This repository presents our fine-tuning experiments for metrical scansion of poetry in Galician, besides complementary experiments like tests with a symbolic system or with prompting.
The repository goes together with the following paper:
Ruiz Fabo, Pablo, Alonso Pérez, Anxo, Rodríguez Fernández, Pablo, Gamallo, Paulo. (2026). Automatic Metrical Scansion of Poetry in a Low-Resource Setting. To appear in LLMs4SSH Workshop @LREC 2026.
Repository structure:
01_corpus: Annotation guidelines, information about corpus formatting before fine-tuning, and inter-annotator agreement code02_prompting: Prompting experiments with GPT 5 (baseline)03_symbolic: Preprocessing + symbolic system (baseline)04_fine-tuning: Code for the fine-tuning experiments.05_analysis: Significance tests and code to obtain results for each meter and for lines with specific types of metaplasms
The fine-tuned models are on the project's Hugging Face group: https://huggingface.co/collections/compellit/scansion-models