FOLK-Gold ― A Gold Standard for Part-of-Speech-Tagging of Spoken German

FOLK-Gold – A GOLD standard for Part-of-Speech Tagging of Spoken German Swantje Westpfahl, Thomas Schmidt Institut für Deutsche Sprache R5, 6-13 68181 Mannheim, Germany E-mail: westpfahl@ids-mannheim.de, thomas.schmidt@ids-mannheim.de Abstract In this paper, we present a GOLD standard of part-of-speech tagged transcripts of spoken German. The GOLD standard data consists of four annotation layers – transcription (modified orthography), normalization (standard orthography), lemmatization and POS tags – all of which have undergone careful manual quality control. It comes with guidelines for the manual POS annotation of transcripts of German spoken data and an extended version of the STTS (Stuttgart Tübingen Tagset) which accounts for phenomena typically found in spontaneous spoken German. The GOLD standard was developed on the basis of the Research and Teaching Corpus of Spoken German, FOLK, and is, to our knowledge, the first such dataset based on a wide variety of spontaneous and authentic interaction types. It can be used as a basis for further development of language technology and corpus linguistic applications for German spoken language. Keywords: German spoken language, POS tagging, GOLD standard 1. Introduction In this paper, we present a GOLD standard of part-of-speech-tagged transcripts of spoken German. It comes with guidelines for the manual annotation of transcripts of spoken data and an extended version of the STTS (Stuttgart Tübingen Tagset) which accounts for phenomena typically found in spontaneous spoken German. The GOLD standard was developed on the basis of the Research and Teaching Corpus of Spoken German, FOLK (Schmidt 2014), and is thus, to our knowledge, the first such dataset based on a wide variety of spontaneous and authentic interaction types. The GOLD standard will be used in the FOLK project itself to further improve the part-of-speech tagging of the corpus. It will also be made available to the research community as a resource for the development of language technology for spontaneous interaction data. The paper is structured as follows: section 2 briefly recapitulates related work, section 3 outlines the method by which data for the GOLD standard was sampled and manually annotated. Section 4 provides some technical information about the format of the GOLD standard and on how it will be made available. Section 5 gives an outlook on further work which we plan to carry out on its basis. 2. Related Work There are several corpora for German spoken data: The Berlin Map Task Corpus (Sauer 2015), the Hamburg Map Task Corpus (HZSK 2010), GeWiss (Gesprochene Wissenschaftssprache - corpus of spoken academic language) (Fandrych et al. 2014), KiDKo (KiezDeutschKorpus - corpus of German youth language) (Rehbein et al. 2014), Tüba-D/S (Tübinger Baumbank des Deutschen/Spontansprache - Treebank of German spoken language) (Universität Tübingen, Seminar für Sprachwissenschaft 2014) are all corpora with transcriptions of specific types of conversations. Only the last two of them are manually annotated with part-of-speech tags, and only in KiDKo was the data annotated part-of-speech tags specific to spoken language (Rehbein/Schalowski 2013). Tüba-D/S consists of transcribed and manually annotated data (360,000 tokens) from the Verbmobil project (Wahlster 2000). The data is tagged …

Content truncated to 500 words