Comparison of bilingual lexicon induction accuracies using (1) matching canonical correlation analysis (MCCA), (2) our supervised discriminative model using only contextual and orthographic features, and (3) our supervised discriminative model using our complete feature set. Accuracy is measured as the percent of test set translations that are correctly matched by each model's full bipartite matching.
Model
.
Accuracy (%)
.
MCCA
15.1
Discriminative Model w/ Context and Orthographic Features Only
24.3
Discriminative Model w/ All Features
42.3
Model
.
Accuracy (%)
.
MCCA
15.1
Discriminative Model w/ Context and Orthographic Features Only