Skip to Main Content
Table 6: 

Performance of the best systems (the variants with attention for each paradigm) on redistributed Rebank train/test splits. Frequency bins are based on the new training set.

ModelAccAcc by cat freq
All≥10010–991–9OOV
n =53,765n =50,754n =989n =292n =1,730
N=1,351N=188N=240N=118N=805
Nonconstructive 
MLP_10 88.76 92.86 55.71 13.24 – 
MLP_1 88.79 92.87 55.61 19.29 – 
 
Sequential 
K+19 80.20 83.49 47.72 25.11 11.62 
RNN 88.73 92.64 52.92 23.52 5.38 
 
Tree-structured 
TreeRNN 88.78 92.54 49.90 20.55 9.62 
AddrMLP 89.01 92.70 54.03 26.48 10.96 
ModelAccAcc by cat freq
All≥10010–991–9OOV
n =53,765n =50,754n =989n =292n =1,730
N=1,351N=188N=240N=118N=805
Nonconstructive 
MLP_10 88.76 92.86 55.71 13.24 – 
MLP_1 88.79 92.87 55.61 19.29 – 
 
Sequential 
K+19 80.20 83.49 47.72 25.11 11.62 
RNN 88.73 92.64 52.92 23.52 5.38 
 
Tree-structured 
TreeRNN 88.78 92.54 49.90 20.55 9.62 
AddrMLP 89.01 92.70 54.03 26.48 10.96 
Close Modal

or Create an Account

Close Modal
Close Modal