Start with everything at zero: the distribution is a flat line that knows nothing and commits to nothing. Each λ you raise is a measured fact the curve now honours — and nothing beyond it. Watch the entropy bar: it leaves its maximum the moment you add the first commitment.
The same principle over three candidates. With the weight at zero, the grammar splits the probability evenly — maximum entropy, it knows nothing. Raise the weight and it commits to whoever violates least, and the entropy drops.