#absolute-discounting

After a context $w$, a corpus shows $N = 10$ following tokens of three word types, with counts 5, 3 and 2. The vocabulary has $V = 10$ words, so 7 types never followed $w$. Absolute discounting with $\delta = 0.5$ subtracts $\delta$ from every non-zero count, gives seen words $(r - \delta)/N$, and shares the freed probability mass equally among the unseen types. What is the probability of each unseen word?