Language Recognition Using Latent Dynamic Conditional Random Field Model with Phonological Features

Table 10

Error of language classifiers trained from the feature vector after removal of one PF attribute (averaged across 28 pairs).

Removed attribute	Error; (%)
None (Full PF)	17.46
High	17.79
Tense	17.80
Voiced	17.87
Strident	18.00
Anterior	18.02
Low	18.05
Vocalic	18.10
Continuant	18.12
Coronal	18.15
Nasal	18.17
Round	18.31
Consonantal	18.44
Back	18.60

The least error result is displayed in bold.