Table 6 

Dependency parsing (LAS) results for UDapter with and without typological feature prediction. # missing shows the number of unavailable features in 289 typological features (syntax + phonology + inventory). Note that the base UDapter uses KNN predictions (Littell et al. 2017) for missing typological features. Results for all low-resource languages are given in Appendix D.

High-Resource Languages
areneufihehiitjakorusvtrzhhr-avg
# missing→5321741245413171651163–
no-pred 84.4 90.0 83.1 89.2 89.3 91.9 93.5 92.5 85.7 92.3 90.2 69.4 82.6 87.2 
typo-pred 84.3 89.4 83.2 89.2 89.4 91.9 93.7 92.9 85.8 92.4 90.0 70.2 83.6 87.4 
KNN 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 
 
Low-Resource Languages (Zero-Shot) 
 be br bho fo hsb kk mr olo sa ta te tl yo lr-avg* 
# missing→ 256 41 229 256 289 275 65 289 112 61 47 31 13 – 
no-pred 59.8 61.9 31.0 61.6 42.5 46.5 45.2 31.0 20.2 44.8 70.5 65.4 42.7 32.0 
typo-pred 75.1 60.7 33.6 66.7 48.3 58.6 44.2 36.7 23.0 44.9 71.0 66.8 42.6 35.0 
KNN 79.3 58.5 37.3 69.2 54.2 60.7 44.4 43.3 22.2 46.1 71.1 69.5 42.7 36.5 
High-Resource Languages
areneufihehiitjakorusvtrzhhr-avg
# missing→5321741245413171651163–
no-pred 84.4 90.0 83.1 89.2 89.3 91.9 93.5 92.5 85.7 92.3 90.2 69.4 82.6 87.2 
typo-pred 84.3 89.4 83.2 89.2 89.4 91.9 93.7 92.9 85.8 92.4 90.0 70.2 83.6 87.4 
KNN 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 
 
Low-Resource Languages (Zero-Shot) 
 be br bho fo hsb kk mr olo sa ta te tl yo lr-avg* 
# missing→ 256 41 229 256 289 275 65 289 112 61 47 31 13 – 
no-pred 59.8 61.9 31.0 61.6 42.5 46.5 45.2 31.0 20.2 44.8 70.5 65.4 42.7 32.0 
typo-pred 75.1 60.7 33.6 66.7 48.3 58.6 44.2 36.7 23.0 44.9 71.0 66.8 42.6 35.0 
KNN 79.3 58.5 37.3 69.2 54.2 60.7 44.4 43.3 22.2 46.1 71.1 69.5 42.7 36.5 
View Large
Close Modal
This Feature Is Available To Subscribers Only

or Create an Account

Close Modal
Close Modal
This site uses cookies. By continuing to use our website, you are agreeing to our privacy policy. No content on this site may be used to train artificial intelligence systems without permission in writing from the MIT Press.
Accept