Universal Dependencies for Punjabi

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5705–5711 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5705 Universal Dependencies for Punjabi Aryaman Arora Georgetown University Washington, D.C., USA aa2190@georgetown.edu Abstract We introduce the first Universal Dependencies treebank for Punjabi (written in the Gurmukhi script) and discuss corpus design and linguistic phenomena encountered in annotation. The treebank covers a variety of genres and has been annotated for POS tags, dependency relations, and graph-based Enhanced Dependencies. We aim to expand the diversity of coverage of Indo-Aryan languages in UD. Keywords: Punjabi, Universal Dependencies, syntax, POS tagging, treebanking 1. Introduction Universal Dependencies (UD) is a community project that maintains a standard scheme for the annotation of grammar (including part-of-speech tags, syntactic relations, and mor- phological features) in a cross-lingually consistent manner, and releases treebanks for many languages annotated with the UD schema (Nivre et al., 2016, 2020). UD is currently on its version 2.9 release, comprised of 217 treebanks cov- ering 122 languages. However, UD coverage is biased towards high-resource languages in NLP, especially in South Asia where only Hindi and Urdu have treebanks with greater than 100k to- kens. South Asian languages with tens of millions of speak- ers (e.g. Punjabi, Kannada, Gujarati, Pashto) often do not have treebanks yet. UD treebanked data is useful for up- stream tasks in NLP, including semantic parsing (Reddy et al., 2017) and natural language understanding (Schuster and Manning, 2016), and syntactic corpora are useful for linguists studying language typology and change (Levshina, 2019). In this paper, we present a new UD treebank for Pun- jabi, a language of South Asia with over 100 million native speakers but otherwise underesearched in computational linguistics and NLP. Punjabi has official status at the re- gional level in both India and Pakistan. We describe the composition of the corpus, thorny linguistic issues that are relevant for UD annotation of other South Asian languages, and future plans for Punjabi NLP. We especially explore the annotation of complex syntactic phenomena which are relevant to many other languages. 2. Related treebanks The Indo-Aryan (IA) languages are unevenly represented in UD, with large automatically-converted (and thus of imperfect quality) treebanks for the two major IA lan- guages Hindi (Tandon et al., 2016) and Urdu (Ehsan and Butt, 2020), and a smattering of smaller manual treebanks for lesser-studied languages such as Marathi (Ravishankar, 2017), Bhojpuri (Ojha and Zeman, 2020), Sanskrit (Hell- wig et al., 2020; Dwivedi and Zeman, 2017), Ashokan Prakrit (Farris and Arora, 2021), and code-mixed Hindi– English (Bhat et al., 2018). Besides UD, there is a substan- tial body of work on syntactic annotation using Paninian Genre Doc. Sent. Tok. misc — 71 1,664 news 4 89 1,614 editorial 2 80 1,570 blog 1 33 806 nonfiction 1 26 566 Total 7 299 6,220 Table 1: Data in the Punjabi UD corpus by genre. Columns are ‘documents’, ‘sentences’, and ‘tokens’. karaka formalisms for many South Asian languages (Bhatt et al., 2009; …

Content truncated to 500 words