This study proposes a generative language model called AfriKI – an abbreviation for Afrikaanse Kunsmatige Intelligensie, or Afrikaans Artificial Intelligence. The approach is based on an LSTM architecture trained on a small corpus of contemporary fiction. Specifically, AfriKI is used as a writing tool to explore machine-in-the-loop, or co-creative, Afrikaans poetry generation. This experimental integration of computational and poetic practices conceives the model as a device for ‘making language strange’. The generated results contain unusual, compelling figures of speech that are interwoven by the human poet to create verse. Thus, the emphasis falls not on full automation but on the augmentation of human creativity through the defamiliarization of language.
About the SADiLaR DH colloqiuims:
SADiLaR organizes a monthly (online) colloquium showcasing research related to digital humanities. Each month a speaker will present their work in the area of digital humanities.