Table 1 Average test accuracy (in percent) by word-specific models with structure (i) through (iv), with and without layer \(\mathcal {C}\)
Models | with \(\mathcal {C}\) | without \(\mathcal {C}\) |
|---|---|---|
Basic NN | 93.40 | 94.25 |
Sum NN | 93.60 | 94.48 |
Cct-V NN | 94.41 | 94.78 |
Cct-T NN | 94.50 | 94.87 |
Cct-T NN whole-paragh | NA | 96.00 |
Attention cct-T whole-paragh | 93.80 | 93.94 |