The Universal Dependencies Treebank of Spoken Slovenian

The Universal Dependencies Treebank of Spoken Slovenian Kaja Dobrovoljc1, Joakim Nivre2 1Institute for Applied Slovene Studies Trojina, Ljubljana, Slovenia 1Department of Slovenian Studies, Faculty of Arts, University of Ljubljana 2Department of Linguistics and Philology, Uppsala University kaja.dobrovoljc@trojina.si, joakim.nivre@lingfil.uu.se Abstract This paper presents the construction of an open-source dependency treebank of spoken Slovenian, the first syntactically annotated collection of spontaneous speech in Slovenian. The treebank has been manually annotated using the Universal Dependencies annotation scheme, a one-layer syntactic annotation scheme with a high degree of cross-modality, cross-framework and cross-language interoper- ability. In this original application of the scheme to spoken language transcripts, we address a wide spectrum of syntactic particularities in speech, either by extending the scope of application of existing universal labels or by proposing new speech-specific extensions. The initial analysis of the resulting treebank and its comparison with the written Slovenian UD treebank confirms significant syntactic differences between the two language modalities, with spoken data consisting of shorter and more elliptic sentences, less and simpler nominal phrases, and more relations marking disfluencies, interaction, deixis and modality. Keywords: dependency treebank, spontaneous speech, Universal Dependencies 1. Introduction It is nowadays a well-established fact that data-driven pars- ing systems used in different speech-processing applica- tions benefit from learning on annotated spoken data, rather than using models built on written language observation. Since the influential syntactic annotation of the Switch- board section of the Penn Treebank (Godfrey et al., 1992; Marcus et al., 1993), several syntactically annotated spoken corpora have emerged, such as the Verbmobil treebanks for English, German and Japanese (Hinrichs et al., 2000), the CGN treebank for Dutch (van der Wouden et al., 2002), the NoTa treebank for Norwegian (Johannessen and Jørgensen, 2006), the PDTSL treebank for Czech (Hajiˇc et al., 2008), and the Rhapsodie treebank for French (Lacheret et al., 2014). However, until now, no syntactically annotated data has been available for spoken Slovenian. In addition to differences in the underlying phrase-based or dependency-based grammatical formalisms, existing spo- ken treebanks vary considerably in their approach to anno- tation of syntactic particularities of spoken language, even though these are not generally considered as language- specific. On one side of the spectrum are annotation schemes providing syntactic analysis of all transcribed lex- ical tokens, typically by introduction of new labels for speech-specific phenomena, while on the other side we find schemes, in which only well-formed, written-like construc- tions are included in the resulting syntactic trees, disregard- ing disfluencies and other types of ’noisy’ structural partic- ularities. This prevalent multi-layer approach has partially been mo- tivated by the data-driven parsing systems themselves, usu- ally adopting a two-pass pipeline architecture, in which the structural particularities are first removed and followed by parsing of normalized transcriptions (Charniak and John- son, 2001). Recent advances in parsing systems using non- monotonic transition-based algorithms, however, show that joint treatment of disfluencies and other syntactic relations actually out-performs state-of-the-art pipeline approaches (Rasooli and Tetreault, 2013; Honnibal and Johnson, 2014). Such heterogeneity of spoken language annotation schemes inevitably leads to …

Content truncated to 500 words