CodE Alltag: A German-Language E-Mail Corpus

CODE ALLTAG — A German-Language E-Mail Corpus Ulrike Krieg-Holz,1 Christian Schuschnig,1 Franz Matthies,2 Benjamin Redling,2 Udo Hahn2 1 Alpen-Adria-Universit¨at Klagenfurt, Universit¨atsstraße 65-67, A-9020 Klagenfurt, Austria 2 Friedrich-Schiller-Universit¨at Jena, F¨urstengraben 30, D-07743 Jena, Germany corresponding author: ulrike.krieg-holz@aau.at Abstract We introduce CODE ALLTAG, a text corpus composed of German-language e-mails. It is divided into two partitions: the first of these portions, CODE ALLTAGXL, consists of a bulk-size collection drawn from an openly accessible e-mail archive (roughly 1.5M e-mails), whereas the second portion, CODE ALLTAGS+d, is much smaller in size (less than thousand e-mails), yet excels with demographic data from each author of an e-mail. CODE ALLTAG, thus, currently constitutes the largest e-mail corpus ever built. In this paper, we describe, for both parts, the solicitation process for gathering e-mails, present descriptive statistical properties of the corpus, and, for CODE ALLTAGS+d, reveal a compilation of demographic features of the donors of e-mails. Keywords: text corpus, German language, e-mails 1. Introduction Much of the success of modern human language technol- ogy can be attributed to the provision of large amounts of linguistic data in terms of corpora of written or spoken lan- guage. With respect to written language, we observe two major streams of work. The first of these aims at build- ing large-sized national reference corpora as a ‘represen- tative’ mirror of authoritative, canonical language use for one national idiom. For the English language, e.g., an American (ANC) (Ide, 2009) and a British (BNC) (As- ton and Burnard, 1997) variant has been assembled, for the German language, currently two synchronous resources exist, DEREKO (Kupietz et al., 2010) and DWDS CORE (Geyken, 2007); as a diachronic extension going back un- til 950 AD, the DTA (Deutsches Text-Archiv) (Geyken and Gloning, 2015) is currently going to be set up as an historic complement. These reference corpora originally started as raw data collections aiming at high coverage, while in the meantime they have been continuously enriched by linguis- tic metadata, such as part-of-speech tags, lemma informa- tion and even parsing data (cf., e.g., Andersen et al. (2008)). The second stream of work is dominated by computational concerns in terms of (meta)language banking. NLP re- searchers took mainly newspaper/newswire material as a basis for in-depth annotation efforts, on a much smaller scale than the national reference corpora though, and built up value-added resources containing, e.g., extensive syn- tactic or semantic metadata associated with linguistic raw data. This approach is most prominently featured in the seminal work on the English PENN TREEBANK (Marcus et al., 1993) and PROPBANK (Palmer et al., 2005), respec- tively. For German, there are currently two major compet- ing syntactic annotation efforts, namely the TIGER (Brants et al., 2004) and the T ¨UBA-D/Z Treebank (Telljohann et al., 2004), whereas propositional information is primarily featured in the SALSA corpus which builds on the syntactic annotations of TIGER (Burchardt et al., 2006). In basically all these efforts related to national reference corpora or metadata banks, for German and other lan- guages, a preference for documents from …

Content truncated to 500 words