Fine-grained Named Entity Annotations for German Biographic Interviews

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4605–4614 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4605 Fine-grained Named Entity Annotations for German Biographic Interviews Josef Ruppenhofer1, Ines Rehbein2, Carolina Flinz3 1. Leibniz-Institute for the German Language, R5, 6-13 , 68161 Mannheim 2. Mannheim University, Data and Web Science Group 3. Universit`a degli Studi di Milano, Department of Studies in Language Mediation and Intercultural Communication ruppenhofer@ids-mannheim.de, ines@informatik.uni-mannheim.de, carolina.flinz@unimi.it Abstract We present a fine-grained NER annotations scheme with 30 labels and apply it to German data. Building on the OntoNotes 5.0 NER inventory, our scheme is adapted for a corpus of transcripts of biographic interviews by adding categories for AGE and LAN(guage) and also adding label classes for various numeric and temporal expressions. Applying the scheme to the spoken data as well as a collection of teaser tweets from newspaper sites, we can confirm its generality for both domains, also achieving good inter-annotator agreement. We also show empirically how our inventory relates to the well-established 4-category NER inventory by re-annotating a subset of the GermEval 2014 NER coarse-grained dataset with our fine label inventory. Finally, we use a BERT-based system to establish some baselines for NER tagging on our two new datasets. Global results in in-domain testing are quite high on the two datasets, near what was achieved for the coarse inventory on the CoNLLL2003 data. Cross-domain testing produces much lower results due to the severe domain differences. Keywords: Named Entity Recognition, spoken language, German, oral history corpora 1. Introduction While Named Entity Recognition (NER) is typically envi- sioned in service of NLP tasks such as information extrac- tion, question answering, automatic translation, etc (Juraf- sky, 2000), we are interested in it also from a corpus lin- guistic and digital humanities perspective. The Datenbank f¨ur Gesprochenes Deutsch (DGD; ‘Database for Spoken German’) that is hosted by the Leib- niz Institute for the German Language is a repository of, and a platform for research on, spoken German (Schmidt, 2014). It contains a large, continuously growing collection of currently 34 variational and conversational corpora, totalling more than 4.000 hours of audiovisual material, which are used to address a wide variety of research questions. As a first step in adding a layer of shallow semantic analysis to these spoken corpora, we want to provide NER tags. Out of all the corpora in the database, we chose to begin with the ISW corpus, which contains transcripts of German-language biographic narrative inter- views with Austrian-born emigrants to Israel. The ISW corpus is part of a series of three interrelated corpora (IS, ISW, ISZ) collected by Prof. Anne Betten and colleagues over more than two decades, mainly in the 1990s and the 2000s. All three corpora have been and continue to be intensely researched from various angles (Betten and Du-Nour (2004; Leonardi and Th¨une (2011; Leonardi et al. (2016), inter alia). Our initial focus on the ISW corpus is motivated by two considerations. First, we generally …

Content truncated to 500 words