📄 Corpas Náisiúnta Na Gaeilge 2022-2029: A Project Overview

Summary

Irish is a low-resource language, and both researchers and the Irish-speaking community have lacked a large, balanced reference corpus. This overview from the Gaois research group at DCU describes Corpas Náisiúnta na Gaeilge (CNG), a national corpus of Irish covering 2000–2024. It combines edited print and online writing with spoken material from broadcasting, lectures, songs and stage drama. In Phase 1 (2022–2024), the team built the main corpus and three companion collections. They also launched the corpas.ie concordancing site, which runs on NoSketchEngine, and trained a part-of-speech tagger with UDPipe and the PAROLE tagset for Irish. Phase 2 (2025) enlarged the monitor and spoken corpora and refreshed the tagging and frequency data. Phase 3 (2026–2029) is set to bring in legislative and historical material. The authors also describe the difficulty of securing licensed texts from publishers and news agencies, a copyright-aware data-sharing policy, and their teaching and outreach work. The project gives Irish a lasting and legally careful infrastructure for linguistic and NLP research.

Key points

• Phase 1 corpus sizes:
  • CNG: 101 million words.
  • CGS (written, editorially processed texts only): 131 million words.
  • CGL (spoken): 9 million words.
  • CMG (monitor): about 1 million words per year.
• Phase 2 added 1 million words to CMG for 2025 and 1 million words to CGL. It also updated the n-gram and word-frequency lists and corrected tagging across all corpora. These outputs went live on the website in Q1 2026.
• The paper's table of outputs is garbled in the extracted text, but these items appear to be planned for Phase 3:
  • an English–Irish parallel corpus of Irish and EU legislation (69.3m Irish, 63.7m English words);
  • four more years of CMG data (4 million words);
  • the Bailiúchán na Scol (1937–1939) collection (10 million words).
• Sharing is limited by copyright:
  • Word embeddings and n-grams are published on the website.
  • Bona fide researchers can request sentence-shuffled subsets, shuffled at least 1,000 times.
  • Spoken data and creative works are excluded from these subsets.
• Training and outreach include:
  • the mandatory 10 ECTS MA module "Taighde Corpais";
  • a project launch in November 2024;
  • a conference on Corpus Linguistics for the Languages of Ireland in November 2025, with 19 research presentations;
  • involvement in CLARIN-ERIC.