Computational Semantics
Computational semantics is the study of how to represent and compute the meaning of words, phrases, and sentences, mapping language to formal meaning representations or to distributional representations of sense.
Definition
Computational semantics is the construction of meaning representations for natural-language expressions, whether as formal logical forms built compositionally from parts or as distributional representations of word and phrase meaning, together with methods to disambiguate and reason about them.
Scope
This topic covers the representation of meaning in NLP: logical/compositional semantics (mapping sentences to logical forms, with the principle of compositionality and quantifier scope), lexical semantics and word sense, semantic role labeling, and distributional and vector representations of meaning. It addresses semantic parsing into executable or logical meaning representations and the resolution of meaning-level ambiguity. The syntactic structures that feed meaning composition are covered under syntactic parsing.
Core questions
- How is the meaning of a sentence built from the meanings of its parts and their syntactic combination?
- How are logical forms and quantifier scope represented and computed?
- How is the correct sense of an ambiguous word determined in context?
- How do distributional and vector representations capture word meaning from usage?
Key concepts
- compositionality
- logical form and quantifier scope
- lambda calculus for meaning
- lexical semantics and word sense
- word-sense disambiguation
- semantic role labeling
- distributional semantics and vectors
- semantic parsing
Key theories
- Compositional (formal) semantics
- Following Montague's program, the meaning of a complex expression is computed as a function of the meanings of its parts and their syntactic combination, allowing sentences to be mapped to logical forms suitable for inference.
- Lexical semantics and word sense
- Words have multiple related senses organized in lexical resources, and word-sense disambiguation uses context to select the intended sense, a prerequisite for accurate meaning representation.
- Distributional semantics
- The meaning of a word can be represented by the contexts in which it occurs, yielding vector representations in which semantic similarity corresponds to proximity, a paradigm captured by the distributional hypothesis that words in similar contexts have similar meanings.
Clinical relevance
Meaning representations support question answering over knowledge bases, natural-language interfaces to databases, dialogue understanding, and textual entailment, by turning sentences into forms that systems can reason over or match; distributional representations underlie much of modern semantic similarity and retrieval.
History
Formal semantics in NLP drew on Montague's compositional treatment of quantification (1973) and logical meaning representation. Lexical resources and word-sense work matured in the 1990s, while the distributional tradition, rooted in Harris's distributional hypothesis, grew into vector-space and embedding methods that now dominate semantic representation.
Key figures
- Richard Montague
- Christopher D. Manning
- Daniel Jurafsky
- Zellig Harris
Related topics
Seminal works
- montague1973
- jurafsky2023
Frequently asked questions
- What is compositionality in semantics?
- Compositionality is the principle that the meaning of a complex expression is determined by the meanings of its parts and the rules used to combine them. It lets systems compute the meaning of novel sentences systematically from their words and structure rather than memorizing whole sentences.
- How do distributional representations capture meaning?
- They rely on the observation that words appearing in similar contexts tend to have similar meanings. By representing each word as a vector summarizing the contexts it occurs in, semantically related words end up with similar vectors, so meaning similarity becomes geometric proximity.