BRAGD is a part-of-speech tagger for Faroese, developed for grammatical annotation of Faroese text. The tagger assigns each word a detailed BRAGD tag that describes both its word class and relevant morphological features, such as gender, number, case, definiteness, degree, tense, mood and person.
The GitHub repository contains the code, data, training and evaluation scripts, results, and documentation for the BRAGD tagset. It can be used to train and run the tagger locally, reproduce the reported results, and further develop or retrain the model.
The released BRAGD model is also available on Hugging Face. That model used both the Sosialurin-BRAGD dataset and the OOD-BRAGD (out of domain) dataset for training. For a simple web-based demonstration, the tagger can be tested through the Marka demo.
Sosialurin-BRAGD: 117,322 word-level tokens (6,099 sentences)
OOD-BRAGD: approx. 10,000 tokens (500 sentences)
Release: 24.03.2026
Contact: mtd@setur.fo
