Table 5.
Classification task performance of the MLP neural network. All data are for leave-one-infant-out partitioning of utterances.
| Type of hidden layer | Protophones | Ages | ||
|---|---|---|---|---|
| Protophone-predictinga | Age-predicting | Age-predicting | Protophone-predicting | |
| Mean % correct | 45.9 (chance=33.3) | 46.6 (chance=33.3) | 35.1 (chance=33.3) | 36.1 (chance=33.3) |
| Standard deviation | 10.3 | 5.7 | 3.6 | 3.0 |
| Mean Cohen’s κ | 0.191 | 0.200 | 0.026 | 0.041 |
| Mean p | <0.001 | <0.001 | 0.118 | 0.018 |
a
With adjustment for unequal category sizes. When there is no adjustment and one infant is reserved for testing, the mean percent correct is 65.3 (chance=74.9) with a standard deviation of 6.9.