ScholarGate
Asistenti

Compositional Semantics and Semantic Parsing

Building the meaning of a sentence from its parts and mapping natural language to formal meaning representations such as logical forms and executable queries.

Gjeni temë me PaperMindSë shpejtiFind papers & topics
Tools & resources
Shkarko diapozitivat
Learn & explore
VideoSë shpejti

Definition

Compositional semantics computes the meaning of a complex expression from the meanings of its constituents and their mode of combination; semantic parsing maps natural language to a formal meaning representation.

Scope

Covers the construction of sentence-level meaning: the principle of compositionality, logical-form representations, the syntax–semantics interface, and semantic parsing that maps utterances to formal representations usable for inference or database queries. It includes both formal and learned approaches. Word-level meaning is covered in lexical semantics and discourse-level meaning in a sibling topic.

Core questions

  • How is sentence meaning assembled compositionally from word meanings and structure?
  • How are quantifiers, scope, and negation represented in logical form?
  • How can semantic parsers be learned from data?
  • How are meaning representations used for inference and querying?

Key concepts

  • compositionality
  • logical form
  • lambda calculus
  • quantifier scope
  • syntax-semantics interface
  • semantic parsing
  • meaning representation
  • inference

Key theories

Montague grammar
Montague's program of treating natural language with the rigor of formal logic, pairing syntactic rules with semantic operations to compute truth-conditional meaning.
Learned semantic parsing
Inducing a parser that maps sentences to logical forms from annotated examples, using probabilistic categorial grammars and structured prediction.

History

Montague's 1970s work showed natural language could be given a precise model-theoretic semantics, inspiring computational implementations surveyed by Blackburn and Bos. From the mid-2000s, semantic parsing turned to machine learning, with Zettlemoyer and Collins learning logical-form mappings from data, a line that continues into neural semantic parsing.

Debates

Hand-built versus learned meaning representations
Whether formal logical forms should be designed by hand or induced from data, and how much explicit logical structure neural systems still need for reliable inference.

Key figures

  • Richard Montague
  • Patrick Blackburn
  • Johan Bos
  • Luke Zettlemoyer

Related topics

Seminal works

  • montague1973
  • blackburn2005
  • zettlemoyer2005

Frequently asked questions

What is a logical form?
A logical form is a formal, unambiguous representation of a sentence's meaning, often in a logic such as the lambda calculus, that a computer can use to reason or to query a database.

Methods for this concept

Related concepts