search

tm4roon / survey

Survey on machine learning.
14 stars 1 forks source link

Lambda Dependency-Based Compositional Semantics #113

Open tm4roon opened 3 years ago

tm4roon commented 3 years ago

Lambda Dependency-Based Compositional Semantics

意味論的構文解析(Semantic Parsing)において、新たな形式表現(Lambda-DCS: Lambda Dependency-Based Compositional Semantics)を提案した論文。従来用いられていたラムダ式(Lambda calculus)に比べ、より簡潔に書くことができる。

以下に、Lambda calculusとLambda-DCSの比較を示す。

Utterance Lambda calculus Lambda-DCS
Unary base case - λx.[x = Seattle] Seattle
Binary base case - λx.λy.PlaceOfBirth(x, y) PlaceOfBirth
Join people born in Seattle λx.PlaceOfBirth(x, Seattle) PlaceOfBirth.Seattle
Join those who had children born in Seattle λx.∃y.Children(x, y) ∧ PlaceOfBirth(y, Seattle) Children.PlaceOfBirth.Seattle
Intersection scientists born in Seattle λx.Profession(x, Scientist) ∧ PlaceOfBirth(x, Seattle) Profession.Scientist ⊓ PlaceOfBirth.Seattle
Union Oregon, Washington and Canadian provinces λx.[x = Oregon] ∨ [x = Washington] ∨ Type(x, CanadianProvince). Oregon ⊔ Washington ⊔ Type.CanadianProvince
Negation states not bordering California λx.Type(x, USState) ∧ ¬Border(x, California) Type.USState ⊓ ¬Border.California
Higher-order functions the number of states count(λx.Type(x, USState)) count(Type.USState)
Higher-order functions the number largest state by area argmax(λx.Type(x, USState), λx.λy.Area(x, y)) argmax(Type.USState, Area)
Lambda abstraction those who had a child who influenced them λx.∃y.Children(x, y).Influenced(y, x) µx.Children.Influenced.x
Lambda abstraction person who has the most children argmax(λx.Type(x, Person), λx.count(λz.Children(x, z))) argmax(Type.Person, R[λx.count(R[Children].x]))

文献情報