Creativity in Human and AI-Enhanced Literary Translation: A Keylogging Experiment

Abstract

Based on a recent experiment with MA students at the University of Innsbruck, Austria, this article analyzes the benefits and drawbacks of lit­er­ary post-editing. Using the keylogging software Inputlog, six experiment participants documented not only the product but also the step-by-step pro­cess that led to their final German translation of a short prose poem by Vir­ginia Woolf entitled “Green,” which was originally published in 1921. Some students worked only with monolingual and bilingual online dictionaries, while others post-edited an automated translation generated with the DeepL next-generation language model, which was launched in September 2024 and is based on a large language model infrastructure. Echoing previous re­search on literary post-editing, the experiment results show that AI-en­hanced literary translation may entail a loss of creativity. At the same time, as the translation of Woolf’s prose poem with its novel images and ex­pres­sions requires translation strategies that move underneath the textual surface level, the time gains that come with partial automation are negligible at best.