Configurations for y ⁺ and y⁻ for semantic parsing. We abbreviate 𝒫(x) = 𝒦(x) : δ(y) = 1, which is the most likely output in the k-best list 𝒦(x) that leads to the correct answer, and 𝒩(x) 𝒦(x) : δ(y) = 0, which is the most likely output in the k-best list 𝒦(x) that leads to the wrong answer.