Building a Universal Dependencies Treebank for Occitan

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2932–2939 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2932 Building a Universal Dependencies Treebank for Occitan Aleksandra Miletic*, Myriam Bras*, Marianne Vergez-Couret**, Louise Esher* Clamenc¸a Poujade*, Jean Sibille* *CNRS UMR 5263 CLLE-ERSS (University of Toulouse), **EA 3816 FoReLLIS (University of Poitiers) {aleksandra.miletic, myriam.bras, louise.esher, clamenca.poujade, jean.sibille}@univ-tlse2.fr marianne.vergez.couret@univ-poitiers.fr Abstract This paper outlines an ongoing effort to create the first treebank for Occitan, a low-resourced regional language spoken mainly in the south of France. We briefly present the global context of the project and report on its current status. We adopt the Universal Dependencies framework for corpus annotation. Our methodology is based on two main principles. Firstly, we rely on pre-processing using existing tools (taggers and parsers) to facilitate the work of human annotators, mainly through a delexicalized cross-lingual parsing approach. Secondly, we use agile annotation to ensure annotation quality. We present the results available at this point: annotation guidelines and an initial corpus annotated with PoS tags, lemmas and syntactic dependencies. Keywords: Occitan, treebank, Universal Dependencies, agile annotation, delexicalized cross-lingual parsing, low-resource lan- guages 1. Introduction and Background Low-resourced regional, non-official or minority languages often find themselves in a similar situation: building NLP resources requires substantial human and financial re- sources, but given their status and their often limited num- ber of speakers, investing in NLP research on these lan- guages is typically not seen as profitable. Nonetheless, these languages are part of the world’s cultural heritage and their preservation and study can shed significant light on various scientific questions, be it in theoretical or con- trastive linguistics, in linguistic typology, or in NLP itself. Luckily, this fact has been recognized both by the NLP community and by cultural institutions, leading to special- ized workshops and conferences, but also to greater finan- cial support from official sources. Occitan is one of the languages that has benefited from this paradigm shift. 1.1. Occitan Occitan is a Romance language spoken in a large area in the south of France, in several valleys in Italy and in the Aran valley in Spain. As illustrated in example (1), it shares nu- merous trademark traits of the Romance language family, such as: overt inflection for number and gender on all mem- bers of the NP; overt inflection for tense, aspect, mood, per- son and number on finite verbs; relatively free word order; and non-obligatory subject pronouns (Olivieri and Sauzet, 2016). As such, it is closer to Catalan, Spanish and Italian than to French or to regional languages of northern France. Like many other low-resourced languages, Occitan is not standardized. It has six varieties organized in dialec- tal groups (Auvernh`as, Gascon, Lengadocian, Lemosin, Provenc¸au and Vivaro-Aupenc). There is no universal spelling standard, but rather two different spelling norms, one called the classical, based on the Occitan troubadours’ medieval spelling, and the other closer to the French lan- guage conventions (Sibille, 2000). This diversity, which manifests itself on the …

Content truncated to 500 words