[BMI776] Back-off model correction

Colin Dewey cdewey at biostat.wisc.edu
Thu Feb 28 12:03:14 CST 2008


Hi all,

There was a question today in class regarding the delta and lambda  
parameters in the back-off models (slides 24-26 of Lecture #11).  I  
checked with the text (Manning & Shutze) and it turns out the the  
lambda and delta parameters are both dependent on the history x_{i- 
n}...x_{i-1}.  So in the case that all counts are greater than k, the  
delta for that history will be zero (i.e., we don't need to back-off  
at all).  I have updated slide 25 with a small note about this.

Thanks,
Colin



More information about the BMI776 mailing list