Skip to content
Loganathan Ramasamy edited this page Jun 5, 2015 · 11 revisions

CzeSL error model tested against CzeSL test data

                 ***************
                 Error detection
                 ***************
Dataset Precision Recall F1-score
czesl-olga-a-w 95.0 96.8 95.9
czesl-czesl-a-w 76.6 83.3 79.8
czesl-czesl-b-w 87.4 66.6 75.6

Data set format : czesl-olga-a-w

czesl - error model

olga - test data set

a - level 1 data

w - level 2 data

                 ***************
                 Error correction
                 ***************
# top-1 top-1 top-1 top-2 top-2 top-2 top-3 top-3 top-3
dataset precision recall F1-score precision recall F1-score precision recall F1-score
czesl-olga-a-w 86.9 96.5 91.5 91.4 96.7 94.0 93.7 96.7 95.2
czesl-czesl-a-w 70.7 82.1 76.0 73.1 82.6 77.6 74.1 82.8 78.2
czesl-czesl-b-w 82.4 65.3 72.9 84.2 65.8 73.9 84.9 66.0 74.3

Error Detection: Richter error model Vs CzeSL error model

Error model-edit-distance Precision Recall F1-score
Richter-1-edit-distance 95.7 90.8 93.2
CzeSL-1-edit-distance 95.3 93.1 94.2
Richter-2-edit-distance 95.0 95.9 95.4
CzeSL-2-edit-distance 95.0 96.8 95.9
Richter-3-edit-distance 95.0 95.9 95.4
CzeSL-3-edit-distance 95.0 96.8 95.9
Richter-4-edit-distance 95.0 95.9 95.4
CzeSL-4-edit-distance 95.0 96.8 95.9
Richter-5-edit-distance 95.0 95.9 95.4
CzeSL-5-edit-distance 95.0 96.8 95.9

The above experiments use unpruned SYN2005 language model created from the Czech National Corpus.

Error Correction: Richter error model Vs CzeSL error model

# top-1 top-1 top-1 top-2 top-2 top-2 top-3 top-3 top-3
Error model-edit_distance precision recall F1-score precision recall F1-score precision recall F1-score
Richter-1-edit-distance 87.9 90.1 89.0 92.3 90.5 91.4 93.7 90.7 92.2
CzeSL-1-edit-distance 87.3 92.5 89.9 91.1 92.8 91.9 93.4 93.0 93.2
Richter-2-edit-distance 86.8 95.5 91.0 91.8 95.7 93.7 93.2 95.8 94.5
CzeSL-2-edit-distance 86.9 96.5 91.5 91.4 96.7 94.0 93.7 96.7 95.2
Richter-3-edit-distance 86.8 95.5 91.0 91.4 95.7 93.5 92.7 95.8 94.2
CzeSL-3-edit-distance 86.9 96.5 91.5 91.4 96.7 94.0 93.7 96.7 95.2
Richter-4-edit-distance 86.8 95.5 91.0 91.4 95.7 93.5 92.7 95.8 94.2
CzeSL-4-edit-distance 86.9 96.5 91.5 91.4 96.7 94.0 93.7 96.7 95.2
Richter-5-edit-distance 86.8 95.5 91.0 91.4 95.7 93.5 92.7 95.8 94.2
CzeSL-5-edit-distance 86.9 96.5 91.5 91.4 96.7 94.0 93.7 96.7 95.2