Source-linked AI summary
Lambda Dependency-Based Compositional Semantics
Percy Liang
TL;DR
Semantic parsing maps utterances to logical forms in formal languages such as lambda calculus. This note introduces lambda DCS, which suppresses explicit variables and makes existential quantification implicit; its logical forms are generally more compact than lambda-calculus forms.
Problem
Semantic parsing maps natural-language utterances to logical forms in a formal language such as lambda calculus.
Source: §1 · p. 1Method
Lambda DCS represents meanings over sets, using joins and adding mu or lambda abstraction for non-tree dependencies or binary constructions.
Source: §2 · pp. 3–4, §2.1 · p. 5Results
Lambda DCS eliminates explicit variables and makes existential quantification implicit, and its forms are generally more compact than lambda-calculus forms.
Source: §1 · p. 1, §2 · p. 3, §2.1 · p. 5Takeaways & Limitations
For logical forms derived from natural language, especially database queries, the authors consider lambda DCS convenient to work with despite its lower expressivity than lambda calculus.
Source: §2.1 · pp. 5–6Lambda DCS is less expressive than lambda calculus, and this note focuses on final-form representation rather than composition, using a simple construction mechanism for short questions.
Source: §2.1 · p. 6
Abstract
from arXiv · showhide
This short note presents a new formal language, lambda dependency-based compositional semantics (lambda DCS) for representing logical forms in semantic parsing. By eliminating variables and making existential quantification implicit, lambda DCS logical forms are generally more compact than those in lambda calculus.
1 Introduction
The note introduces lambda DCS as a potentially more compact formal language for semantic-parsing logical forms, removing explicit variables while retaining variables and lambda abstraction when needed.
1 Introduction
- The example represents “people who have lived in Seattle” as PlacesLived.Location.Seattle using Freebase’s reified living events.In the example, the living event connects a person to its location and can also carry start and end dates.Source: §1 · p. 1
- Lambda DCS avoids explicit variables but can use variables and lambda abstraction to handle phenomena such as anaphora.Unlike DCS, which is restricted to tree-structured logical forms, lambda DCS can handle these cases.Source: §1 · p. 1
- The formalism was designed for natural-language interfaces to Freebase and focuses on noun phrases and wh-questions, not utterance-to-form composition.The note assumes basic familiarity with lambda calculus.Source: §1 · p. 1
2 Definition
Lambda DCS builds logical forms from sets of entities or entity pairs, using implicit existential quantification to avoid explicit variables. It adds abstraction where needed, while remaining less expressive than lambda calculus.
2 Definition
- Join chains such as Children.PlaceOfBirth.Seattle encode entities reaching Seattle through child relations, with existentially shared intermediates forming graph paths.A chain p1...pk.e matches entities that reach e by following the listed property edges.Source: §2 · p. 3
- Intersection combines unary forms into tree-structured graph patterns whose branch points correspond to intersections.The paper presents intersection as conjunction.Source: §2 · p. 3
- Union represents disjunction between unary forms, as in a set containing Oregon, Washington, and Canadian provinces.The paper gives the general operation as u1 ⊔ u2.Source: §2 · p. 3
- Negation applies to a unary form, while higher-order functions operate on entity sets for aggregation, superlatives, and generalized quantification.Examples include counting states and selecting the largest state by area.Source: §2 · p. 4
- Mu abstraction handles non-tree-structured dependencies such as bound anaphora, while lambda abstraction constructs binaries for comparison functions such as child counts.Lambda abstraction applied to unaries produces binary forms; mu abstraction produces unary forms.Source: §2 · pp. 4–5
2.1 Discussion
- Lambda DCS centers logical-form construction on sets and implicitly existential variables, but its expressivity remains below lambda calculus.The paper argues it is convenient for natural-language-derived forms, especially database queries.Source: §2.1 · pp. 5–6
- Lambda DCS adapts DCS's tree structure and default existential quantification, but uses unary and binary forms and omits DCS's mark-execute construct.Its graph-pattern perspective also connects it to SPARQL, while higher-order operations make that perspective less helpful.Source: §2.1 · pp. 5–6
- The paper focuses on representing logical forms rather than constructing them compositionally, and its construction mechanism targets short questions with limited compositional demands.The authors leave open the possibility of construction mechanisms suited to complex sentences.Source: §2.1 · p. 6