CyTag Neural Upgrade and DigiGrid Enhancement
Welsh Government-funded project developing a neural version of CyTag, the Welsh part-of-speech tagger, and deploying it as a reusable, open-source toolkit through DigiGrid.
Funder: Welsh Government
Period: April 2026 – March 2027
Role: Co-Investigator
Research theme: Welsh Language Technology & Multilingual NLP
Both CEFR-Cymraeg and Proffiliadur rely on CyTag, a high-quality, linguistically-informed, rule-based tagger developed as part of CorCenCC, for Welsh-specific tokenisation and part-of-speech tagging. CyTag remains one of the most accurate Welsh POS taggers available, but it is built on legacy rule-based technology that is difficult to maintain and extend, and hard to integrate with modern neural NLP libraries and deployment pipelines. Left unaddressed, this becomes a bottleneck for scaling Proffiliadur and future Welsh language tools.
This project modernises this core piece of Welsh NLP infrastructure. The objectives are to:
- Develop a neural version of CyTag that provides the tokenisation and POS tagging functionality required by Proffiliadur, compatible with modern NLP ecosystems
- Integrate the neural CyTag and an improved CEFR predictor into Proffiliadur, improving its robustness, maintainability, and scalability
- Deploy Proffiliadur and the neural CyTag as open-source, publicly accessible tools, and strengthen DigiGrid as a national platform for hosting Welsh language technology resources
The first version will prioritise the core tokenisation and tagging functionality that Proffiliadur needs, rather than fully replicating all of CyTag’s rule-based behaviour, so that a reliable and extensible replacement can be deployed quickly.
