TGDA 2.0 – An Enriched Texas German Dialect Corpus for (Comparative) Corpus Analyses

The Texas German Dialect Archive (TGDA) is one of the largest resources documenting an extraterritorial variety of German, containing audio recordings from 800+ speakers (1,000+ hours). It has long been accessible through the project’s website (https://tgdp.org/dialectarchive) and the speechislands.org portal, both for viewing/browsing and for simple querying of the data. However, corpus linguistic approaches to the corpus have been limited because the data are lacking annotation layers that would allow queries abstracting over the concrete (and not always predictable) form of transcribed tokens. With a smaller, representative sample of the corpus — the Texas German Sample Corpus (TGSC) — Blevins (2022) studied how the transcribed data could be language-tagged and orthographically “normalized” (i.e. mapped from idiosyncratic or variety-specific to standard orthography forms). This is a crucial step both for enabling more reliable queries and for preparing further (automatic) annotation, such as lemmatization and part-of-speech tagging, which in turn makes the data much more usable for new approaches to query and analysis. Using the training data of the TGSC (for normalization and language tagging) and the FOLK1 - trained STTS 2.0 tagger (for lemmatization and POS tagging), as well as additional information from DeReWo2 and the Corpus of Contemporary American English (COCA) (Davies 2008-), several layers of annotation have been added to the TGDA transcripts. In this presentation, we will explain the workflow behind using the results of the TGSC study to enhance the larger TGDA, including adding automatic language tagging, normalization, lemmatization and part-of-speech tagging layers to all transcribed data. This enhanced TGDA data is now available using ZuMult (https://zumult.org/), which allows users to explore transcripts as well as basic speaker metadata. We will also report on the quality (i.e. error-rates) of automatically annotated data, as well as demonstrate how such an enriched corpus can be used and how it can serve as a model for enabling new corpus-based approaches to comparative speech island analysis (Boas 2016, 2021). 1 The Forschungs- und Lehrkorpus Gesprochenes Deutsch (Schmidt 2016).2 https://www.ids-mannheim.de/digspra/kl/projekte/methoden/derewo/ References Boas, Hans C. (2016): Variation im Texasdeutschen: Implikationen für eine vergleichende Sprachinselforschung. In: Alexandra N. Lenz (ed.): German Abroad. Perspektiven der Variationslinguistik, Sprachkontakt- und Mehrsprachigkeitsforschung. Göttingen: Vandenhoeck & Ruprecht. Boas, Hans C. (2021): Zur Vergleichbarkeit von Sprachinseldaten. Ein Plädoyer für eine “bottom-up” Methodologie im Rahmen der Konstruktionsgrammatik und der Frame Semantik. In C. Földes (ed.), Kontaktvarietäten des Deutschen im Ausland. Tübingen: Narr Francke Attempto. 66-88. Blevins, Margaret (2022): The language-tagging and orthographic normalization of Germanlanguage contact varieties. Dissertation, University of Texas, Austin. Davies, Mark. (2008-) The Corpus of Contemporary American English (COCA). Available online at www.english-corpora.org/coca/. Schmidt, Thomas (2016): Construction and dissemination of a corpus of spoken interaction: tools and workflows in the FOLK project. In: Marc Kupietz & Alexander Geyken (eds) Corpus Linguistic Software Tools, Journal for Language Technology and Computational Linguistics (JLCL 31/1), p. 127-154.