At the Hybrid Ecologies Lab, I researched if it was possible to teach machines creative intelligence. Could they learn subjective data such as writing style?
Every writer has a linguistic fingerprint, one that can be qualified in terms of their diction, flow, and structure. These qualities are the domain of stylometry. Our research goal was to craft a creative intelligence interface to study human-computer interactions with writing. What if we had search and recommendation engines that would traverse and suggest based on writing style rather than content?
To tackle this digital humanities and human-computer interaction question, we used techniques from research in natural language processing (NLP), machine learning, human-computer interaction, and stylometry. I engineered features for the literature data, web scraped data sets for training, and visualized data in 2D and hyperdimensional space.
The images above visualizes some of the output of the technical approaches I took. In the bottom right image, each row represents different excerpts from the same author, and one can see that there is a distinct difference between a Paulo Coelho versus a F. Scott Fitzgerald.
The table below summarizes my technical and design research approach.