Shared Task on the Disambiguation of German Verbal Idioms at KONVENS 2021

The processing of multiword expressions (MWEs) has gained a lot of attention in recent years. Not least thanks to the shared tasks on verbal MWE identification organized by the PARSEME network. A phenomenon that because of its inherently rare nature is quite infrequent in datasets such as the PARSEME corpus are literal readings of verbal idioms (VIDs). This makes it difficult to efficiently train systems capable of identifying them. To alleviate this issue for German we built a VID corpus with a higher than usual number of literal readings. Together with the SemEval 2013 Task 5b dataset this formed the corpus for the Shared Task on the Disambiguation of German Verbal Idioms at KONVENS 2021. This paper describes the organization of the competition as well as the results of the participating systems.