In a neural network, opposing concepts are more similar (related) than minor diversions from those.
Hence, there is a gradual scale (gradient) with on the one extreme similarity and on the other extreme complete difference.
Exactly in the middle between those, we find the most unrelated, which is 50% different.
So, as life and death are semantically strongly interconnected, they function as both synonym and antonym of each other.
They exist in closer proximity than life to living, or death to sleeping.