Projects
NLP · Computational Narratology
Narrative Deconstruction
Do classical discourse features alone reveal genre structure? An unsupervised UMAP + HDBSCAN study over 400 short stories — using pacing, emotional arc, and discourse rhythm, with no large language models.
ClassicalNMI = 0.366ClassicalARI = 0.093Classicalk = 27 clustersFull + EBNMI = 0.407Full + EBARI = 0.085
Story embedding space — coloured by genre
Each dot is one story. Position encodes structural similarity (pacing, emotional arc, discourse rhythm) — not topic or lexical similarity. Stories that “feel” structurally alike land close together.