Universal Dependencies for Western Sierra Puebla Nahuatl

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5011–5020 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5011 Universal Dependencies for Western Sierra Puebla Nahuatl Robert Pugh†, Marivel Huerta Mendez∗, Mitsuya Sasaki∗, Francis M. Tyers† † Indiana University ∗ Independent {pughrob, ftyers}@iu.edu {marivelhm1, mitchara}@gmail.com Abstract We present a morpho-syntactically-annotated corpus of Western Sierra Puebla Nahuatl that conforms to the annotation guidelines of the Universal Dependencies project. We describe the sources of the texts that make up the corpus, the annotation process, and important annotation decisions made throughout the development of the corpus. As the first indigenous language of Mexico to be added to the Universal Dependencies project, this corpus offers a good opportunity to test and more clearly define annotation guidelines for the Mesoamerican linguistic area, spontaneous and elicited spoken data, and code-switching. Keywords: nahuatl, universal dependencies, treebank, syntax, morphology 1. Introduction Linguistically-annotated corpora are critical resources for natural language processing and computational lin- guistics. Statistical models for virtually all tasks in these areas, including word- and sentence-tokenization, mor- phological segmentation and analysis, and syntactic pars- ing, are commonly trained using collections of annotated text. Rules-based systems, as well, can leverage anno- tated corpora as ground-truth for performance evaluation or as a reference for rule development. The Universal Dependencies (UD) project1 (Nivre et al., 2016) is a widely-used annotation framework whose aim is to provide a consistent schema for morphological and syntactic phenomena for all of the world’s languages. An annotated UD corpus contains rich information for all aspects of a standard NLP pipeline including tokeniza- tion, part-of-speech tagging, and morphological and syn- tactic analysis, making it a highly valuable resource for the development of NLP applications. Since the annota- tion schema is intended to be language-independent, the resulting annotated corpora can (and in fact is intended to) be leveraged for multilingual NLP systems and cross- lingual transfer for downstream tasks. In descriptive lin- guistics research, questions about syntactic patterns and tendencies in a language can be approached quantita- tively with a large enough corpus (Kiss and Thomas, 2019; Tyers and Henderson, 2021). UD corpora are also useful for large-scale, multilingual corpus analysis (Naranjo and Becker, 2018; Levshina, 2019). Given UD’s goal of achieving consistent, cross- linguistically viable annotation guidelines for all of the world’s languages, it is crucial to prove out the existing guidelines on a diverse set of languages and domains. While the set of UD treebanks represents an impressive level of linguistic diversity, there are still a number of language families and linguistic areas that have yet to be analyzed with UD. Particularly relevant to our work, indigenous languages of Latin America are 1http://www.universaldependencies.org under-represented in UD both in terms of the number of languages within a treebank, and in terms of the sizes of the existing treebanks. With respect to the genre in existing treebanks, the overwhelming majority of the datasets represent written language, much of which is edited and polished (Müller-Eberstein et al., 2021). In …

Content truncated to 500 words