Corpus REDEWIEDERGABE

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 803–812 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 803 Corpus REDEWIEDERGABE Annelen Brunner1, Stefan Engelberg1, Fotis Jannidis2, Ngoc Duyen Tanja Tu1, Lukas Weimer2 1Leibniz-Institut für Deutsche Sprache Mannheim; 2Julius-Maximilians-Universität Würzburg 1R5 6-13, 68161 Mannheim, Germany; 2Am Hubland, 97074 Würzburg, Germany {brunner, engelberg, tu}@ids-mannheim.de; {fotis.jannidis, lukas.weimer}@uni-wuerzburg.de Abstract This article presents the corpus REDEWIEDERGABE, a German-language historical corpus with detailed annotations for speech, thought and writing representation (ST&WR). With approximately 490,000 tokens, it is the largest resource of its kind. It can be used to answer literary and linguistic research questions and serve as training material for machine learning. This paper describes the composition of the corpus and the annotation structure, discusses some methodological decisions and gives basic statistics about the forms of ST&WR found in this corpus. Keywords: corpus, annotation, speech thought writing representation, machine learning 1. Introduction and Motivation The corpus REDEWIEDERGABE is a historical corpus of fictional and non-fictional German texts from 1840 to 1919 annotated with various forms of speech, thought and writing representation (ST&WR). It was created by the project ‘Redewiedergabe’ (www.redewiedergabe.de, funded by the Deutsche Forschungsgemeinschaft). The corpus can be used for quantitative literary and linguistic studies (e.g. about the development of ST&WR forms, differences between fiction and non-fiction) and also serve as training material for the development of automatic recognizers for ST&WR. The use of the corpus as training material for machine learning was a main focus of the project ‘Redewiedergabe’, which influenced some aspects of its design and structure. This paper describes the composition of the corpus and the annotation structure, discusses some methodological decisions and gives basic statistics about the forms of ST&WR found in this corpus. 2. Related Work ST&WR has been extensively researched in German linguistics and literary studies. In linguistics, different types of representation are described structurally (e.g. Fabricius-Hansen, 2002; Weinrich, 2007) and specific linguistic features have been studied, especially the use of subjunctive verb mode in the context of representation (e.g. Zifonun, Hoffmann and Strecker, 2011; Fabricius- Hansen, Solfjeld and Pitz, 2018) but also other aspects like verb selection in framing phrases (Hauser, 2008; Tu, Engelberg and Weimer, 2020). In literary studies, different modes of representation for a fictional character’s voice are a central part of many narrative theories (e.g. Stanzel, 2008; Genette, 2010; Leech and Short, 2013; Martinez and Scheffel, 2016). Some studies focus on a particular aspect such as the representation of thoughts (e.g. Cohn, 1978; Palmer, 2004) or free indirect ST&WR in particular (e.g. Banfield, 1982; Fludernik, 1993) which received much attention in literary studies. Our corpus strives to capture the phenomenon ST&WR in a structured way by systematic annotation. A comparable 1 https://textgrid.de/digitale-bibliothek. 2 https://repos.ids-mannheim.de/mkhz-beschreibung.html. effort is the corpus by Semino and Short (2004) who annotated a corpus of modern English texts according to the ST&WR schema defined by Leech and Short (1981). A direct predecessor of our corpus is Brunner (2015), a corpus of 13 …

Content truncated to 500 words