Universal Dependencies for Turkish

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 3444–3454, Osaka, Japan, December 11-17 2016. Universal Dependencies for Turkish Umut Sulubacak ⋆ and Memduh G¨okırmak † Department of Computer Engineering Istanbul Technical University 34469 Istanbul, Turkey {⋆sulubacak | †gokirmak}@itu.edu.tr Francis M. Tyers HSL-fakultehta UiT Norgga ´arktalaˇs universitehta N-9018 Tromsø, Norway francis.tyers@uit.no C¸ a˘grı C¸ ¨oltekin Department of Linguistics University of T¨ubingen 72074 T¨ubingen, Germany ccoltekin@sfs.uni-tuebingen.de Joakim Nivre Department of Linguistics and Philology Uppsala University 75126 Uppsala, Sweden joakim.nivre@lingfil.uu.se G¨uls¸en Eryi˘git Department of Computer Engineering Istanbul Technical University 34469 Istanbul, Turkey gulsen.cebiroglu@itu.edu.tr Abstract The Universal Dependencies (UD) project was conceived after the substantial recent interest in unifying annotation schemes across languages. With its own annotation principles and abstract inventory for parts of speech, morphosyntactic features and dependency relations, UD aims to facilitate multilingual parser development, cross-lingual learning, and parsing research from a language typology perspective. This paper presents the Turkish IMST-UD Treebank, the first Turkish treebank to be in a UD release. The IMST-UD Treebank was automatically converted from the IMST Treebank, which was also recently released. We describe this conversion pro- cedure in detail, complete with mapping tables. We also present our evaluation of the parsing performances of both versions of the IMST Treebank. Our findings suggest that the UD frame- work is at least as viable for Turkish as the original annotation framework of the IMST Treebank. 1 Introduction The Universal Dependencies (UD)1 project is an international collaborative project to make cross-lin- guistically consistent treebanks available for a wide variety of languages. Currently in version 1.3, the UD project covers 40 languages, including two Turkic languages: Kazakh, which was annotated from scratch, and Turkish, the creation of which is described in this paper. The universal annotation guidelines of UD are based on the Google Universal Part-of-Speech Tagset (Petrov et al., 2012) for parts of speech, the Interset framework (Zeman, 2008) for morphologi- cal features, and Stanford Dependencies (De Marneffe et al., 2006; Tsarfaty, 2013; De Marneffe et al., 2014) for dependency relations. The objective of harmonizing annotation guidelines as far as possible is to make comparison of parsing results and investigating cross-linguistic methods across languages easier. This is achieved by a number of principles, including the primacy of content words, distinguishing core arguments from modifiers and distinguishing clausal constituents from nominals. The IMST-UD Treebank was first released in UD version 1.3 and became the first Turkish tree- bank to be included in a UD release. The treebank was created by automatic conversion of the IMST This work is licenced under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/ 1http://universaldependencies.org/ 3444 Treebank (Sulubacak et al., 2016), which is itself a reannotation of the METU-Sabancı Turkish Tree- bank (Oflazer et al., 2003; Atalay et al., 2003). Although the annotation framework of the IMST Tree- bank was revised, it is still fundamentally similar to that of the METU-Sabancı Treebank and radically different from the UD framework in both morphology and syntax. In this paper, …

Content truncated to 500 words