Introducing the Welsh Text Summarisation Dataset and Baseline Systems

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5097–5106 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5097 Introducing the Welsh Text Summarisation Dataset and Baseline Systems Ignatius Ezeani1, Mahmoud El-Haj1, Jonathan Morris2 and Dawn Knight2 1UCREL NLP group, School of Computing and Communications, Lancaster University, 2Cardiff University {i.ezeani, m.el-haj}@lancaster.ac.uk, {knightd5, morrisj17}@cardiff.ac.uk Abstract Welsh is an official language in Wales and is spoken by an estimated 884,300 people (29.2% of the population of Wales). Despite this status and estimated increase in speaker numbers since the last (2011) census, Welsh remains a minority language undergoing revitalisation and promotion by Welsh Government and relevant stakeholders. As part of the effort to increase the availability of Welsh digital technology, this paper introduces the first Welsh summarisation dataset, which we provide freely for research purposes to help advance the work on Welsh text summarisation. The dataset was created by Welsh speakers through manually summarising Welsh Wikipedia articles. In addition, the paper discusses the implementation and evaluation of different summarisation systems for Welsh. The summarisation systems and results will serve as benchmarks for the development of summarisers in other minority language contexts. Keywords: summarisation, Welsh, corpus, word embeddings 1. Introduction It is estimated that over a quarter (29.2%) of the popu- lation in Wales aged over 3 consider themselves to be Welsh speakers 1. Although this estimate represents an increase in the proportion of the population who re- ported speaking Welsh at the last (2011) census2, his- torically the language has been in decline and repre- sents a minority language in Wales despite having of- ficial status. This decline has led to the development of language policy designed to safeguard the language and promote its use among the population (Carlin and Chríost, 2016). The most recent Welsh Government strategy for the revitalisation of Welsh has infrastructure (and particu- larly digital infrastructure) as a main theme along with increasing the number of speakers and increasing lan- guage use3. The aim is to ensure that the Welsh lan- guage is at the heart of innovation in digital technology to enable the use of Welsh in all digital contexts (Welsh Government 2017: 71). A system that could assist in the automatic summarisation of long documents would prove beneficial to the culture revitalisation efforts cur- rently taking place. Over time, there have been various approaches to au- tomatic text summarisation, but when looking at those approaches in detail, we can see that they are mainly split between single-document summarisation (finding the most informative sentences in a document) and multi-document summarisation (finding a summary that combines the main themes across thematically diverse set of documents) with the majority of work 1https://gov.wales/welsh-language-data-annual- population-survey-july-2020-june-2021 2https://statswales.gov.wales/Catalogue/Welsh- Language/Census-Welsh-Language 3Welsh Government: Cymraeg 2050 - A million Welsh speakers: https://gov.wales/sites/ default/files/publications/2018-12/ cymraeg-2050-welsh-language-strategy.pdf being applied to the English language, as a global lingua franca (Goldstein et al., 2000; Svore et al., 2007; Svore et al., 2007; Litvak and Last, 2008; El-Haj et al., 2011; El-Haj and Rayson, …

Content truncated to 500 words