Discourse comprehension is the cognitive process through which people understand connected spoken, written, or signed language as an organized account of people, objects, events, ideas, and intentions. Understanding a discourse involves identifying words and sentence meanings, maintaining relevant information across sentences, establishing who or what expressions refer to, generating warranted inferences, and constructing a coherent representation of the situation being described. This representation allows a person to answer questions, anticipate consequences, explain causes, and remember the essential meaning even when the original wording is forgotten. The foundational construction–integration account of Kintsch (1988) and the situation-model synthesis of Zwaan and Radvansky (1998) explain why discourse comprehension extends beyond the interpretation of isolated sentences.
Consider a three-sentence story: “Maya left the windows open before going to work. Dark clouds gathered during the afternoon. When she returned, the carpet was soaked.” Nothing explicitly states that rain entered through the windows. A comprehender can nevertheless link the developing weather, the open windows, and the wet carpet into a plausible event sequence. The inference depends on background knowledge, the causal organization of the events, and the reader's goal. Another explanation, such as a burst pipe, remains logically possible. Good comprehension involves constructing a defensible interpretation while remaining able to revise it when new evidence arrives. This interaction between linguistic information and knowledge is central to research on narrative inferences (Graesser, Singer, and Trabasso, 1994).
What Discourse Comprehension Means in Psychology
Discourse, text, and connected language
A discourse is an extended communicative sequence whose components acquire meaning through relationships with one another. It can be a conversation, a spoken explanation, a paragraph, a scientific argument, a news report, a narrative, a signed discussion, or an account delivered through words and images together. Researchers sometimes reserve “text” for a written product and use “discourse” more broadly for connected language in context, although terminology varies by tradition. The common psychological problem is that the meaning of a later expression depends partly on what has already been said and on what participants assume about the world.
The unit of study extends beyond a sentence boundary. “Jordan found the key. It was under the chair” requires reference resolution; “The experiment failed. Therefore, the team repeated it” requires a causal or justificatory relation; “She smiled, but her hands trembled” presents potentially conflicting evidence about an emotional state. The comprehension system integrates these connections into an evolving interpretation. A sequence of grammatically correct sentences can remain difficult to understand when these relations are missing, unexpected, or obscured. Conversely, ordinary speech often contains incomplete sentences and repairs while conveying an intelligible overall message.
Three representations: surface form, textbase, and situation model
A useful analytical distinction separates the surface form of a message, the propositions conveyed by its linguistic structure, and the represented situation. Surface form concerns the exact words and syntax; a textbase organizes the explicit semantic propositions and their relations; a situation model integrates those propositions with relevant knowledge about agents, time, space, causal relations, and goals. These are theoretical levels of analysis, rather than three independent storage boxes in the brain. In Kintsch's construction–integration model, interacting representations and constraints help explain how multiple activated meanings resolve into a coherent structure.
A person may remember that a visitor missed a train because of traffic without remembering which sentence stated the traffic delay or whether the narrator called the traveler a visitor, passenger, or colleague. That ability reflects the preservation of meaning across linguistic changes. Yet verbal details remain important when exact phrasing carries legal, scientific, poetic, or interpersonal significance. The balance between wording, propositional content, and event representation varies with tasks and goals; it cannot be reduced to a universal rule that people always retain gist and always discard form.
Comprehension as an evolving process
Discourse understanding unfolds while the message arrives. A listener cannot defer every interpretive decision until a full conversation is complete; a reader must connect the current sentence with a changing context while selecting what to maintain, retrieve, or revise. Early interpretations guide attention and expectations, but later information can alter the implied agent, sequence, or explanation. Understanding therefore involves both continuity and updating. The central empirical questions concern which connections are established immediately, which are inferred only when needed, and which are remembered after processing. These questions motivate competing theories and experimental paradigms.
Coherence: How Sentences Become a Meaningful Whole
Local coherence and referential continuity
Local coherence links nearby propositions through repeated topics, shared participants, semantic relations, and plausible transitions. In “The musician carried a violin onto the stage. The instrument was old,” readers connect “the instrument” with “a violin” even though the two expressions differ. The relation is established using lexical knowledge and discourse context. A sentence that unexpectedly introduces an unrelated object can interrupt the continuity of the current representation. Local coherence therefore concerns relationships among adjacent or closely related portions of discourse, including those whose referents are connected by category membership or a familiar event sequence.
Linguistic cohesion contributes useful cues: pronouns, repeated terms, synonymous descriptions, connectives such as “because” and “however,” temporal adverbs, and conventional paragraph structure. Cohesion and coherence refer to related but distinguishable properties. Cohesion is visible in the linguistic material, while coherence is a property of a comprehender's constructed interpretation. A highly cohesive paragraph can be misleading or logically inconsistent; a coherent conversation can omit explicit connectives because participants infer relationships from common knowledge. Research on inference making explains how readers bridge these gaps rather than requiring every connection to be spelled out Graesser et al. (1994).
Global coherence and the discourse's organizing question
Global coherence concerns how episodes and propositions contribute to a larger theme, explanatory structure, argument, or communicative goal. An article about coastal erosion can introduce waves, infrastructure, and planning decisions because each section contributes to a common account of environmental change. A narrative maintains a recognizable plot even when scenes shift in time and location. Readers construct a higher-level representation of which information matters, which outcomes follow from which events, and why an author has included a particular detail. This requires selective attention and integration over longer intervals than a single sentence.
Global coherence depends partly on genre and purpose. A recipe is organized around actions that achieve a practical outcome. A scientific report separates the research question, method, observations, and interpretation. A personal story often centers on agents' goals and consequences. A conversation coordinates a series of turns within participants' changing social purposes. The same sentence can contribute different information depending on which discourse structure contains it. The practical implication is that understanding a paragraph demands knowledge of how its ideas fit together, not merely recognition of its topic words.
Causal coherence: Why did this happen?
Causal relations frequently determine whether a narrative makes sense. In “The road froze overnight. The bus skidded at the intersection,” general knowledge supplies a possible causal link. Readers may also connect an earlier intention to a later action: someone who promises to bring medicine and subsequently visits a pharmacy is pursuing a recognizable goal. The explanatory relation may involve an enabling condition, a cause, a motive, or an outcome; these relations are distinct. Contemporary synthesis emphasizes how narrative causal structure shapes comprehension and memory, while leaving open which neural mechanisms support different kinds of causal learning Chen and Bornstein (2024).
Causal interpretation can be mistaken. Events that occur in sequence do not always stand in a causal relation, and a plausible story can conceal missing evidence. Scientific discourse comprehension therefore includes evaluating whether an inference follows from the premises, is a background assumption, or is an unsupported explanation. The very process that makes narratives intelligible can create confident but unwarranted inferences when readers import familiar scripts into an unfamiliar event. Successful comprehension involves identifying relations and monitoring their evidential strength.
Contrast, elaboration, and rhetorical relations
Discourse also organizes information through comparison, concession, explanation, exemplification, problem–solution structure, and evaluation. “The treatment improved sleep; however, daytime functioning did not change” signals a contrast between outcomes. “The device failed because the sensor overheated” supplies an explicit explanation. “In particular” introduces a narrower instance, whereas “therefore” asserts a conclusion. Connectives can narrow interpretive possibilities, but their effects depend on syntax, world knowledge, and the larger argumentative structure. An author can omit a connective while leaving a relation inferable, or insert a connective that encourages readers to treat a weak argument as stronger than it is.
The cognitive benefit of recognizing rhetorical relations is organizational: a reader understands which propositions support, qualify, exemplify, or challenge others. This is especially important for expository and scientific materials, where the main difficulty may arise from abstract relations rather than sequences of observable events. A complete account of discourse comprehension accordingly includes arguments and explanations as well as stories.
Reference Resolution: Who or What Does the Language Mean?
Pronouns, descriptions, and antecedents
Anaphora occurs when an expression depends on a previously introduced referent. “Leah lent Ava her bicycle. She needed it for work” presents multiple candidates for “she,” and contextual information guides interpretation. Resolving such references involves linguistic constraints, prominence, recency, grammatical information, event plausibility, and the listener's assumptions about the speaker's intentions. Pronouns are especially informative because their short forms often require substantial contextual work. The process is rapid, though the weight assigned to different cues varies by language, expression, and task.
Eye-tracking experiments demonstrate that listeners can exploit grammatical-gender information during the unfolding interpretation of pronouns. In Arnold, Eisenband, Brown-Schmidt, and Trueswell (2000), the time course of pronoun resolution was investigated through eye movements, showing how discourse interpretation can be studied before an overt answer is given. Such work establishes that reference tracking is a measurable online process. It does not imply that every ambiguous pronoun is immediately resolved or that all languages rely on the same combination of cues.
Accessibility, prominence, and discourse focus
Not all previously mentioned entities remain equally accessible. A main protagonist, a recently discussed object, or a participant currently central to an action can remain easier to retrieve than a peripheral detail. Repetition, grammatical position, topic continuity, and changes in the narrative scene can influence accessibility. When a new sentence begins “He finally arrived,” readers must evaluate which male participant is most relevant and whether earlier events justify this choice. The same description can be unambiguous within one discourse and confusing within another.
Reference processing helps maintain a stable cast of people and objects while the topic develops. It is also a frequent location for repair: “She — I mean the second researcher — contacted the clinic” explicitly updates the intended referent. Ambiguity may be resolved through a later clause, a speaker's gesture, or a listener's question. Discourse comprehension therefore includes both immediate resolution and recovery when the initial model proves incomplete.
Discourse-level ambiguity and brain responses
Electrophysiological research finds neural responses to referential problems that arise only from a wider discourse context. Van Berkum, Brown, Hagoort, and Zwitserlood (2003) examined event-related brain potentials associated with discourse-referential ambiguity in spoken language. Their results demonstrate why sentence-level semantic anomaly tasks alone cannot characterize all components of comprehension. A person can understand each individual word while facing uncertainty about the person, object, or event to which the utterance applies.
Referential failure has practical consequences. Instructions such as “Place it beside the other one” can work when shared visual attention identifies the objects and fail when two plausible alternatives are present. In technical writing, clear nominal references and stable terminology reduce unnecessary ambiguity. In ordinary conversation, effective reference may depend more on shared attention and quick clarification than on repeated explicit descriptions.
Inference: Understanding What Was Implied
Bridging inferences maintain coherence
A bridging inference supplies a connection needed to relate incoming information to the developing representation. “The picnic began in sunshine. Everyone hurried indoors when the thunder started” invites an inference about approaching stormy weather. If a reader cannot connect thunder, possible rain, and the decision to move, the two sentences remain adjacent but inadequately integrated. Bridging inferences commonly concern causal relations, event continuity, object identity, and information that a later sentence presupposes. They connect the explicit text without requiring a detailed imagined scene for every word.
An inference can be supported to different degrees. A strong bridge may be necessary for the story to make sense; an attractive elaboration may simply enrich it. Researchers assess inference activation with reading times, recognition probes, reaction-time priming, eye movements, and the ability to detect contradictions introduced later. Yet each method measures a different aspect of processing, and rapid recognition of a related concept does not automatically prove that the reader consciously represented a particular proposition.
Elaborative, predictive, and explanatory inferences
Elaborative inferences add plausible detail beyond what coherence requires. “The hiker packed a flashlight” can activate expectations about darkness, a cave, a power failure, or an overnight trip; none is guaranteed by the sentence. Predictive inferences concern what may happen next, while explanatory inferences connect events with reasons or enabling conditions. Their occurrence depends on contextual constraints, the strength of available knowledge, and task goals. A reader answering a causal “why” question may construct a richer model than someone scanning for a departure time.
The constructionist account of Graesser, Singer, and Trabasso (1994) emphasizes a search for meaning that fits goals, coherence, and explanation. It provides a theoretical explanation of why some inferences become central to narrative understanding. It does not establish that every conceivable inference is automatically generated during reading. Comprehenders are selective, and a richly coherent retrospective explanation may contain connections that arose only after a direct question prompted them.
The minimalist hypothesis and the scope of automatic inference
McKoon and Ratcliff (1992) advanced a minimalist hypothesis: automatic inference tends to favor readily available information and what is required for local coherence, while more elaborate inference may depend on strategic or goal-directed processing. The resulting theoretical dispute concerns timing and conditions, rather than the simple existence of inferences. Constructionist and minimalist models can make different predictions about when a distant causal explanation should become accessible during ordinary reading and when it should emerge only after prompting.
An evidence-based synthesis recognizes a graded continuum: some referential and bridging computations are rapid, others require effort or re-reading, and still others appear primarily during later recall or evaluation. Task instructions, text length, knowledge, and measurement windows all matter. Apparent disagreement among findings can reflect distinct operational definitions of inference generation. The useful scientific question is which inference is active, at what time, under which reading goal, and according to which experimental measure.
Inferences, misinterpretation, and source monitoring
An inferred cause can later be misremembered as explicitly stated. When a report says that a laboratory lost power and a sample was spoiled, readers may infer that the outage spoiled the sample; a later test must distinguish this inference from the actual wording. Discourse comprehension includes such constructive processes, so memory tests should separate verbatim recognition, recall of stated propositions, and responses that depend on reasonable but unstated implications. In journalism, policy, and clinical communication, this distinction is important because a compelling narrative can make an unverified relationship feel established.
An effective comprehension check asks both “What does the text say?” and “What does the text suggest, and on what grounds?” The former tests explicit representation; the latter assesses inference and justification. Readers who can articulate the evidential basis for a connection are better positioned to notice alternative explanations or contradictions. This does not eliminate inference, which is indispensable, but makes it more transparent.
Situation Models: Representing the Described World
The event model as an integrated representation
A situation model organizes the state of affairs to which language refers. In a story about an engineer repairing a bridge, the representation includes agents, their goals, the physical location, the sequence of repair activities, and the causal significance of a discovered defect. These elements form a structure that can be updated as the story proceeds. The model can be built from prose, conversation, audiovisual narrative, or other meaningful inputs. In Zwaan and Radvansky's (1998) review, situation models provide a framework for explaining language comprehension and subsequent retrieval of event information.
A situation model is useful because many later questions concern relationships that are never expressed in a single sentence: who knew a fact at a certain point, why an action changed, whether an object is still in a room, or what would follow if a condition changed. Such judgments require the reader to consult the developing event representation. The model is also fallible. A reader may retain the broad story while losing a location change or incorrectly carrying a previous character's goal into the next scene.
Time, space, protagonists, causality, and intention
The event-indexing model of Zwaan, Langston, and Graesser (1995) describes narrative events as connected along dimensions of time, space, protagonist, causality, and intentionality. When a story jumps to another year, city, character, causal episode, or goal, the comprehender may have to update the active situation model. Multiple changes can increase the demands of integrating the new passage with prior context, although effects depend on the design and task.
These dimensions are analytically separable but interact in real narratives. A change of place may coincide with a new goal; a time jump may make earlier causal information newly relevant. Readers can compensate with genre knowledge, explicit transition phrases, or a stable protagonist. Situation-model theory therefore provides more than a list of narrative features: it generates questions about what is represented, when it becomes accessible, and how discontinuities influence attention, reading time, and later memory.
Narrative perspective and the limits of simulation
Readers sometimes mentally adopt the spatial or psychological perspective of a narrator or character. A scene described from inside a vehicle may privilege different objects than an overhead description of the same route. Whether such perspective-taking involves detailed perceptual simulation, abstract relational coding, or a mixture depends on the materials and method. Mental imagery is one possible resource for discourse comprehension, but constructing a situation model does not require vivid visual imagery in every individual or modality.
Likewise, understanding a character's belief does not require endorsing that belief. A reader can track a protagonist who mistakenly thinks a key is in the drawer while retaining the information that the key was moved. This demands keeping perspectives and knowledge states organized. Such representational flexibility connects discourse comprehension to social cognition, yet the basic ability to track fictional beliefs should be analyzed separately from individual clinical diagnoses or broad theories of interpersonal competence.
Major Psychological Models of Discourse Comprehension
Construction–integration theory
Kintsch's construction–integration framework proposes that incoming linguistic information activates candidate meanings, propositions, and knowledge-based associations; an integration process then strengthens mutually compatible information and weakens inconsistent alternatives. The distinction helps explain why an ambiguous word may initially activate several possibilities before context favors one, and why a distant earlier sentence can influence the interpretation of a later proposition. The theory combines a representational account of propositions with a computational account of mutual constraint and activation Kintsch (1988).
Its explanatory scope includes coherence formation, the influence of background knowledge, and the relation between immediate processing and later recall. Some details of the original implementation are model-specific: demonstrating that readers integrate context does not prove the brain literally implements a particular spreading-activation algorithm. The model's enduring contribution lies in treating comprehension as the construction of a constrained representation rather than a passive sequence of word lookups.
Constructionist and minimalist inference theories
The constructionist perspective emphasizes goal-driven explanation, global coherence, and inferences motivated by the desire to understand why events occur. The minimalist perspective predicts that automatic inferences are more restricted, with goal-directed reasoning accounting for many richer interpretations. Comparing them requires experimentally separating automatic activation, prompted explanations, and longer-term memory. Findings in favor of one mechanism under one task do not disprove the contribution of the other mechanism in a different situation Graesser et al. (1994); McKoon and Ratcliff (1992).
Both perspectives benefit from explicit statements about reading goals and timescales. A fast semantic probe immediately after a sentence asks a different question from an open-ended explanation after an entire story. When experimental tasks differ, theoretical labels can obscure the fact that the underlying evidence concerns different stages of interpretation. Scientific synthesis is strongest when predictions specify the temporal window, stimulus type, and inferred content.
Event-indexing and event-segmentation approaches
Event-indexing accounts describe the changing dimensions along which a narrative situation is monitored, while event-segmentation theory addresses how continuous experience is parsed into meaningful episodes. Both concern temporal structure and updating, though one emphasizes dimensions within a constructed narrative and the other emphasizes transitions between event representations. Zacks, Speer, Swallow, Braver, and Reynolds (2007) synthesize evidence that event boundaries organize perception and memory, proposing a role for changes in ongoing predictions.
These frameworks apply most directly when discourse represents sequences of events. An abstract proof or philosophical argument may require integration of propositions whose importance is logical rather than chronological. Consequently, models of event comprehension and models of discourse generally overlap without having identical explanatory domains. Comprehensive psycholinguistics must represent both kinds of organization and specify the contexts in which each theoretical framework is most informative.
Predictive, constraint-based, and good-enough interpretations
Context can make upcoming referents, events, and words more or less expected. Predictive accounts propose that a comprehender uses prior information to anticipate possible continuations. Constraint-based approaches emphasize the simultaneous influence of syntax, semantics, probability, and discourse information. Good-enough interpretations describe circumstances in which people maintain a useful interpretation that is incomplete or partially inaccurate, especially under limited attention or when the current task does not demand exhaustive analysis. These labels concern different explanatory levels and should not be treated as competing all-or-nothing positions.
A reduced neural response to an expected word can be consistent with easier contextual integration, anticipatory processing, or both, depending on the experiment. Predictive-processing claims become stronger when designs distinguish pre-activation before the word appears from the reduced cost of integrating a word after it appears. Discourse-level electrophysiological findings show how rapidly wider context influences meaning, but they do not by themselves establish a single universal algorithm for prediction van Berkum, Hagoort, and Brown (1999).
How Comprehension Unfolds in Real Time
From a new word to an updated interpretation
During continuous reading or listening, a word enters a representation that already contains partial expectations and retrieved context. The reader identifies its likely meaning, connects it with grammatical structure, determines whether it introduces a new entity or refers back to an existing one, and evaluates how its proposition changes the discourse. These operations can overlap and interact. A phrase such as “the other witness” presupposes an earlier witness or a mutually understood set of witnesses; comprehension can fail if no appropriate referent is available.
Many interpretive decisions are provisional. “The driver told the mechanic that he…” creates an ambiguous reference, which may be constrained by the rest of the sentence. “After the hearing” can signal a time shift whose full relevance becomes clear only later. The sequence of mental operations is therefore partly incremental and partly reconstructive. A model that predicts an immediate, perfectly specified interpretation for every fragment would underestimate ambiguity and revision.
Evidence from the N400 and contextual meaning
The N400 is an event-related potential component sensitive to aspects of semantic processing and contextual fit. In influential experiments, van Berkum, Hagoort, and Brown (1999) showed that words congruent with an individual sentence can produce increased N400 responses when they violate the wider story. A later spoken-language experiment likewise showed discourse-level semantic information affecting the N400 during unfolding sentences van Berkum, Zwitserlood, Hagoort, and Brown (2003).
These studies establish that discourse context affects relatively early semantic processing, rather than being consulted only after sentence completion. However, an ERP effect is a measured neural signal, not a direct image of a fully formed conscious prediction. Its amplitude can reflect several interacting factors, including expectancy and integration demands. Experimental interpretation must therefore distinguish what the data directly establish from the computational mechanism proposed to explain them.
Repair, contradiction, and model revision
An incoming fact can contradict a previously established situation model. Suppose a story presents a character leaving a backpack in a café and later reports that the backpack was in the character's office. The reader may infer an omitted transfer, question the new information, or revise the earlier assumption. Whether the contradiction is immediately detected depends on how recently the initial fact was encountered, how central the object is to the story, and the reader's goals. Failure to notice a contradiction on a first pass does not mean that the reader lacks the relevant knowledge; it may reflect limited retrieval at the moment of integration.
Revision can also be triggered by an author's explicit correction or the recognition of unreliable narration. A coherent interpretation is consequently a dynamic achievement. Monitoring involves evaluating whether the current situation representation remains compatible with incoming claims and deciding when uncertainty warrants re-reading, questioning, or seeking additional information. For expository and scientific materials, contradiction detection is a key part of critical comprehension.
Event Boundaries, Attention, and Memory
How continuous narratives are segmented
Long stories and real-world experiences rarely arrive with objective labels identifying every episode. People nevertheless divide continuous activity into meaningful units, often near changes in location, actors, goals, or action patterns. The event-segmentation account of Zacks et al. (2007) proposes that event boundaries relate to changes in predictive processing and influence memory organization. A scene change in a film, an abrupt new action in a conversation, or a temporal marker in a written story may prompt an update in the active event model.
Event boundaries can help organize recall because information within a segment shares a local structure, while transitions mark relationships between segments. Their effect is not a guarantee of stronger memory for every detail; performance varies with attention, event structure, and testing conditions. The theory addresses the segmentation of ongoing experience generally, so applications to text must consider how words convey boundaries that perception might otherwise reveal through changes in movement or scene.
Narrative causality and selective remembering
Memorable events are frequently connected through causes and consequences. A detail that explains a protagonist's later choice can have a central role even when it is briefly mentioned. In their 2024 review, Chen and Bornstein discuss how causal structure guides comprehension, event memory, and proposed computational accounts of learning from narratives. This synthesis helps connect classic discourse psychology with current naturalistic neuroscience while clarifying that causal relations are properties of story structure and the comprehender's interpretation, rather than just temporal adjacency.
An event model can preserve both what happened and why it mattered. During summary, people typically select information that contributes to the central causal or argumentative organization of the material. In a technical failure report, the relevant event may be a sensor anomaly because it explains the subsequent shutdown; an entertaining side detail may be less important to the reader's practical goal. The selection of central information is therefore partly shaped by the question a person is trying to answer.
Prior Knowledge, Schemas, and Understanding
World knowledge supplies missing connections
Discourse often leaves predictable information unstated. The sentence “The server overheated, so the service went offline” relies on knowledge about the relationship between temperature and equipment function. A science student may recognize additional mechanisms, while a novice may only understand a general causal association. Knowledge changes which explanations are available and which gaps are noticed. This effect is fundamental to construction–integration theory, where prior knowledge influences the candidate associations and constraints from which a coherent representation emerges Kintsch (1988).
Schemas summarize recurring structures such as visiting a restaurant, conducting an experiment, or attending a trial. They support efficient inference by providing expected roles and sequences. Their utility has a corresponding cost: a comprehender may insert conventional details that the current discourse never stated. Reliable interpretation therefore requires the flexibility to use familiar schemas while checking whether they match the particular case. People can share a language yet construct different interpretations because they possess different knowledge or assume different background conditions.
Goals determine which relations are worth constructing
A reader studying for an examination, checking medication instructions, or enjoying fiction approaches the same language with different priorities. The student may retain definitions and causal explanations; the person following instructions must track actions, conditions, and timing; the fiction reader may attend to motivation and surprise. Goal differences can change inference generation, comprehension monitoring, and what is later remembered. The constructionist emphasis on meaning-seeking goals captures this variation without implying that deliberate effort is necessary for every linguistic link Graesser et al. (1994).
Task sensitivity complicates laboratory generalization. A short paragraph followed by a yes–no question can encourage processing tailored to that question. Naturalistic conversation imposes turn-taking and social demands absent from silent reading. A research design that does not specify its task can therefore yield results whose scope is unclear. Knowledge, motivation, attention, and communicative situation all belong in a complete account of discourse comprehension.
When prior knowledge creates bias
Knowledge-based inference sometimes converts probability into apparent certainty. A story may resemble a familiar social script, leading readers to overlook unusual facts or assume a conventional motivation. Causal coherence can be especially persuasive when a proposed explanation forms an intuitive chain. Critical discourse comprehension includes identifying which propositions were given, which came from prior knowledge, and which represent competing explanations. This is relevant to news, medical information, legal narratives, and scientific argumentation where an internally coherent account may still lack empirical support.
Teaching readers to identify evidence, uncertainties, and alternative explanations is consistent with this psychological framework, but the effectiveness of any specific training program must be tested rather than assumed. Interventions that improve comprehension in one age group, domain, or task may not generalize unchanged to another. Evidence on strategy combinations underscores the role of background knowledge and task design in reading comprehension outcomes Peng et al. (2024).
Narratives, Explanations, Arguments, and Conversation
Narrative comprehension
Narrative discourse organizes events through time, character, causality, and intention. The reader must often infer why an action occurred and how it changed another character's opportunities or beliefs. A beginning establishes a situation and expectations; later developments update the event model; an ending can resolve open questions or deliberately leave them unsettled. Narrative comprehension thus offers a rich setting for studying how people maintain representations over long intervals. Models of causal coherence and event indexing are particularly informative for this genre Zwaan, Langston, and Graesser (1995); Chen and Bornstein (2024).
Literary narrative adds additional challenges: unreliable narrators, shifts in perspective, irony, nonlinear chronology, and underdetermined motives. Readers can construct multiple compatible interpretations when the discourse provides insufficient evidence for a single solution. Such cases expose a fundamental property of comprehension: it generates an organized representation under uncertainty, rather than simply retrieving a uniquely encoded meaning. Literary sophistication and scientific inference are different contexts in which this general capacity is exercised.
Expository and scientific comprehension
Expository texts explain concepts, mechanisms, classifications, or arguments. Understanding an article on climate systems may require connecting a general principle to several examples, interpreting a contrast, and distinguishing evidence from interpretation. These operations rely on coherence and inference but place greater emphasis on abstract conceptual relations, unfamiliar terms, and the hierarchy of claims. A student can recognize each sentence and still fail to form the correct global explanation when concepts are not connected.
A scientific text also has epistemic structure. A hypothesis, experimental result, interpretation, and conclusion play distinct roles. Causal language must be evaluated against the study design. A correlation described in one paragraph does not become proof of causation merely because a subsequent paragraph offers a plausible mechanism. Scientific discourse comprehension is therefore a combination of linguistic integration, conceptual knowledge, and reasoning about evidence. This article treats these capacities as interconnected while reserving specialized learning-disorder diagnosis for its established clinical domain.
Dialogue and shared context
Conversations introduce speaker turns, changing intentions, reference to a shared physical setting, interruptions, and rapid correction. A statement such as “The blue one?” may be comprehensible because speakers are jointly examining objects. Dialogue partners also use prior turns to decide what a response answers, challenges, or clarifies. The discourse model includes not only described events but an unfolding interaction: who proposed an idea, which question is active, and which information is mutually available.
Social norms and interpersonal goals influence conversational interpretation. Nevertheless, cognitive discourse comprehension and applied conflict management are distinct scientific questions. The former investigates how connected utterances acquire meaning; the latter may focus on negotiation, emotion regulation, power, or relationship outcomes. The shared context provides grounds for cross-disciplinary connections without transferring every communication problem into psycholinguistics.
Listening, reading, signing, and multimodal messages
Discourse comprehension occurs across modalities. Listening is constrained by the temporal flow of sound, intonation, pauses, and the possibility of overlapping speakers; reading offers visual layout, punctuation, and often the opportunity to re-read; signed discourse includes linguistic structure together with spatial reference and visual attention. Captions, diagrams, gestures, and videos can supply relevant context, but they can also compete for attention. The modality changes how information arrives and which cues support coherence, without removing the need to track referents and build an organized interpretation.
Prosodic emphasis may disambiguate contrast or mark an important referent. Paragraph boundaries and headings may signal shifts in topic. A map accompanying travel instructions can externalize spatial relations that would otherwise need to be inferred from prose. The scientific question is how these cues are integrated under specific tasks, not whether one medium is universally superior. Controlled experiments, naturalistic observation, and accessibility research can each illuminate different parts of that question.
The Cognitive Neuroscience of Connected Language
Language networks interact with wider cognitive systems
Sentence-level language processing involves neural systems strongly associated with linguistic representations, while extended narrative comprehension also recruits systems involved in memory, event organization, and the representation of situations. Naturalistic neural studies show that connected language produces coordinated responses across regions and temporal scales. Silbert, Honey, Simony, Poeppel, and Hasson (2014) compared neural responses during the production and comprehension of extended narrative speech, reporting distributed and partly coupled activity across speakers and listeners.
Such coupling does not establish that two individuals have identical private experiences or that their interpretations are perfectly aligned. It indicates measurable correspondence in responses to a shared communicative sequence. A brain network identified during narrative listening also need not be devoted exclusively to language, because memory, attention, and situation construction are active during other meaningful experiences. Anatomical claims should therefore specify tasks, contrasts, and analysis methods.
Hierarchies of processing timescales
Some neural responses track rapidly changing acoustic or lexical input, while others preserve information over longer stretches of connected speech. Lerner, Honey, Silbert, and Hasson (2011) investigated a hierarchy of temporal receptive windows using a narrated story. Their findings support the idea that meaningful language is processed through interactions among representations integrating information over different durations. A discourse model can thus retain an earlier topic or event while new words are interpreted over shorter intervals.
A temporal hierarchy should not be confused with a literal one-way conveyor belt in which every brain area performs exactly one stage. Neural measurements provide temporally aggregated or indirect signals, and different experimental analyses operationalize “integration window” differently. The robust insight is that extended comprehension requires retaining and connecting information across multiple timescales; the precise computational organization remains a research question.
Hippocampal contributions and event organization
Recent research connects narrative comprehension with hippocampal systems supporting event encoding and relational integration. In an fMRI study involving a temporally scrambled movie, Park, Song, and Shim (2025) analyzed memory for event content and reconstructed order, finding different predictive relationships for hippocampal connections with posterior medial and ventromedial prefrontal regions. The authors used model-based indices to identify narrative coherence and sequencing moments. These findings provide evidence for differentiated memory processes during narrative reconstruction under the study's design.
The findings are promising rather than a complete map of discourse comprehension in every modality. Reconstructing a scrambled audiovisual narrative is a special task that engages inference and memory; the same functional-connectivity patterns cannot automatically be generalized to ordinary reading, conversation, or all populations. A correlation between connectivity and performance also does not establish that a particular pathway is necessary or sufficient. Converging lesion, behavioral, and intervention evidence remains important for stronger causal interpretation.
Individual Differences and Development
Vocabulary, knowledge, and working-memory demands
People vary in word knowledge, processing fluency, attention, memory retrieval, and familiarity with a topic. These differences can alter how many discourse relations become accessible and how effectively a person maintains a changing representation. Vocabulary supports interpretation of explicit propositions, while topic knowledge makes plausible causal or referential bridges easier to construct. Working-memory demands rise when several entities, time shifts, or embedded explanations must be coordinated, although their effects are intertwined with familiarity and the availability of external cues.
A poor answer to a story question can have multiple explanations: unfamiliar vocabulary, missed pronoun reference, an absent inference, difficulty maintaining earlier information, insufficient knowledge, or a question that assumes more than the text justified. The same surface score need not reveal a single underlying mechanism. Research and educational assessment therefore benefit from tasks that distinguish explicit recall, inference, event ordering, and explanation.
Development of coherent discourse understanding
Children acquire increasing skill in tracking protagonists, using connectives, drawing inferences, and organizing narratives as language knowledge and experience expand. Development occurs through interacting changes in vocabulary, grammar, knowledge, attention, and participation in conversation and reading. An age difference on a comprehension task is not automatically evidence of a discrete developmental stage; tasks differ in linguistic complexity, assumed knowledge, and response requirements. Studies that compare cohorts or longitudinal trajectories must account for these factors.
Adults support comprehension when they clarify who a pronoun refers to, explain why two events are connected, discuss the meaning of an unfamiliar word, or invite a child to retell the central causal sequence. These are conceptually plausible supports that align with the mechanisms described here. The efficacy of specific instructional programs requires independent evaluation. Research on strategy instruction indicates that outcomes depend on combinations of techniques and background-knowledge support rather than a single universally superior approach Peng et al. (2024).
Aging and context-dependent compensation
Discourse comprehension across adulthood reflects interacting changes in knowledge, goals, sensory access, and processing resources. A person may draw effectively on extensive experience while finding it harder to track a dense passage with many competing referents. Early theoretical work by Hasher and Zacks (1988) highlighted inhibitory and working-memory accounts of age-related discourse performance. Later research places increasing emphasis on variation among people and tasks, making broad assertions about uniform age-related deterioration scientifically inadequate.
Clinical research sometimes investigates narrative comprehension in neurocognitive disorders. A systematic review of studies involving Alzheimer's disease and mild cognitive impairment identified difficulties on several discourse measures and called for further evaluation of their predictive utility. Such findings concern selected clinical samples and research instruments. They do not transform ordinary forgetting, slow reading, or difficulties with one story into diagnostic indicators. Diagnosis and clinical management belong to appropriately validated clinical assessment pathways.
Multilingual comprehension and cultural knowledge
Comprehending discourse in a second language can place additional demands on vocabulary access, grammatical processing, idiomatic knowledge, and cultural background. Readers can possess strong reasoning and discourse skills in one language while struggling with a passage that assumes unfamiliar vocabulary or institutional conventions in another. Testing connected-language understanding therefore requires attention to the language of assessment and to the knowledge presupposed by the text. Differences should not be assigned to a single generic bilingual cognitive advantage or disadvantage.
Narratives and explanatory genres may be organized differently across communities and communicative practices. Successful understanding still requires building a representation of relevant entities and relations, but linguistic cues and genre expectations vary. Comparisons that do not match proficiency, literacy history, topic familiarity, and task requirements risk attributing observed differences to bilingualism when another factor is responsible. This issue is especially important in educational contexts, where the instructional language may be less familiar than the student's strongest language.
How Researchers Measure Discourse Comprehension
Online behavioral methods
Self-paced reading records how long readers spend on successive text segments, enabling researchers to study costs associated with reference resolution, contradiction, or event shifts. Eye tracking provides fixation and regression measures that can reveal when attention returns to an earlier phrase or how listeners use context while looking at relevant objects. Probe recognition and priming tasks assess whether an implied concept or referent is accessible at a particular point. These methods are powerful because they constrain claims about processing time, but each requires careful controls for word length, lexical frequency, stimulus predictability, and task-specific strategies.
Reading-time differences indicate changes in processing demand rather than identifying a unique mental operation. A slower sentence could reflect unusual vocabulary, syntactic complexity, effortful inference, or reanalysis. Eye movements can index attention while remaining an indirect measure of mental representation. Experiments are strongest when competing theories make distinct predictions and multiple measures converge on the same interpretation.
Offline comprehension and memory measures
Retelling, summarization, main-idea extraction, event ordering, and questions about explicit and inferred information assess what people can reconstruct after processing. A carefully designed comprehension test distinguishes literal facts from bridging inferences and wider situation-model understanding. A retelling may show whether readers have grasped the central causal organization, but it also depends on expressive-language ability, memory, and willingness to elaborate. A multiple-choice question may be answered through recognition or elimination without reflecting the same representation that supports a free explanation.
Assessment quality depends on specifying the construct. A task that asks only for names and dates measures a limited aspect of discourse memory. A task requiring explanation of motives, evidence, and event relations samples different abilities. Comparing such measures can reveal whether a person retained individual propositions yet missed the global structure, or understood the meaning but struggled to reproduce details. Results require interpretation in light of language proficiency and response format.
Electrophysiology and neuroimaging
Electroencephalography and event-related potentials provide time-sensitive measures of responses to linguistic and referential information. The N400 studies of van Berkum and colleagues (1999; 2003) demonstrate how the wider discourse can shape meaning processing early in a sentence. Functional MRI offers spatially distributed measures of activity and connectivity across extended narratives, including the timescale and event-memory findings of Lerner et al. (2011) and Park et al. (2025).
No single brain method reads the content of a person's complete mental model directly. ERP components require interpretive models, and fMRI captures indirect signals with temporal limitations. Naturalistic tasks improve relevance to extended discourse but increase the difficulty of separating lexical, emotional, visual, and causal variables. Controlled experiments provide cleaner manipulation of selected factors while simplifying communication. A cumulative account draws on both.
Naturalistic stories and computational analysis
Researchers increasingly use stories, films, podcasts, and conversations to capture information integration over realistic durations. Computational methods can annotate changes in characters, event boundaries, causal connections, lexical surprisal, and semantic similarity. These measurements help formulate hypotheses about where comprehension becomes easier or harder. The value of the method depends on whether its variables correspond to the psychological processes being claimed. A text model that assigns high semantic similarity to two sentences need not have represented their actual causal relation.
An important distinction concerns the evidence provided by models. A model's prediction of reading time or neural response establishes a relationship between its computed feature and a human measurement under specific conditions. It does not automatically validate the model as a complete account of human understanding. Contemporary work using language-model-derived indices of narrative coherence illustrates the promise of such analyses while underscoring their dependence on task and operationalization Park et al. (2025).
Discourse Comprehension, Literacy, and Practical Communication
Reading comprehension and discourse comprehension
Reading comprehension involves building meaning from written language, with word recognition and literacy processes contributing to access to the text. Discourse comprehension describes the integration of connected meanings across sentences and communicative contexts, including spoken and signed language. The two domains overlap strongly when a reader interprets a paragraph or story. Yet difficulties in written decoding are analytically distinct from weaknesses in inference, reference tracking, or knowledge integration. Clinical dyslexia and other specific learning disorders have their own diagnostic frameworks and canonical scientific ownership.
Instruction focused on comprehension can address main ideas, text structure, inference, retelling, and monitoring. A 2024 Bayesian network meta-analysis of comprehension strategies for students with reading difficulties found that combined strategies and background-knowledge instruction mattered; adding more strategies did not reliably produce greater benefits. These findings support intentional instructional design, with attention to the populations and tasks studied, rather than a universal recipe that can be assumed effective for every reader.
Making connected explanations easier to understand
A clear explanation introduces entities before referring to them ambiguously, signals shifts in time and topic, and makes central causal relations explicit where a reader could otherwise misinterpret them. A document can improve coherence by identifying its organizing question, introducing essential background knowledge, and distinguishing claims, evidence, examples, and conclusions. This is an application of discourse theory: it reduces unnecessary inferential work and makes the intended global structure easier to reconstruct. Clear headings and paragraph transitions help readers locate a relationship when they need to re-read.
These principles are strongest when applied flexibly. Repeating every referent by its full name can make prose awkward, and excessive connective phrases can obscure the argument. The goal is to provide the linguistic and conceptual cues that the audience actually needs. Scientific clarity also requires explicitly separating observations from hypotheses, because smooth causal storytelling can make tentative explanations seem more conclusive than the data warrant.
Checking whether a person understood
Asking someone to repeat a sentence verbatim tests different abilities from asking them to explain what happened, why an event mattered, or what evidence supports a conclusion. A useful comprehension check may ask the reader to identify the central question, paraphrase the causal sequence, resolve an ambiguous referent, and distinguish an explicit statement from an inference. Misunderstandings can then be traced to their source: vocabulary, missing knowledge, ambiguous language, or an unsupported inference. In conversation, clarification can repair a problem before it becomes an entrenched error.
Checking comprehension should avoid equating fluency, eye contact, confidence, or speed with understanding. People can produce persuasive summaries while missing important constraints, and slow responses can reflect careful evaluation rather than lack of comprehension. The scientifically appropriate approach is to sample the specific representations and reasoning that the task requires, using transparent questions and relevant context.
Connected Language and Large Language Models
What computational comparison can establish
Language models can assign probabilities to word sequences, generate summaries, track some referential dependencies, and answer questions about multi-sentence passages. These capabilities provide useful tools for constructing stimuli and comparing computational predictors with human behavior. They also introduce a methodological challenge: accuracy on discourse questions can reflect patterns learned from training distributions, sensitivity to surface cues, or task-specific regularities. A benchmark result does not by itself reveal the representations or cognitive mechanisms that produced it.
Human discourse comprehension is studied through convergent evidence from behavior, development, memory, conversation, neural responses, and controlled manipulations. Computational systems can be tested against some of these measures, but text output alone cannot establish that a model shares the human route to reference tracking, causal inference, event simulation, or lived subjective experience. These are separate empirical and philosophical questions. The proper research comparison specifies input conditions, outcomes, error patterns, and the explanatory level being tested.
Using models to annotate narratives
A recent example comes from Park, Song, and Shim (2025), who employed a pretrained language model to help locate moments associated with narrative coherence in a study of event sequencing and hippocampal connectivity. The resulting measure provided one analytic tool within a larger behavioral and neural investigation. Its success as a predictor does not make the model a direct observation of mental content or proof that the computation is implemented identically in the human brain.
Computational annotation can be particularly useful when naturalistic narratives contain too many events for entirely manual coding. Researchers still need human validation, carefully separated training and test materials, and controls distinguishing coherence from topical similarity. Claims about causal reasoning require explicit tests of interventions, counterfactual sensitivity, and alternative explanations. Discourse comprehension offers productive shared benchmarks for people and artificial systems precisely because the underlying competence has many interacting components.
Evidence Strength, Open Questions, and Scientific Limits
Well-established findings
Connected-language understanding depends on information extending beyond the current sentence. Reference tracking, contextual semantic integration, coherence building, and knowledge-supported inference can be demonstrated experimentally. Situation models and event structures provide well-developed explanatory frameworks supported by convergent behavioral and memory evidence Kintsch (1988); Zwaan and Radvansky (1998); van Berkum et al. (1999). The existence of discourse-level effects is more secure than the claim that a single computational model fully explains them.
Contested and method-dependent questions
Researchers continue to debate how frequently predictive or elaborative inferences occur automatically, how richly situation models are specified, how context constrains meaning at the earliest stages, and how processing strategies change with goals or tasks. Naturalistic neuroimaging raises further questions about which neural patterns reflect language-specific integration versus domain-general memory and event processing. Many studies use a particular language, participant population, and stimulus type; results should be generalized in proportion to this evidence. Theories can be compatible on one result and diverge on a carefully selected experimental prediction.
Preliminary connections to broader computational theories
Recent narratives-and-neuroscience work connects causal organization with value-guided learning and credit assignment Chen and Bornstein (2024). Such connections offer promising theoretical links between story understanding, memory, and models of action. They are not an established universal mechanism for all comprehension. Studies of hippocampal connectivity and machine-derived coherence measures likewise offer informative task-specific evidence Park et al. (2025), with causal and cross-domain generalization remaining open.
Clinical boundaries and responsible interpretation
Discourse tasks can reveal differences in memory, language, or inferential performance in selected clinical groups, but performance on a narrative task is not a stand-alone diagnosis. Hearing differences, second-language proficiency, fatigue, literacy experience, educational opportunity, task anxiety, and background knowledge can affect results. Aphasia, developmental language disorder, dyslexia, neurocognitive disorders, and other clinical conditions require assessment appropriate to their distinct diagnostic criteria. A psycholinguistic description of an underlying mechanism cannot substitute for clinical evaluation or authorize a treatment claim.
Frequently Asked Questions About Discourse Comprehension
Is discourse comprehension the same as understanding sentences?
Sentence comprehension supplies essential lexical, grammatical, and propositional information, while discourse comprehension connects that information across sentences and communicative turns. A person can interpret each isolated sentence correctly yet misunderstand the overall sequence if a pronoun is assigned to the wrong character or a crucial causal link is missed. The relationship is hierarchical and interactive: sentence interpretation informs discourse construction, and discourse context guides the interpretation of incoming sentences.
What is the difference between cohesion and coherence?
Cohesion concerns linguistic signals of connection, such as pronouns, repetition, and connectives. Coherence concerns how a comprehender organizes a meaningful relationship among the ideas and events. A passage may contain cohesive markers yet express an implausible argument; a conversation can be coherent despite leaving several logical links implicit. Writers use cohesion to support coherence, while listeners and readers also construct coherence through inference and knowledge.
Why do people infer information that was never stated?
Connected discourse is economical: speakers and writers ordinarily omit information their audience can supply. Inferences connect stated events to prior context and world knowledge, allowing the reader to understand causes, references, and goals. They are essential to fluent comprehension, but some go beyond what the evidence guarantees. Careful understanding includes recognizing when a conclusion is a plausible interpretation rather than an explicit fact or necessary implication Graesser et al. (1994).
What is a situation model, and why does it matter?
A situation model is the integrated mental representation of the state of affairs described by discourse. It allows a reader to track people, locations, events, time, causal relationships, and intentions. This representation supports event ordering, question answering, explanation, and transfer of a story's meaning into a new context. It is a theoretical construct supported by behavioral evidence, not a literal photograph or universally complete simulation of the described world Zwaan and Radvansky (1998).
Does good discourse comprehension require strong working memory?
Maintaining and updating several pieces of information can impose memory demands, especially in lengthy or unfamiliar texts. Yet performance also depends on vocabulary, topic knowledge, organization, attention, and the availability of cues or re-reading. Strong background knowledge can make complex relationships easier to integrate, while poor text structure can burden even knowledgeable readers. Working memory is one contributor within an interacting system, not a stand-alone measure of understanding.
Does the brain predict every next word?
Context shapes the ease of interpreting upcoming information and can support predictions, but the strength and specificity of prediction vary. Some findings that are consistent with prediction may also reflect easier integration after a word appears. Rapid discourse-sensitive N400 responses demonstrate contextual effects without proving that every word is consciously or unconsciously preselected in advance van Berkum et al. (1999).
Can discourse comprehension be improved?
Instruction can target identification of main ideas, relationships among events, inference, text structure, retelling, and monitoring of understanding. The usefulness of specific strategies depends on readers' needs, prior knowledge, age group, and materials. A Bayesian network meta-analysis found that the effects of reading-comprehension strategy combinations were sensitive to background-knowledge instruction and did not support a simple “more strategies is always better” rule Peng et al. (2024).
Can a language model's accurate summary prove human-like comprehension?
An accurate summary demonstrates performance on a specified output task. It does not settle which internal representations the system used or whether its mechanisms resemble human discourse processing. Comparisons require matched tasks and explanatory tests; the question of subjective experience is separate again. Computational language models are valuable scientific tools, but their output quality alone cannot answer every psychological or philosophical question about understanding.
Synthesis: Connected Language as an Integrated Cognitive Achievement
Discourse comprehension enables people to transform a sequence of expressions into an organized interpretation of events, ideas, and intentions. Its central operations include reference resolution, local and global coherence, causal and other inferences, construction of situation models, and updating as new information arrives. These operations are constrained by linguistic structure and supported by knowledge, attention, memory, goals, and communicative context. The most productive scientific accounts link fine-grained experiments to extended narrative behavior rather than treating either scale as sufficient by itself.
Classic theory explains why connected meaning requires representations beyond the sentence; contemporary behavioral and neuroscience research shows how event structure and temporal integration contribute to that process. The essential distinction is between understanding the words that were uttered and understanding the relationships those words establish across time and context. Research on discourse comprehension investigates how the mind constructs, tests, remembers, and revises those relationships, and how that capacity supports reading, listening, conversation, education, and scientific reasoning.
Related Articles
Cluster Language, Thought and Psycholinguistics — the parent scientific synthesis covering language processing, psycholinguistic mechanisms, and the wider research landscape.
Semantic Memory: Facts, Meaning, and Knowledge — the memory systems and knowledge representations that support discourse-level inference.
Tip-of-the-Tongue Phenomenon: Why Words Feel Just Out of Reach — a related investigation of lexical retrieval, distinct from the integration of extended meaning.
Pragmatics in Psycholinguistics: Meaning Beyond Literal Words — how context and speaker intentions contribute to linguistic interpretation.
References
Arnold, J. E., Eisenband, J. G., Brown-Schmidt, S., and Trueswell, J. C. (2000). The rapid use of gender information: Evidence of the time course of pronoun resolution from eyetracking. Cognition, 76(1), B13–B26. https://doi.org/10.1016/S0010-0277(00)00073-1
Chen, J., and Bornstein, A. M. (2024). The causal structure and computational value of narratives. Trends in Cognitive Sciences, 28(8), 769–781. https://doi.org/10.1016/j.tics.2024.04.003
Graesser, A. C., Singer, M., and Trabasso, T. (1994). Constructing inferences during narrative text comprehension. Psychological Review, 101(3), 371–395. https://doi.org/10.1037/0033-295X.101.3.371
Hasher, L., and Zacks, R. T. (1988). Working memory, comprehension, and aging: A review and a new view. Psychology of Learning and Motivation, 22, 193–225. https://doi.org/10.1016/S0079-7421(08)60041-9
Kintsch, W. (1988). The role of knowledge in discourse comprehension: A construction–integration model. Psychological Review, 95(2), 163–182. https://doi.org/10.1037/0033-295X.95.2.163
Lerner, Y., Honey, C. J., Silbert, L. J., and Hasson, U. (2011). Topographic mapping of a hierarchy of temporal receptive windows using a narrated story. Journal of Neuroscience, 31(8), 2906–2915. https://doi.org/10.1523/JNEUROSCI.3684-10.2011
McKoon, G., and Ratcliff, R. (1992). Inference during reading. Psychological Review, 99(3), 440–466. https://doi.org/10.1037/0033-295X.99.3.440
Park, J., Song, H., and Shim, W. M. (2025). Hippocampal systems for event encoding and sequencing during ongoing narrative comprehension. Communications Biology, 8, 954. https://doi.org/10.1038/s42003-025-08377-1
Peng, P., Wang, W., Filderman, M. J., Zhang, W., and Lin, L. (2024). The active ingredient in reading comprehension strategy intervention for struggling readers: A Bayesian network meta-analysis. Review of Educational Research, 94(2), 228–267. https://doi.org/10.3102/00346543231171345
Silbert, L. J., Honey, C. J., Simony, E., Poeppel, D., and Hasson, U. (2014). Coupled neural systems underlie the production and comprehension of naturalistic narrative speech. Proceedings of the National Academy of Sciences. https://pubmed.ncbi.nlm.nih.gov/25267658/
van Berkum, J. J. A., Brown, C. M., Hagoort, P., and Zwitserlood, P. (2003). Event-related brain potentials reflect discourse-referential ambiguity in spoken language comprehension. Psychophysiology, 40, 235–248. https://doi.org/10.1111/1469-8986.00025
van Berkum, J. J. A., Hagoort, P., and Brown, C. M. (1999). Semantic integration in sentences and discourse: Evidence from the N400. Journal of Cognitive Neuroscience, 11(6), 657–671. https://doi.org/10.1162/089892999563724
van Berkum, J. J. A., Zwitserlood, P., Hagoort, P., and Brown, C. M. (2003). When and how do listeners relate a sentence to the wider discourse? Evidence from the N400 effect. Cognitive Brain Research, 17(3), 701–718. https://doi.org/10.1016/S0926-6410(03)00196-4
Zacks, J. M., Speer, N. K., Swallow, K. M., Braver, T. S., and Reynolds, J. R. (2007). Event perception: A mind–brain perspective. Psychological Bulletin, 133(2), 273–293. https://doi.org/10.1037/0033-2909.133.2.273
Zwaan, R. A., Langston, M. C., and Graesser, A. C. (1995). The construction of situation models in narrative comprehension: An event-indexing model. Psychological Science, 6(5), 292–297. https://doi.org/10.1111/j.1467-9280.1995.tb00513.x
Zwaan, R. A., and Radvansky, G. A. (1998). Situation models in language comprehension and memory. Psychological Bulletin, 123(2), 162–185. https://doi.org/10.1037/0033-2909.123.2.162