Computational Semantics
Computing the meaning of words, sentences, and discourse: representing lexical and compositional meaning, labeling who did what to whom, and resolving reference and coherence across text.
Definition
Computational semantics is the study of how the meaning of linguistic expressions can be represented and computed automatically, from individual word senses to the interpretation of connected discourse.
Scope
Covers the machine treatment of meaning — lexical semantics and word-sense disambiguation, compositional construction of sentence meaning, predicate-argument structure via semantic role labeling, and discourse and pragmatic phenomena such as coreference and coherence. It spans both logic-based and distributional approaches. Syntactic structure is covered in the parsing area and learned representations in statistical-and-neural NLP.
Sub-topics
Core questions
- How are word meanings represented and disambiguated in context?
- How is sentence meaning composed from the meanings of its parts?
- How can predicate-argument structure be recovered automatically?
- How are reference and coherence tracked across a discourse?
Key concepts
- word sense
- compositionality
- logical form
- predicate-argument structure
- semantic role
- coreference
- discourse coherence
- distributional meaning
Key theories
- Compositional meaning representation
- Building the meaning of a sentence systematically from the meanings of its constituents and their syntactic combination, often into logical forms.
- Distributional semantics
- Inferring meaning from patterns of word co-occurrence in corpora, complementing logic-based approaches with empirically grounded similarity.
History
Computational semantics inherited logic-based meaning representation from Montague's formal semantics and inference traditions, while the statistical era added distributional and corpus-based meaning. The two strands increasingly converge, with neural models learning representations that approximate both compositional and distributional meaning.
Debates
- Logical versus distributional meaning
- Whether meaning is best captured by explicit logical forms supporting inference or by distributional vectors capturing similarity; hybrid and neural approaches now aim to combine their strengths.
Key figures
- Patrick Blackburn
- Johan Bos
- Christopher Manning
- Richard Montague
Related topics
Seminal works
- blackburn2005
- manning1999
- jurafsky2025
Frequently asked questions
- What is compositionality?
- Compositionality is the principle that the meaning of a complex expression is determined by the meanings of its parts and the way they are combined, allowing systems to interpret sentences they have never seen before.