Following are the results obtained for the Word Segmentation. The validation was done a different test corpora also and it turns that the precision recall (when all errors are equal) is high than the precision recall when errors are computed as in morpheme boundary case. After tryin out on various test data, there was not much change obtained. Following are the results and the corresponding files can be downloaded from the Codes link above:
ATTEMPT1
test : hi-wordseg.txt
ground truth : hi-wseg-gt.txt
All error equal:
Recall : 65.3282345443%
Precision : 88.5146804836%
Morphology like error:
Recall : 76.2470308789%
Precision : 57.3214285714%
ATTEMPT2
test : gt1_test.txt
ground truth : gt1_truth.txt
All error equal:
Recall : 56.574185766%
Precision : 90.8914728682%
Morophology like error :
Recall : 65.2303120357%
Precision : 44.4782168186%
ATTEMPT3
test : gt2_test.txt
ground truth : gt2_truth.txt
All error equal:
Recall : 57.1287128713%
Precision : 91.2974683544%
Morphology like error:
Recall : 66.6666666667%
Precision : 45.3052631579%