Bin Yuan 14-11-03

Accomplished this week

• last week's building HCLG method is wrong, can't decode correctly, now fixed
• test tagged HCLG, using seed corpus, baseline wer is 56%, tagged wer is 60%(merge weight 1)

Planned for next week

• use large size corpus to build tagged and try more merge weight
• record a test set for address tag lm and test on it