In which condition backoff weight can be positive?

[Biswajit]

example:

I know that if prob. is greater than "1", it can be positive value in log domain.

~~~~~~~~~~~~~~~~
1gram:
-4.1258 above 0.0169

2gram:
-1.6579 above
-1.6579 above all
-1.6579 above him
-1.6579 above his
-1.6579 above the
~~~~~~~~~~~~~~~~~~~~

Total count of above is "5". Total vocabulary size: 6171
How toolkit is calculating BW here with Good Turing?
Freq of those bigrams are 1.
How it will estimate new count for 1 by GT because it is the highest frequency. I'll not get count for 2 as well.

[Biswajit]

Any clarification on this ??

[Nickolay V. Shmyrev]

For the most precise and exact explanation of the calculation process you can read the code.