A Framenet and Frame Annotator for German Social Media

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3942–3949 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3942 A Framenet and Frame Annotator for German Social Media Eckhard Bick Institute of Language and Communication, University of Southern Denmark Campusvej 55, 5230 Odense M, Denmark eckhard.bick@mail.dk Abstract This paper presents PFN-DE, a new, parsing- and annotation-oriented framenet for German, with almost 15,000 frames, covering 11,300 verb lemmas. The resource was developed in the context of a Danish/German social-media study on hate speech and has a strong focus on coverage, robustness and cross-language comparability. A simple annotation scheme for argument roles meshes directly with the output of a syntactic parser, facilitating frame disambiguation through slot-filler conditions based on valency, syntactic function and semantic noun class. We discuss design principles for the framenet and the frame tagger using it, and present statistics for frame and role distribution at both the lexicon (type) and corpus (token) levels. In an evaluation run on Twitter data, the parser-based frame annotator achieved an overall F-score for frame senses of 93.6%. Keywords: German Framenet, Automatic frame annotation, Semantic roles, Constraint Grammar 1. Introduction In corpus linguistics, semantic annotation is a valuable addition to ordinary, morphosyntactic tagging, lemmatization and dependency relations. This is true in particular for corpus research transcending lexicographical issues, language variation and usage statistics (in our case, German/Danish hate speech). Still, with the possible exception of named-entity recognition (NER), automatic semantic annotation tools are still difficult to find for most languages. Ontologies like WordNet (Fellbaum 1998) have a good coverage and effective classification principle for nouns (hyponymy), but usually do not provide the structural-relational information necessary to disambiguate senses/synsets. And though troponymy goes part of the way, semantically bundling and annotating verb senses is even harder, because verb semantics is closely related to clause structure and argument relations. Thus, for English, VerbNet (Kipper et al., 2006), Berkeley FrameNet (Baker et al., 1998, Johnson & Fillmore, 2000, Ruppenhofer et al., 2010) and PropBank1 (Palmer et al., 2005) address entire predications or frames, assigning a semantic/frame class to the core lexeme (typically but not necessarily a verb) and semantic roles (also called case/thematic roles, Fillmore 1968) to its arguments and possibly adjuncts. A prototypical frame consists of a full verb and its nominal, adverbial or subclause complements. However, for a comprehensive semantic annotation, a frame annotator must also address noun- or adjective-based frames and semantically transparent or verb-incorporated material. In his comparison of the WordNet and FrameNet approaches, Boas (2005) stresses the added level of abstraction provided by the latter (e.g. cross-POS frames), as well as the systematic link between semantic information and lexico-syntactic patterns. Thus, both FrameNet and PropBank provide morphosyntactic 1 Methodologically, VerbNet and the English Berkeley FrameNet were conceived as lexicographical projects, one frame at a time, while PropBank and the German SALSA framenet depart from corpus data, one sentence at a time. Coverage problems are therefore different in nature: In the former, a common …

Content truncated to 500 words