Universal Dependencies for Norwegian

Universal Dependencies for Norwegian Lilja Øvrelid, Petter Hohle Department of Informatics, University of Oslo {liljao,pettehoh}@ifi.uio.no Abstract This article describes the conversion of the Norwegian Dependency Treebank to the Universal Dependencies scheme. This paper details the mapping of PoS tags, morphological features and dependency relations and provides a description of the structural changes made to NDT analyses in order to make it compliant with the UD guidelines. We further present PoS tagging and dependency parsing experiments which report first results for the processing of the converted treebank. The full converted treebank was made available with the 1.2 release of the UD treebanks. Keywords: Dependency treebanks, Universal Dependencies, dependency parsing 1. Introduction With the increasing popularity of dependency-based repre- sentations of syntactic structure in recent years, a wealth of different dependency annotation schemes have surfaced. It has been shown that the choice of dependency scheme influences parsing results (Schwartz et al., 2012) as well as downstream applications (Elming et al., 2013) and even though attempts have been made to contrast differ- ent schemes theoretically (Ivanova et al., 2012), it is clear that the diversity of representation makes comparisons dif- ficult. Cross-linguistically even more so, and it can often be difficult to tease apart aspects of annotation scheme from typological differences in cross-lingual learning (Søgaard, 2011; Skjærholt and Øvrelid, 2012). Universal Dependencies (UD) (de Marneffe et al., 2014; Nivre, 2015) is a recent community-driven effort to cre- ate cross-linguistically consistent syntactic annotation. UD is based on the Stanford dependency scheme (de Marneffe et al., 2006) which has become a widely used dependency scheme for English in recent years. A number of existing dependency treebanks have been converted to UD (Pyysalo et al., 2015; Nivre, 2014) and new data has also been an- notated from scratch in order to enable multilingual parser development, cross-lingual learning and typological stud- ies of syntactic structure. Treebanks involved in this effort represent a diverse range of languages such as English, Ger- man, Swedish, Spanish, Italian, Persian, Japanese, and the UD release 1.2 contains treebanks for as many as 33 differ- ent languages of varying sizes. This paper describes a fully automatic conversion proce- dure for the Norwegian Dependency Treebank (NDT) to UD. Due to differences both in the tag set, as well as struc- tural analyses, the conversion requires non-trivial transfor- mations of the dependency trees, in addition to mappings of tags and labels that make reference to a combination of various kinds of linguistic information. This paper details the mapping of PoS tags, morphological features and de- pendency relations and provides a description of the struc- tural changes made to NDT analyses in order to make it compliant with the UD guidelines. We further present PoS- tagging and dependency parsing experiments which con- trast the performance of a state-of-the-art PoS-tagger and dependency parser on the original and converted treebank. Head Dependent Finite verb Complementizer Finite auxiliary Lexical verb Infinitival marker Lexical verb Preposition Prepositional complement Noun Determiner First conjunct Subsequent conjuncts Table 1: Annotation choices in the …

Content truncated to 500 words