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.
| Model . | Acc . | Acc by cat freq . | |||
|---|---|---|---|---|---|
| All . | ≥100 . | 10–99 . | 1–9 . | OOV . | |
| n =53,765 . | n =50,754 . | n =989 . | n =292 . | n =1,730 . | |
| N=1,351 . | N=188 . | N=240 . | N=118 . | N=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 |
| Model . | Acc . | Acc by cat freq . | |||
|---|---|---|---|---|---|
| All . | ≥100 . | 10–99 . | 1–9 . | OOV . | |
| n =53,765 . | n =50,754 . | n =989 . | n =292 . | n =1,730 . | |
| N=1,351 . | N=188 . | N=240 . | N=118 . | N=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 |