Towards Measuring Linguistic Creativity in Literary and Non-literary Text. First results and insights

Our project aims to operationalize and measure linguistic creativity across genres. We want to provide a robust empirical account for measuring stylistic originality and, beyond that, the linguistic creativity of sentences, paragraphs, and full texts. By means of a case study on spatial descriptions, we develop data-driven metrics and models to automatically identify creative signs in accordance with human creativity ratings. Approaching literariness and linguistic creativity from the perspectives of rhetorics, computational stylistics, and NLP, we disentangle literariness at the level of the sign from creativity at the level of usage. The poster reports on the findings of the first stage of the four-year research project. It will show how different assessments of linguistic creativity relate to one another, discussing how specific textual features and creativity ratings correlate and whether perplexity, determined by a language model, is an appropriate method for predicting these assessments in different linguistic registers.