sandrarrey/Bertinho_readability

This model is a fine-tuned version of Bertinho (bertinho-gl-base-cased) that classifies text readability into four levels in Galician (ILG/RAG spelling). It was trained on an in-house dataset that includes the iRead4Skills Dataset 1(https://zenodo.org/records/13127399), automatically translated into Galician using SalamandraTA (BSC-LT/salamandraTA-7b-instruct). It was then evaluated on a set of texts from the Corlega corpus (https://zenodo.org/records/15441342).

Evaluation on the test set: Accuracy: 0.55 Neighbor accuracy: 0.95 Precision: 0.56 Recall: 0.55 f1-score: 0.54

Published in the READIxTSAR @ LREC 2026 Workshop under the title 'A Comparative Study of Multilingual Fine-tuning and Prompting for Automatic Text Readability Classification in Galician' (Sandra Rodríguez Rey and Marcos Garcia).