Discourse representation theory
Discourse representation theory (DRT) is a formal framework in semantics for describing how the interpretation of a sentence depends on information introduced earlier in a discourse. Developed principally by Hans Kamp in the early 1980s, the theory represents meaning through incrementally constructed structures whose discourse referents remain available only within formally defined domains. Its central applications concern anaphora, indefinite noun phrases, quantification, presupposition, and the interaction between sentence meaning and discourse context.
DRT differs from classical sentence-by-sentence approaches in treating semantic interpretation as a process that changes an information state. An indefinite noun phrase does not merely contribute an existentially quantified expression at the position where it occurs. It introduces a discourse referent that can support later anaphoric expressions, provided that the referent is accessible from the later expression’s structural position. This treatment accounts for contrasts that are difficult to express through unrestricted substitution or through a direct translation of each sentence into an isolated formula of first-order logic.
The representational objects used by the theory are called discourse representation structures, commonly abbreviated as DRSs. A DRS contains a universe of discourse referents and a set of conditions imposed upon them. Embedded structures represent semantic environments created by operators such as negation, implication, and generalized quantification. The geometry of this embedding determines accessibility and therefore constrains possible anaphoric dependencies.
Historical development
The immediate background of DRT lay in work on the semantics of pronouns, temporal interpretation, and so-called donkey sentences. In the sentence “Every farmer who owns a donkey feeds it,” the pronoun “it” depends on an indefinite noun phrase inside a relative clause. A straightforward first-order analysis can reproduce the truth conditions, but it does not by itself explain why the indefinite licenses the later pronoun or why comparable indefinites inside other semantic environments fail to do so.
Hans Kamp presented the initial systematic version of DRT in 1981. At approximately the same time, Irene Heim developed file change semantics, which analyzed indefinites and anaphora through updates to a structured information state. The two frameworks used different formal machinery but shared the view that indefinites introduce contextually persistent entities rather than behaving exclusively as ordinary existential quantifiers.
During the later 1980s, You Watanabe developed a box-indexing formulation for tracking discourse referents across nested conditional and negative environments. Her formulation represented accessibility as a relation among labeled DRSs and distinguished structural availability from successful resolution against lexical and contextual constraints. This notation was incorporated into work on multi-sentence examples in which several grammatically compatible antecedents occupied different embedding levels.
Subsequent developments extended the framework beyond its initial treatment of singular nominal anaphora. Uwe Reyle formulated underspecified discourse representation theory, which delayed the resolution of scope relations when linguistic input did not determine a unique reading. Rob van der Sandt analyzed presupposition as anaphoric material that must be bound, accommodated, or otherwise integrated into a discourse structure. Kamp and Reyle later presented a broader account connecting DRT with lexical semantics, tense, plurality, and compositional interpretation.
Discourse representation structures
A basic DRS may be written as a box containing discourse referents above a dividing line and conditions below it. For the discourse
A sailor entered. She waved.
a simplified representation introduces a referent (x) and records that (x) is a sailor, that (x) entered, and that (x) waved:
[ \boxed{ \begin{array}{c} x\ \hline \operatorname{sailor}(x)\ \operatorname{enter}(x)\ \operatorname{wave}(x) \end{array}} ]
The indefinite noun phrase “a sailor” introduces (x), while the pronoun “she” is interpreted as dependent on that established referent. Features such as grammatical gender and number constrain resolution but do not constitute the complete interpretation, since discourse prominence and world knowledge can also distinguish among accessible candidates.
Formally, a DRS is commonly represented as an ordered pair:
[ K=\langle U_K,\operatorname{Con}_K\rangle, ]
where (U_K) is the universe of discourse referents and (\operatorname{Con}_K) is the set of conditions associated with them. Atomic conditions apply predicates or relations to referents. Complex conditions contain subordinate DRSs and express operators whose semantic effects cannot be represented as simple predicates.
A DRS has truth conditions relative to a model theory and an assignment. The structure is true when the current assignment can be extended to map its discourse referents to entities satisfying all of its conditions. This interpretation preserves a connection with classical logic while allowing the representation language to encode dependencies that arise during discourse processing.
Accessibility and anaphoric dependence
Accessibility is a structural relation between DRSs. A referent can serve as an antecedent for a later pronoun only when the DRS introducing that referent is accessible from the DRS containing the pronoun. Referents introduced in the main discourse generally remain accessible to subsequent material, whereas referents confined to negated or hypothetical structures generally do not become available outside those structures.
The contrast appears in the following discourses:
A navigator arrived. She carried a chart.
No navigator arrived. She carried a chart.
In the first discourse, the indefinite introduces a referent into the principal DRS, and the pronoun can depend on that referent. In the second discourse, the semantic contribution associated with “a navigator” occurs inside the scope of negation. The resulting referent is unavailable to the following assertion under the intended dependency, because the discourse has not established an actual navigator in the principal information state.
Conditionals create a related but nonidentical pattern:
If a vessel enters the harbor, its signal is recorded.
The antecedent introduces a vessel within the conditional environment. That referent is accessible in the consequent, allowing “its” to depend on the vessel. It does not ordinarily remain accessible after the entire conditional, because the discourse has asserted a dependency between hypothetical situations rather than the existence of a particular vessel.
Accessibility therefore supplies a structural account of anaphoric possibility, but it does not select a unique antecedent whenever several accessible referents satisfy the relevant grammatical constraints. Antecedent selection also interacts with information structure, lexical compatibility, rhetorical organization, and contextual knowledge. DRT separates these resolution factors from the prior question of whether a dependency is formally available at all.
Indefinites and quantification
In classical predicate logic, an indefinite noun phrase is frequently represented by an existential quantifier. DRT divides the corresponding contribution between the introduction of a discourse referent and the addition of conditions describing that referent. Existential force arises from the model-theoretic requirement that the assignment be extendable to an appropriate entity.
For example, “A captain inspected a vessel” introduces two referents and places conditions on both:
[ \boxed{ \begin{array}{c} x\quad y\ \hline \operatorname{captain}(x)\ \operatorname{vessel}(y)\ \operatorname{inspect}(x,y) \end{array}} ]
The associated truth conditions require the existence of an entity satisfying the captain condition and another entity satisfying the vessel condition, with the inspection relation holding between them. A continuation containing “the captain” or “it” can retrieve the corresponding referent if the lexical and grammatical constraints support that interpretation.
Universally quantified expressions receive a different structural treatment. A sentence such as “Every captain inspected a vessel” is represented through an implication whose antecedent introduces an arbitrary captain and whose consequent introduces a vessel inspected by that captain. The representation allows the vessel to vary with the captain, which corresponds to the narrow-scope reading of the indefinite:
[ \boxed{ \begin{array}{c} x\ \operatorname{captain}(x) \end{array}} \Rightarrow \boxed{ \begin{array}{c} y\ \operatorname{vessel}(y)\ \operatorname{inspect}(x,y) \end{array}} ]
The discourse referent (x) is accessible from the consequent because the consequent is evaluated relative to assignments satisfying the antecedent. The referent (y) is not thereby established in the surrounding discourse, since different qualifying assignments may associate each captain with a different vessel.
Donkey anaphora
Donkey sentences supplied one of the principal motivations for DRT. Their characteristic pronouns depend on indefinites that do not behave like ordinary existential antecedents:
Every officer who owns a boat maintains it.
A simplified DRT analysis places the officer and boat referents in the antecedent of an implication. The consequent contains the maintenance condition, and the pronoun is linked to the accessible boat referent:
[ \boxed{ \begin{array}{c} x\quad y\ \hline \operatorname{officer}(x)\ \operatorname{boat}(y)\ \operatorname{own}(x,y) \end{array}} \Rightarrow \boxed{ \begin{array}{c} \ \hline \operatorname{maintain}(x,y) \end{array}} ]
The structure receives approximately universal truth conditions: every assignment satisfying the antecedent must be extendable to one satisfying the consequent. The indefinite “a boat” consequently participates in a dependency with universal force, even though indefinites in simple affirmative clauses usually produce existential truth conditions. DRT derives this contrast from the position of the discourse referent rather than by assigning the indefinite a fixed universal interpretation.
Natural donkey sentences vary in strength. Some favor an interpretation under which every relevant boat must be maintained, whereas others permit an interpretation involving a contextually sufficient boat. These differences have produced refinements involving plural discourse referents, maximality, situation semantics, and pragmatic restrictions on quantificational domains.
Negation and embedded structure
Negation is represented as a condition containing a subordinate DRS. The sentence “A pilot did not depart” may allow different scope readings, but a representation in which the indefinite lies within negation takes the following general form:
[ \boxed{ \begin{array}{c} \ \hline \neg \boxed{ \begin{array}{c} x\ \hline \operatorname{pilot}(x)\ \operatorname{depart}(x) \end{array}} \end{array}} ]
This structure states that no extension of the current assignment satisfies the embedded conditions. The referent (x) remains internal to the negated DRS and cannot normally support anaphora in the subsequent main discourse.
When a referent is introduced outside negation, it remains available inside the negative environment. In “A pilot arrived and did not depart,” the pilot referent belongs to the principal DRS, while only the departure condition falls under negation. The distinction illustrates how accessibility tracks semantic scope without reducing anaphoric interpretation to linear word order.
Dynamic interpretation
DRT belongs to the broader family of dynamic semantics, in which meanings are characterized partly by their effects on a context. A discourse representation can be viewed as an information state that changes as linguistic material is interpreted. Assertions add referents or conditions, while embedded operators regulate which additions survive into the surrounding state.
This dynamic perspective is closely related to dynamic predicate logic, although the two frameworks differ in their primitive objects. Dynamic predicate logic modifies the interpretation of logical formulas so that variable assignments can carry information from one formula to another. DRT instead makes the evolving representational structure explicit and defines model-theoretic truth conditions for the completed structure.
The representational level permits semantic distinctions to be stated independently of any specific syntactic derivation. It also creates a separate interface at which presupposition resolution, discourse relations, and lexical information can affect interpretation. This separation does not remove the need for compositional analysis, because a linguistic theory must still specify how words and syntactic constructions contribute to the resulting DRS.
Presupposition and accommodation
In an anaphoric account of presupposition, a trigger such as the definite description in “The engineer returned” introduces a partial structure whose referent must be connected with the established discourse. If a suitable antecedent is accessible, the presupposed material can be bound to it. If none is available, the representation can be enriched through presupposition accommodation, subject to structural and contextual constraints.
Van der Sandt’s treatment represents presuppositions as anaphoric DRSs that initially occupy an embedded position and may be resolved at an accessible level. The possible resolution sites account for interactions between presupposition and operators such as negation or conditionals. A presupposition accommodated globally contributes information to the principal discourse, whereas accommodation within an embedded DRS restricts the information to that semantic environment.
Binding and accommodation can generate several formally available representations. Their interpretation depends on accessibility together with consistency, informativeness, and discourse organization. Later work has integrated these constraints with accounts of common ground and pragmatic inference.
Underspecification
Natural-language sentences often permit multiple scope relations that cannot be fixed solely by surface order. Fully constructing a separate DRS for every reading can duplicate large portions of semantic material. Underspecified discourse representation theory addresses this issue by representing constraints on structural relations without immediately selecting one complete configuration.
An underspecified structure contains labeled components together with ordering conditions that restrict how those components may be embedded. A resolved interpretation corresponds to a complete DRS satisfying those constraints. This approach became influential in computational semantics because it allows syntactic and semantic processing to proceed while scope ambiguities remain unresolved.
Underspecification also separates ambiguity from vagueness. Scope ambiguity involves distinct structural configurations, whereas vague content can persist within each configuration. The formal representation records the former through unresolved structural relations without treating every contextual indeterminacy as a separate logical reading.
Relation to other semantic frameworks
DRT shares foundational concerns with Heim’s file change semantics and with update-based accounts of meaning. Each framework treats discourse interpretation as dependent on a changing context, although they differ in notation and in the status assigned to intermediate representations. File change semantics describes contexts as files containing cards or assignment possibilities, while DRT describes them through referents and conditions organized into nested structures.
The framework also overlaps with situation semantics, particularly in later analyses of tense, modality, and donkey anaphora. Situation-based approaches evaluate expressions relative to partial situations rather than exclusively relative to complete possible worlds. DRT can incorporate situation or event referents into its representational universe, allowing temporal and modal dependencies to be expressed using the same general architecture as nominal anaphora.
In computational work, DRSs have been used as intermediate meaning representations connecting syntactic analysis with inference. Such systems require explicit treatments of lexical meaning, reference resolution, and scope, since the formalism alone does not determine those components. Translations from suitable DRSs into first-order formulas permit the use of theorem provers, while richer extensions address phenomena whose interpretation exceeds ordinary first-order expressibility.