The same texts were used for the database that fed the Artificial Intelligence in assisted creation.

In this digital anthology, we give everyone the ability to make their way through a vast collection of poems. The path forward is ‘flagged’ based on the preferences identified as users advance through the anthology. To build such an anthology, the machine offers a range of clear preferences that demarcate the first part of the route. Then, a less obvious similarity – using artificial neural networks to represent words – makes it possible to continue the journey based on the initial logic with, nevertheless, the promise of surprises and happy discoveries along the way with which to fill the visitor’s personal anthology.

Concept:
Antonio Rodriguez, UNIL 
Andrei Popescu-Belis, HEIG-VD 
Aris Xanthos, UNIL

Text editing:
Melina Marchetti, UNIL 

Development:
Andrei Popescu-Belis, HEIG-VD


Reference:
Maria Eriksson (et al.), Spotify Teardown: Inside the Black Box of Streaming Music, Cambridge (Ma), MIT Press, 2019.