Text Generator for Style Imitation
Generating texts in the style of Robert Walser and Emmy Hennings
This project explored how machine-learning language models can generate short texts that approximate the writing style of specific authors.
For the Swiss National Library exhibition “Aufgeschrieben. Stift, Taste, Spracherkennung”, two separate language models were trained:
- one on historical texts by Robert Walser
- one on historical texts by Emmy Hennings
In the exhibition, visitors could select an author and complete a prompt such as “Today I’m writing about …” with a single word. The system then fed the prompt to the chosen model and generated a continuation—often extending the input into one or more sentences.
The experiment demonstrates, in an accessible way, how training data shapes a model’s vocabulary and syntax, and how “style imitation” depends on both the underlying corpus and the user’s prompt. It is intended for research communication and teaching, and encourages critical discussion about authorship, originality, and the limits of generative AI in the humanities.
Back to top