TwittIrish: A Universal Dependencies Treebank of Tweets in Modern Irish

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 6869 - 6884 May 22-27, 2022 c⃝2022 Association for Computational Linguistics TwittIrish: A Universal Dependencies Treebank of Tweets in Modern Irish Lauren Cassidy1, Teresa Lynn1, James Barry1, Jennifer Foster2 1,2 School of Computing, Dublin City University 1 ADAPT Centre 1 {lauren.cassidy, teresa.lynn, james.barry}@adaptcentre.ie 2 {jennifer.foster}@dcu.ie Abstract Modern Irish is a minority language lacking sufficient computational resources for the task of accurate automatic syntactic parsing of user- generated content such as tweets. Although language technology for the Irish language has been developing in recent years, these tools tend to perform poorly on user-generated con- tent. As with other languages, the linguistic style observed in Irish tweets differs, in terms of orthography, lexicon, and syntax, from that of standard texts more commonly used for the development of language models and parsers. We release the first Universal Dependencies treebank of Irish tweets, facilitating natural lan- guage processing of user-generated content in Irish. In this paper, we explore the differences between Irish tweets and standard Irish text, and the challenges associated with dependency parsing of Irish tweets. We describe our boot- strapping method of treebank development and report on preliminary parsing experiments. 1 Introduction Irish is a minority language spoken mostly in small communities in Ireland called ‘Gaeltachtaí’ (CSO, 2016) but social media sites, such as Twitter, pro- vide a platform for Irish speakers to communicate electronically from any location. Users may reach a wide audience quickly, unconstrained by the con- ventions of standard language upheld by editors in publications, revealing the orthographic, lexical, and syntactic variation abundant in informal Irish. Analysis of up-to-date, real-world language data can provide an insight into how Irish is used in everyday communication and how such informal texts compare to prescriptive norms of standardised language to which published texts tend to adhere. User-generated content (UGC), such as tweets, is a valuable, highly available resource for training syntactic parsers that can accurately process social media text. UGC is a genre with features different from those of both spoken language and standard- ised written language more traditionally found in natural language processing (NLP) corpora. Plank (2016) notes the advantages of utilising fortuitous data in order to create more adaptive, robust lan- guage technology. Given that the accuracy of syntactic parsing tools has been shown to decline when evaluated on noisy UGC data (Foster et al., 2011; Seddah et al., 2012) and that domain1 adaptation has been shown to improve parser performance for dependency anno- tation of English tweets (Kong et al., 2014) and POS-tagging in Irish tweets (Lynn et al., 2015), the need for genre-specific resources is clear in order to reliably process this variety of data. The prereq- uisite, therefore, for research in this area is a data set of Irish UGC. This research attempts to fill this gap through the development of TwittIrish, a tree- bank of Irish tweets, within Universal Dependen- cies (UD) (Nivre et al., 2020), a cross-lingually con- …

Content truncated to 500 words