GRAIN-S: Manually Annotated Syntax for German Interviews

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5169–5177 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5169 GRAIN-S: Manually Annotated Syntax for German Interviews Agnieszka Falenska1, Zolt´an Czesznak1, Kerstin Jung1, Moritz V¨olkel1, Wolfgang Seeker2, Jonas Kuhn1 1 University of Stuttgart, Institute for Natural Language Processing (IMS) 2 Retresco, Berlin {firstname.lastname}@ims.uni-stuttgart.de, wolfgang.seeker@retresco.de Abstract We present GRAIN-S, a set of manually created syntactic annotations for radio interviews in German. The dataset extends an existing corpus GRAIN and comes with constituency and dependency trees for six interviews. The rare combination of gold- and silver-standard annotation layers coming from GRAIN with high-quality syntax trees can serve as a useful resource for speech- and text-based research. Moreover, since interviews can be put between carefully prepared speech and spontaneous conversational speech, they cover phenomena not seen in traditional newspaper-based treebanks. Therefore, GRAIN-S can contribute to research into techniques for model adaptation and for building more corpus-independent tools. GRAIN-S follows TIGER, one of the established syntactic treebanks of German. We describe the annotation process and discuss decisions necessary to adapt the original TIGER guidelines to the interviews domain. Next, we give details on the conversion from TIGER-style trees to dependency trees. We provide data statistics and demonstrate differences between the new dataset and existing out-of-domain test sets annotated with TIGER syntactic structures. Finally, we provide baseline parsing results for further comparison. Keywords: syntax, treebank, non-canonical data 1. Introduction Treebanks, i.e. structurally annotated corpora, play an im- portant role both in the language sciences (linguistics, psy- cholinguistics) and in speech and language technology. They serve as gold-standard data for testing hypotheses or evaluating automatic systems, provide the signal in su- pervised training of machine learning models, or inform processes of adaptation, generation of synthetic data, etc. Therefore, for more and more languages corpora annotated for syntactic structure have been provided to the research community – not least in response to the Universal Depen- dencies initiative (Nivre et al., 2016). It is known that for language-technological systems trained with supervised machine learning, there is a relatively strong dependency on the text genre, language register, con- tent domain and other dimensions of the material in the training corpus (Sekine, 1997). For research into tech- niques for model adaptation and for building more corpus- independent tools, it is important to have test data that rep- resent relevant variations of existing treebanks for the same language. For example, since adaptation of text-processing tools to spoken language is of central importance to many research and application contexts, the availability of man- ually annotated syntactic structures on samples of spoken utterances is crucial. In this contribution we present GRAIN-S(yntax) – a set of manually created syntactic annotations for GRAIN, a corpus of German RAdio INterviews (Schweitzer et al., 2018). The nature of the interview situation differentiates GRAIN-S from existing German treebanks. The utter- ances we find in the corpus can be put halfway between carefully prepared speech and spontaneous conversational speech: the interviewers’ …

Content truncated to 500 words