Video Recordings

I will summarize the parsing shared task we organized this spring. 11 years after the first CoNLL shared task in dependency parsing, the present one has arguably been one of the largest CoNLL shared tasks ever, both in size and diversity of test data (81 treebanks, 49 languages) and in number of participants (over 50 teams, 32 submissions). There were several novel aspects: end-to-end parsing from raw text, low-resource and surprise languages, unified annotation across languages, blind evaluation on a remote server. The results set the new state of the art for dependency parsing of most languages, including Czech.