CEFR-Cymraeg: A Dataset and Baseline Models for Welsh Language Proficiency Assessment

Welsh Government-funded project building the first CEFR-annotated dataset and baseline models for automatically assessing the proficiency level of Welsh texts.

Funder: Welsh Government
Period: April 2025 – March 2026
Role: Co-Investigator
Research theme: Welsh Language Technology & Multilingual NLP

Automated tools for assessing the proficiency level of Welsh learner texts have been lacking, limiting support for Welsh-language education and assessment. This project addressed that gap with two objectives:

1. Collect and annotate a comprehensive dataset of Welsh texts classified by CEFR level
2. Develop baseline machine learning models for predicting the CEFR level of Welsh texts

The result, CEFR-Cymraeg, is the first CEFR-annotated language proficiency dataset for Welsh (A1–B2), together with baseline classification models, enabling automated language proficiency assessment for Welsh learners.

Selected Publications

• CEFR-Cymraeg: A Dataset and Baseline Models for Language Proficiency Assessment in Welsh