BLiMP-NL is a data set for evaluating the linguistic knowledge of language models. It is divided into BLiMP-NL small and BLiMP-NL large. Both contain minimal pairs for 22 grammatical phenomena in Dutch, further divided into 84 paradigms. The difference between the small and the large data set is that there are 10 minimal pairs per paradigm in the small data set and 100 minimal pairs per paradigm in the large data set. All minimal pairs have been evaluated by native speakers of Dutch in a self-paced reading task also including an acceptability judgement on a 7-point scale.
Publisher
Radboud Universiteit
Organisations
Interfacultary Research - Institute for Logic, Language and Computation (ILLC)
