Analyse orientée corpus d'universaux linguistiques sur Universal Dependencies

International audienceThis paper presents experiments aiming at verifying Greenberg's universals based on Universal Dependencies (UD) corpora (de Marneffe et al., 2021). We adopt a corpus-based approach that allows us to highlight inconsistencies between corpora of the same language and to explore the causes of these inconsistencies. In addition to intra-language inconsistency, our analysis on 141 corpora, i.e. 74 languages, also shows cross-language inconsistency and questions the adaptability of UD annotations for some linguistic concepts