Slide 9
Slide 9 text
Permutation Acceptance
事例(𝒑𝒊, 𝒉𝒊) に対して:
q個の並び替えにモデルMが正しいラベル𝑦!
を割り当てる確率Pr(&)
(sat The on cat mat the, The was cat fat)
(mat the on The sat cat, fat The cat was)
(cat mat the The sat on, cat was fat The)
𝑞 = 3の場合
Pr( (
𝑃+, *
𝐻+),-.=
1
𝑞
/
( 0
1!"∈ 3
4!,6
7!"∈6
8!)
((𝑀 ̂
𝑝+:, (
ℎ+: = 𝑦+) → 1)
M
𝑗 ∈ [1, 𝑞]
neutral
entailment
neutral
Pr & =
2
3
出⼒
ラベル
9
モデルM neutral
(The cat sat on the mat, The cat was fat)
仮説⽂𝒉𝒊
前提⽂𝒑𝒊
モデルM