Unsmurfed: How an LLM Interprets the Smurfs' Distributional Language
(2026) Digital Humanities Benelux 2026 — Location: Maastricht (2.June.2026)
Files
unsmurfed_DHB2026_escouflaire.pdf
Open Access - Adobe PDF
- 2.74 MB
Details
- Authors
- Abstract
- (en) We use the French model CamemBERT to replace every occurrence of the word schtroumpf- ("smurf-") with its contextually most probable word in a corpus of 300 pages (five albums) of the Belgian Smurfs comics. Our multimodal pipeline consists of comic-centered OCR, image captioning and automated token prediction. By generating ten versions of all speech bubbles in about 3000 panels (top-1 to 10 most likely predictions), our experiment exposes how an LLM performs the very task that enables humans to naturally understand the Smurfs’ playful language: inferring meaning from context.
- Affiliations
Citations
Escouflaire, L. (2026, June 4). Unsmurfed: How an LLM Interprets the Smurfs’ Distributional Language. Digital Humanities Benelux 2026, Maastricht. https://doi.org/10.5281/zenodo.19235071