Semantics is the systematic study of meaning in language. Within linguistics, it investigates the meanings of words, phrases, and sentences and the principles governing their interpretation. It also overlaps with philosophy of language, particularly in questions about reference, truth, and the relationship between expressions and the world. Semantic theories seek to explain both individual expressions and speakers’ ability to understand unfamiliar combinations of familiar words. (plato.stanford.edu)
Scope and neighboring fields
Semantics is closely connected to syntax, which studies how expressions are structurally organized. Changing structure can change meaning even when the vocabulary remains constant: “The dog chased the cat” and “The cat chased the dog” describe different events. Semantic interpretation therefore requires more than a list of dictionary definitions; it must explain how grammatical organization determines the contributions of an expression’s parts. (plato.stanford.edu)
A neighboring field, pragmatics, examines language use and interpretation in context. A common distinction contrasts conventional linguistic meaning with what speakers communicate through intentions, shared knowledge, and conversational circumstances. The boundary is not absolute: expressions such as “I,” “here,” and “now” have stable linguistic rules but require contextual information to identify their referents. Different theories allocate aspects of this contextual interpretation differently between semantics and pragmatics. (plato.stanford.edu)
Word meaning
Lexical semantics investigates the meanings of words and relationships among them. These include synonymy, or similarity of meaning; antonymy, or opposition; and hyponymy, the relationship between a narrower category and a broader one, as between “sparrow” and “bird.” Such relationships organize vocabulary into interconnected systems rather than isolated entries. The lexical database WordNet, for example, groups word senses into synonym sets and connects them through semantic relations. (plato.stanford.edu)
An important distinction separates polysemy, in which a word has related senses, from homonymy, in which the same form corresponds to distinct, unrelated meanings. “Bank,” referring to a financial institution or a river’s edge, illustrates the latter. These distinctions require analysis of individual senses rather than treating every written word as a single semantic unit. (web.stanford.edu)
Meaning is also more than reference. Gottlob Frege distinguished an expression’s sense—roughly, its mode of presenting something—from its referent. “The morning star” and “the evening star” both refer to Venus, but convey different ways of identifying it. This distinction helps explain why discovering that two expressions refer to the same object can be informative. (plato.sydney.edu.au)
Composition and sentence meaning
The principle of compositionality states that the meaning of a complex expression is determined by the meanings of its constituents and their structure. It provides a framework for explaining linguistic productivity: speakers can interpret sentences they have never encountered because familiar expressions are combined through familiar patterns. Compositionality is a methodological constraint on semantic theories, not a claim that meanings can simply be added together. (plato.stanford.edu)
Formal semantics uses tools from logic and mathematics to specify these interpretations precisely. In a simple model-theoretic treatment, a name denotes an individual and a predicate denotes a set of individuals or a relation. A sentence such as “Mira sleeps” is true under an interpretation when the individual assigned to “Mira” belongs to the set assigned to “sleeps.” (plato.stanford.edu)
Truth-conditional semantics characterizes declarative sentences through the conditions under which they are true. It also analyzes entailment: one sentence entails another when the second is true in every relevant interpretation in which the first is true. Quantifiers, negation, tense, and scope complicate this account. “Every student read a book,” for instance, can allow different books for different students or identify one book read by everyone. (plato.stanford.edu)
Montague semantics, developed by Richard Montague around 1970, established an influential mathematical framework connecting natural-language syntax with interpretation. Its treatment of intensional expressions uses possible worlds to represent alternatives to actuality, supporting analyses of necessity, possibility, and expressions involving beliefs or desires. (plato.stanford.edu)
Context and discourse
Semantic interpretation also extends beyond individual sentences. Dynamic semantics models meaning in terms of changes to an informational context. In “A woman entered. She sat down,” the first sentence introduces an individual that the pronoun in the second sentence can subsequently identify. This approach makes discourse connections part of the semantic explanation rather than analyzing each sentence independently. (plato.sydney.edu.au)
Presupposition concerns information treated as background, whereas conversational implicature concerns conclusions inferred from how an utterance is used. Saying “Some students passed” often suggests that not all passed, but that suggestion can be explicitly canceled. Distinguishing asserted content, background assumptions, and inferred messages is a central task at the semantics–pragmatics interface. (plato.stanford.edu)
Computational approaches
In natural language processing, semantic analysis includes identifying word senses, measuring similarity, and constructing representations usable by computational systems. The distributional hypothesis motivates learning meanings from linguistic environments: words occurring in similar contexts tend to have related meanings. Word embeddings represent these patterns as vectors in a vector space, enabling numerical comparisons of semantic similarity. (web.stanford.edu)
Distributional representations and logical representations capture different properties. Vector proximity can reflect similarity or association, while truth-conditional analyses explicitly describe reference and inferential relationships. Consequently, numerical similarity is not equivalent to synonymy, and a vector representation does not by itself specify the truth conditions of a sentence. (web.stanford.edu)