All That Glitters is Not Gold: A Gold Standard of Adjective-Noun Collocations for German

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4368–4378 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4368 All That Glitters is Not Gold: A Gold Standard of Adjective-Noun Collocations for German Yana Strakatova1, Neele Falk1, Isabel Fuhrmann2, Erhard Hinrichs1, Daniela Rossmann3 1,3University of Tübingen, 2Berlin-Brandenburg Academy of Sciences 1firstname.lastname@uni-tuebingen.de 2fuhrmann@bbaw.de 3daniela.rossmann@student.uni-tuebingen.de Abstract In this paper we present the GerCo dataset of adjective-noun collocations for German, such as alter Freund ‘old friend’ and tiefe Liebe ‘deep love’. The annotation has been performed by experts based on the annotation scheme introduced in this paper. The resulting dataset contains 4,732 positive and negative instances of collocations and covers all the 16 semantic classes of adjectives as defined in the German wordnet GermaNet. The dataset can serve as a reliable empirical basis for comparing different theoretical frameworks concerned with collocations or as material for data-driven approaches to the studies of collocations including different machine learning experiments. This paper addresses the latter issue by using the GerCo dataset for evaluating different models on the task of automatic collocation identification. We compare lexical association measures with static and contextualized word embeddings. The experiments show that word embeddings outperform methods based on statistical association measures by a wide margin. Keywords: MultiWord Expressions & Collocations, Semantics, Statistical and Machine Learning Methods 1. Introduction Lexical collocations (further simply collocations) such as golden memory and old friend are multi-word units that consist of a base, chosen freely by a speaker, and a col- locate, the choice of which is restricted depending on the base (Mel’čuk, 2012). In that aspect, they differ from free phrases, e.g. golden crown or old lady, where neither of the constituents is lexically constrained. However, collocations are not fully lexicalized as opposed to semantically opaque idiomatic expressions such as golden ticket ‘good opportu- nity’ or old flame ‘former romantic partner’. In the recent decades, collocations have been extensively studied with the focus predominantly on their statistical properties and meth- ods of automatic collocation extraction (Church and Hanks, 1990; Smadja, 1993; Evert, 2004; Pecina, 2008a; Bouma, 2009; Evert et al., 2017; Garcia et al., 2019). Identifying and extracting collocations, either manually or automatically, is a challenging task that requires clear definitions of concepts and reliable tools and resources. In spite of the growing interest in collocations, there exist only a few resources that can serve as gold standards in collocation research. This paper reports on the construction of the dataset an- notated by experts that comprises 4,732 instances of collo- cations and non-collocations (free phrases, idioms, named entities, terms). The dataset includes only adjective-noun phrases as they have been studied less extensively than ver- bal collocations. In our current research, we focus on the German language since digital resources and tools are avail- able for German that identify relevant adjective-noun co- occurrences. The DWDS (short for Digitales Wörterbuch der deutschen Sprache) (DWDS, 2019) and its collocation extraction application the Wortprofil1 serve as our empiri- cal basis, …

Content truncated to 500 words