The Potsdam Commentary Corpus 2.2: Extending Annotations for Shallow Discourse Parsing

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1061–1066 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1061 The Potsdam Commentary Corpus 2.2: Extending Annotations for Shallow Discourse Parsing Peter Bourgonje, Manfred Stede Universit¨at Potsdam, Applied Computational Linguistics Potsdam, Germany {bourgonje, stede}@uni-potsdam.de Abstract We present the Potsdam Commentary Corpus 2.2, a German corpus of news editorials annotated on several different levels. New in the 2.2 version of the corpus are two additional annotation layers for coherence relations following the Penn Discourse TreeBank framework. Specifically, we add relation senses to an already existing layer of discourse connectives and their arguments, and we introduce a new layer with additional coherence relation types, resulting in a German corpus that mirrors the annotation scheme of the PDTB (which is a much larger corpus, though). The aim of the extended annotations is to improve usability of the corpus for the task of shallow discourse parsing. In this paper, we provide inter-annotator agreement figures for the new annotations and compare corpus statistics based on the new annotations to the equivalent statistics extracted from the PDTB. Keywords: Discourse Relations, Discourse Parsing, Corpus Annotation 1. Introduction The Potsdam Commentary Corpus (PCC) was first intro- duced by Stede (2004) as a collection of 175 newspaper editorials. It comprises over 34k words in over 2,100 sen- tences, sourced from a German regional newspaper. The main idea behind the PCC was to provide a single-genre corpus, collected in an ”unbalanced” way, to specifically address research questions on subjectivity and argumenta- tion. It was manually annotated on several different levels independently, allowing investigation of different linguistic phenomena related to subjectivity and argumentation and the ways they interact on syntactic, semantic and discourse level. As such, the original version was annotated for sen- tence syntax, coreference, and rhetorical structure. In Stede and Neumann (2014), an updated version (PCC 2.0) was presented. In addition to revisions on the rhetorical struc- ture and coreference layers, it introduced a new layer con- sisting of discourse connectives and their arguments. The annotation scheme for this layer was based on that of the Penn Discourse TreeBank (PDTB) (Prasad et al., 2008). In contrast to the PDTB however, the connectives and argu- ments layer of the PCC 2.0 contained instances of explicit relations only, and also it did not contain the senses for its explicit relations. More recently, Bourgonje and Stede (2018b) introduced the PCC 2.1, in which a new layer con- taining information-structural aboutness topics and auto- matically produced dependency parses using the Universal Dependencies scheme were added. As part of the 2.1 re- lease, the entire corpus was made available in the interac- tive corpus browser ANNIS3 (Krause and Zeldes, 2016). In this paper, we present the PCC 2.2. In addition to a re- vision of the connectives and arguments layer, resulting in correction of some minor inconsistencies, and a conversion from inline XML to standoff XML format (more conve- nient for automatic processing), the key contributions …

Content truncated to 500 words