Skip to content

Latest commit

 

History

History
20 lines (13 loc) · 987 Bytes

File metadata and controls

20 lines (13 loc) · 987 Bytes

Results UD

back to main README

We provide results on UD version 2.7 and 2.8 (as of now, if you need another version, e-mail the first author). These results are obtained by running the scripts in scripts/2.*. This means that they are trained without multi-words, but evaluated with multi-words for fair comparison! Furthermore, they are trained with the default settings of MaChAmp 0.2, and the average over 3 seeds.

The results for version 2.8 are shared on (just replace the 8 with a 7 for the older scores):

The first link contains the full output as well as the output of the official conllu evaluation script. The latter link contains one file per UD-version and task, and contains the scores of the 5 models used in the paper for each dataset averaged over 3 seeds.