Skip to Main Content
Table 1

Dependency parsing results (LAS) of baselines and UDapter for high-resource languages. The last column shows average LAS of the 13 high-resource languages (hr-avg). Results for previous work are taken from Kulmizev et al. (2019) [1] and Kondratyuk and Straka (2019) [2,3].

areneufihehiitjakorusvtrzhhr-avg
Previous work: 
uuparser-bert[1] 81.8 87.6 79.8 83.9 85.9 90.8 91.7 92.1 84.2 91.0 86.9 64.9 83.4 84.9 
 
udpipe [2] 82.9 87.0 82.9 87.5 86.9 91.8 91.5 93.7 84.2 92.3 86.6 67.6 80.5 85.8 
udify [3] 82.9 88.5 81.0 82.1 88.1 91.5 93.7 92.1 74.3 93.1 89.1 67.4 83.8 85.2 
 
Monolingually trained (one model per language): 
mono-udify 83.5 89.4 81.3 87.3 87.9 91.1 93.1 92.5 84.2 91.9 88.0 66.0 82.4 86.0 
 
Multilingually trained (one model for all languages): 
multi-udify 80.1 88.5 76.4 85.1 84.4 89.3 92.0 90.0 78.0 89.0 86.2 62.9 77.8 83.0 
adapter-only 82.8 88.3 80.2 86.9 86.2 90.6 93.1 91.6 81.3 90.8 88.4 66.0 79.4 85.0 
udapter 84.4 89.7 83.3 89.0 88.8 92.0 93.5 92.8 85.9 92.2 90.3 69.6 83.2 87.3 
areneufihehiitjakorusvtrzhhr-avg
Previous work: 
uuparser-bert[1] 81.8 87.6 79.8 83.9 85.9 90.8 91.7 92.1 84.2 91.0 86.9 64.9 83.4 84.9 
 
udpipe [2] 82.9 87.0 82.9 87.5 86.9 91.8 91.5 93.7 84.2 92.3 86.6 67.6 80.5 85.8 
udify [3] 82.9 88.5 81.0 82.1 88.1 91.5 93.7 92.1 74.3 93.1 89.1 67.4 83.8 85.2 
 
Monolingually trained (one model per language): 
mono-udify 83.5 89.4 81.3 87.3 87.9 91.1 93.1 92.5 84.2 91.9 88.0 66.0 82.4 86.0 
 
Multilingually trained (one model for all languages): 
multi-udify 80.1 88.5 76.4 85.1 84.4 89.3 92.0 90.0 78.0 89.0 86.2 62.9 77.8 83.0 
adapter-only 82.8 88.3 80.2 86.9 86.2 90.6 93.1 91.6 81.3 90.8 88.4 66.0 79.4 85.0 
udapter 84.4 89.7 83.3 89.0 88.8 92.0 93.5 92.8 85.9 92.2 90.3 69.6 83.2 87.3 
Close Modal

or Create an Account

Close Modal
Close Modal