140622 - Xi Ma

Last Week

1.Extract different thresholds corpus from weibo and baiduzhidao using cross-entropy method.

2.Using different thresholds corpus to train language models, test ppl and wer of these models.

3.Using new test set to test ppl and wer of our language models.

This Week

1.Familiar with the entire process development language model.

2.Continue to optimize the language models.

3.Continue to update vocabulary.