gaHealth: An English–Irish Bilingual Corpus of Health Data

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6753–6758 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6753 gaHealth: An English–Irish Bilingual Corpus of Health Data S´eamus Lankford 1, Haithem Afli 1, ´Orla N´ı Loinsigh 2, Andy Way 2 1 Department of Computer Science, Munster Technological University, Cork, Ireland. 2 Adapt Centre, Dublin City University, Dublin, Ireland {seamus.lankford, haithem.afli}@mtu.ie {andy.way, orla.niloinsigh}@adaptcentre.ie Abstract Machine Translation is a mature technology for many high-resource language pairs. However in the context of low-resource languages, there is a paucity of parallel data datasets available for developing translation models. Furthermore, the development of datasets for low-resource languages often focuses on simply creating the largest possible dataset for generic translation. The benefits and development of smaller in-domain datasets can easily be overlooked. To assess the merits of using in-domain data, a dataset for the specific domain of health was developed for the low-resource English to Irish language pair. Our study outlines the process used in developing the corpus and empirically demonstrates the benefits of using an in-domain dataset for the health domain. In the context of translating health-related data, models developed using the gaHealth corpus demonstrated a maximum BLEU score improvement of 22.2 points (40%) when compared with top performing models from the LoResMT2021 Shared Task. Furthermore, we define linguistic guidelines for developing gaHealth, the first bilingual corpus of health data for the Irish language, which we hope will be of use to other creators of low-resource data sets. gaHealth is now freely available online and is ready to be explored for further research. Keywords: Health data, parallel corpus, machine translation, Irish 1. Introduction Improvements in performance in natural language process- ing (NLP) tasks are typically to be seen when deep learning models are used. However, deep learning requires large amounts of data for model training. Consequently, the availability of large amounts of textual data has become fundamental to the success of NLP applications, such as language modelling (Buck et al., 2014) and Neural Ma- chine Translation (NMT) (Bahdanau et al., 2014; Sennrich et al., 2016). A popular method of developing such corpora for Machine Translation (MT) tasks is to crawl and parse bilingual web pages to general parallel corpora. However, given the na- ture of low-resource languages, there are often insufficient web sites available in both the languages of study. Accord- ingly, a lack of web content typically hinders the develop- ment of NLP applications for low-resource languages. The motivation underpinning our present work comes from the challenges we faced in developing high-performing MT models in low-resource settings (Afli et al., 2017). In this work, we developed the first bilingual corpus of health data for the English-Irish pair. A procedure was created to ex- tract, clean and select appropriate sentences to build a bilin- gual corpus. In addition, we built a high-performing MT model for translating in-domain health data. 2. Related work 2.1. Transformer Transformer (Vaswani et al., 2017) is an architecture for transforming an input …

Content truncated to 500 words