Developing the Old Tibetan Treebank

Proceedings of Recent Advances in Natural Language Processing, pages 304–312, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_035 304 Developing the Old Tibetan Treebank Christian Faggionato SOAS, University of London cf36@soas.ac.uk Marieke Meelen University of Cambridge mm986@cam.ac.uk Abstract This paper presents a full procedure for the development of a segmented, POS-tagged and chunk-parsed corpus of Old Tibetan. As an extremely low- resource language, Old Tibetan poses non-trivial problems in every step to- wards the development of a search- able treebank. We demonstrate, how- ever, that a carefully developed, semi- supervised method of optimising and extending existing tools for Classical Tibetan, as well as creating specific ones for Old Tibetan, can address these issues. We thus also present the very first Tibetan Treebank in a variety of formats to facilitate research in the fields of NLP, historical linguistics and Tibetan Studies. 1 Introduction In historical linguistics, there are currently two types of morpho-syntactically annotated cor- pora or ‘Treebanks’, one based on constituency parses, the other on dependency parses. The former includes pioneering work in the Penn- Helsinki tradition, resulting in the Old and Middle English (Taylor and Kroch, 1994), Ice- landic (IcePaHC) (Rögnvaldsson et al., 2012) and Portuguese (Tycho Brahe) (Galves, 2018) treebanks. The latter represents syntactic structure in the form of dependencies and is often used for applied NLP tasks and early Indo-European languages, e.g. the PROIEL Treebank family (Eckhoff et al., 2018). In this paper we present a constituency-based tree- bank for an under-resourced language: Old Tibetan (see Green et al. (2012) and El-Haj et al. (2015) for similar examples of creating under-resourced treebanks). Old Tibetan is an extremely under- resourced and under-researched language from an NLP point of view. We chose to focus our attention on the Old Tibetan corpus (7- 11th c.) since it consists of a small col- lection of documents compared to the vast amounts of translated and original Classical Tibetan texts. Nonetheless, the Old Tibetan corpus is still heterogeneous enough to repre- sent natural language. The majority of Old Ti- betan texts (known to date; new inscriptions and texts are still being discovered) has now been digitised in one way or another (images, OCR and/or transcribed) and annotating this data is fundamental for the understanding of diachronic and synchronic issues in Tibeto- Burman languages. As the first attempt to create a Tibetan Treebank, developing a seg- mented, POS-tagged and chunk-parsed corpus of Old Tibetan provides new opportunities in Tibetan scholarly history, literature and lin- guistics. Old Tibetan was the language spoken in the Yarlung Valley from where the Tibetan em- pire started its initial expansion. Writing was mainly introduced to facilitate administrative tasks, and the earliest Old Tibetan texts rep- resent the most detailed sources for the history of early Tibet (Hill, 2010). The earliest cur- rently available, securely datable Old Tibetan document dates to ca. 763 CE. However, the digital resources for Old Tibetan are inade- quate (problematic transcriptions, translitera- tions and no digitised secondary resources such as dictionaries, etc.). The core of the Old …

Content truncated to 500 words