Question regarding the denominators in Kneser-Ney Smoothing

Web: https://www.reddit.com/r/LanguageTechnology/comments/scyef9/question_regarding_the_denominators_in_kneserney/

Hi everyone!

I am currently studying smoothing techniques, specifically Kneser-Ney smoothing. I understand that it helps to handle the case where the next word hasn't appeared in the given context previously. For eg, the corpus could have non zero trigram counts of 'This is a', but no occurrence of the 4-gram 'This is a car'.

The count C(This is a) is captured in the denominator of the lambda term as well, and this lambda term is multiplied with the recursion …

languagetechnology