A Diachronic Treebank of Russian Spanning More Than a Thousand Years

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5251–5256 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5251 A Diachronic Treebank of Russian Spanning More Than a Thousand Years Aleksandrs Berdicevskis1, Hanne Eckhoff2 1Språkbanken (The Swedish Language Bank), Universify of Gothenburg 2Faculty of Medieval and Modern Languages, University of Oxford aleksandrs.berdicevskis@gu.se, hanne.eckhoff@mod-langs.ox.ac.uk Abstract We describe the Tromsø Old Russian and Old Church Slavonic Treebank (TOROT) that spans from the earliest Old Church Slavonic to modern Russian texts, covering more than a thousand years of continuous language history. We focus on the latest additions to the treebank, first of all, the modern subcorpus that was created by a high-quality conversion of the existing treebank of contemporary standard Russian (SynTagRus). Keywords: Russian, Slavonic, Old Church Slavonic, treebank, diachrony, conversion 1. Introduction The Tromsø Old Russian and OCS Treebank (TOROT, Eckhoff and Berdicevskis 2015) has been available in various releases since its beginnings in 2013, and its East Slavonic part now contains approximately 230K words.1 This paper describes the TOROT 20200116 release,2 which adds a conversion of the SynTagRus treebank. Former TOROT releases consisted of Old East Slavonic and Middle Russian texts from the 11th–19th century, covering a broad range of text types and genres such as chronicles, lives of saints, charters, birchbark letters, personal correspondence and tales. Now it also includes contemporary standard Russian. The new release thus turns it into a treebank covering every attested stage of Russian (contemporary standard Russian, Middle Russian, Old East Slavonic) and Old Church Slavonic, spanning more than a thousand years. This paper is structured as follows: Section 2 discusses the TOROT treebank design. Section 3 is a description of SynTagRus. Section 4 describes the conversion, Section 5 concludes. For detailed examples, we will be using tables with a format similar to simplified CONLL: id, Russian form, English gloss, head, relation. For SynTagRus relations, we will be using our translations of their original Russian names into English. 2. TOROT 2.1 The PROIEL dependency format TOROT is a part of the PROIEL family of ancient Indo- European treebanks (Eckhoff et al. 2018) and is annotated using the PROIEL enhanced dependency format (Haug and Jøhndal 2008). The format was designed specifically for the needs of ancient Indo-European languages, which are characterised by rich inflection and relatively free word order. The scheme differs from more classical approaches to dependency grammar, such as the Prague Dependency Treebank scheme (Hajič et al. 2018), in three main ways: use of secondary dependencies, limited use of empty nodes 1 Word count here is not the same as token count (which is more traditional in NLP), since word count does not include punctuation marks. and a richer inventory of syntactic relation labels. Secondary dependencies are used to indicate external subjects, for example in control structures, and to indicate shared dependents, for example in structures with coordinated verbs. Empty nodes are allowed to give more information on elliptic structures and asyndetic coordination. The empty nodes …

Content truncated to 500 words