significance of the difference between two values

Hello all,

I'm working on a skew-normal distribution and would like to know if the difference between 2 values of the data set is significant.

It's the distribution of Mutual Information between couples of lemmas (100,000 of them). You can see the distribution on the jpg image.

What I aim is (for instance) being able to say values A=-1 and B=0,7 are not so different so that we cannot really say couple A is really more rejected than B. but C=-1,0 and D=4,5 are. Is there any test to achieve this?

Thank you for the help.