Presuppositions are propositions that an utterance takes for granted as part of the common ground between speaker and hearer. The sentence "The king of France is bald" presupposes that there exists a king of France, and this presupposition survives negation: "The king of France is not bald" carries the same presupposition. Unlike entailments, which are canceled by negation, and implicatures, which can be explicitly canceled, presuppositions persist across a range of linguistic environments. Modeling presupposition is essential for computational systems that aim to track the information state of discourse participants and generate contextually appropriate language.
Presupposition Triggers and Projection
Factive verbs: "know that P" → P is true
Change-of-state verbs: "stop V-ing" → was V-ing
Clefts: "It was X who Y" → someone Y-ed
Iteratives: "again" → happened before
Projection: presuppositions survive embedding under
negation, questions, conditionals, and modals
Presuppositions are triggered by specific linguistic constructions. Definite descriptions presuppose the existence and uniqueness of their referent. Factive verbs like "know," "regret," and "realize" presuppose the truth of their complement. Change-of-state verbs like "stop" and "begin" presuppose the prior state. Iteratives like "again" presuppose a previous occurrence. The projection problem — predicting which presuppositions survive when triggers are embedded in complex sentences — has been a central challenge in formal semantics since Karttunen (1973). For example, "If the king of France exists, the king of France is bald" does not presuppose that there is a king of France, while "If it rains, the king of France will be upset" does.
Theories of Presupposition
Several theoretical frameworks address presupposition projection. Karttunen's (1974) approach treats presuppositions as conditions on the context: a sentence is felicitous only if its presuppositions are already in the common ground. Heim's (1983) satisfaction theory, building on dynamic semantics, models presuppositions as requirements that must be satisfied by the local context at the point of evaluation. When presuppositions are not in the common ground, accommodation — the process by which listeners silently add the presupposed content to their belief state — typically occurs. Lewis (1979) introduced the concept of accommodation as a conversational repair mechanism.
Presupposition is a powerful rhetorical tool because it introduces information as background rather than asserted content, making it harder to challenge. The question "Have you stopped cheating on exams?" presupposes prior cheating, and any direct answer accepts this presupposition. Politicians and advertisers exploit presupposition strategically: "When we win this election..." presupposes victory. Computational detection of presupposition-based framing in political discourse and media is an emerging application, combining presupposition trigger identification with analysis of argumentative structure.
Computational Approaches
Computational modeling of presupposition addresses several tasks: identifying presupposition triggers in text, determining what content is presupposed, predicting projection behavior in complex sentences, and detecting when accommodation is required. Rule-based systems encode trigger inventories and projection rules from formal semantics. Statistical approaches learn projection patterns from annotated corpora. Recent neural approaches frame presupposition identification as a natural language inference task, asking whether a candidate presupposition is entailed by the sentence and its negation.
Presupposition interacts with other discourse phenomena in computationally significant ways. Anaphoric presuppositions (presuppositions that include referential expressions) must be resolved against the discourse context, connecting presupposition to coreference resolution. In dialogue, presupposition failure — when a speaker presupposes something the listener does not accept — triggers clarification requests or corrections, requiring dialogue systems to track and verify presuppositions in the common ground. Text generation systems must manage presuppositions to avoid introducing unwarranted background assumptions, particularly in sensitive applications like automated journalism and legal document generation.