Language Report Welsh

Abstract

In this chapter, based on Prys et al. (2022), an update to the META-NET White Paper (Evas 2014), we present Language Technology (LT) for the Welsh language, providing an overview of the status of Welsh in Wales and a summary of the Welsh writing system and typology. We describe key tools and our recommendations for Welsh LT and associated resource development.

SUBTLEX-CY: A new word frequency database for Welsh

“…Prys & Watkins, 2022) was used to obtain PoS information of each word in the subtitles. As mentioned earlier, the accuracy of this PoS tagger is much better than other Welsh PoS taggers (G. Prys & Watkins, 2022). However, the accuracy is lower than PoS taggers available for English, for example, Stanford CoreNL (Manning et al, 2014) and spaCy (Honnibal et al, 2020).…”

“…The TagTeg PoS tagger (G. Prys et al, 2020;G. Prys & Watkins, 2022) was used to obtain PoS information of each word in the subtitles.…”

“…To be able to calculate word counts based on the role that words play in sentences, the text corpus was processed with a part-of-speech (PoS) tagger that tokenizes the text and assigns a PoS to each token (e.g., noun, verb, punctuation). There are several PoS taggers for Welsh: WNLT2 3 (Welsh Natural Language Toolkit), CyTag, 4 and TagTeg 5 (G. Prys et al, 2020;G. Prys & Watkins, 2022).…”

“…WNLT2 and CyTag are rule-based PoS taggers, whereas TagTeg is a statistical tagger based on spaCy (Honnibal et al, 2020) and is trained using an annotated corpus. G. Prys and Watkins (2022) tested the accuracy of these PoS taggers using a corpus of 500 Welsh sentences (7,675 tokens). The results showed that TagTeg reached a token accuracy of 92%, which is significantly higher than the other two PoS taggers.…”