Abstract

Comprehension is a type of cognition: grasping the meaning of what is read or heard by building a coherent mental representation of it. Readers and listeners represent a text at several levels, from its exact wording through its gist to a model of the situation it describes. Building that situation model draws on inferences linking sentences to each other and to prior knowledge, on limited working-memory capacity and on the quality of word knowledge. Skilled reading needs both accurate decoding and general language comprehension, and brain responses show that comprehenders sometimes anticipate upcoming words, within measurable limits. Three interactive demonstrations show how readers track shifts of time, place and character in a story, how bridging inferences fill gaps between sentences, and how decoding and listening comprehension combine.

Keywords: comprehension, situation model, inference, reading comprehension, working memory

Recognising every word in a passage does not guarantee understanding it. Bransford and Johnson (1972) reported a set of investigations into the contextual prerequisites for understanding, examining both comprehension and later recall. Understanding draws on resources well beyond word recognition: a narrative review of reading comprehension research pointed to vocabulary, monitoring of one's own understanding, inference and background knowledge, and concluded that gains in comprehension call for sustained, collaborative effort, since comprehension is hard both to teach and to measure and the scores of adolescents in the United States have stagnated (Elleman & Oslund, 2019).

Comprehension is the receptive side of language use. Its study sits at the centre of psycholinguistics, which asks how a stream of words, printed or spoken, becomes a representation of meaning, and how that representation is shaped by memory, knowledge and the goals of the reader or listener.

Key Takeaways

  • Comprehension builds representations at more than one level: the surface code (the exact words), the textbase (the ideas the text states) and the situation model (the reader's mental picture of what the text is about), whose part in both comprehension and later memory a major review surveyed.
  • The event-indexing model, the proposal that readers track time, place, characters, causes and goals in a story, drew support from studies of simple narratives, and its first author later reported further supporting studies and an extension to film understanding.
  • Inferences, the connections a reader supplies that the text leaves unsaid, are central to understanding; constructionist and minimalist accounts long disputed how many are drawn automatically, and even the constructionist position holds that only some are constructed.
  • Working memory, the limited mental workspace for holding and processing information, matters: in a meta-analysis of 77 studies, span tasks that combine processing with storage predicted comprehension better than storage-only spans, and math-based versions also predicted well.
  • The simple view of reading, the proposal that reading comprehension is the product of decoding (turning print into words) and language comprehension, is supported with qualification: both components predict reading comprehension, but one cross-validation needed linear as well as product terms, and in beginning readers the weight of each component differs between English and more transparent alphabetic orthographies.
  • Prediction, the brain's anticipation of upcoming words, has limits: a nine-laboratory replication confirmed a graded N400 (a brain response linked to meaning) effect on nouns but not the earlier effect on the preceding article.

What Comprehension Is

The act or fact of grasping the meaning, nature, or importance of; understanding. (American Heritage Dictionary, 4th ed) Includes understanding by a patient or research subject of information disclosed orally or in writing.

Scope note for the MeSH descriptor Comprehension (D032882), U.S. National Library of Medicine.

MeSH lists two entry terms under this descriptor. Understanding is the everyday word for the same achievement, grasping what something means. Readability names a property of texts rather than readers: how easily a text can be understood, which this article takes up in its section on applied comprehension.

In cognitive psychology, comprehension is studied as the construction of a representation, and the field distinguishes several levels at which that representation can be held. The distinction among these levels is usually traced to van Dijk and Kintsch's monograph Strategies of discourse comprehension (van Dijk & Kintsch, 1983). In the first of two experiments, memory for the wording, the meaning and the situational content of brief texts decayed at different rates over delays of up to 4 days; in the second, readers with expertise also drew on their situation model, whereas novices judged test sentences by their surface and textbase features (Kintsch et al., 1990). The surface code preserves the exact words and syntax; the textbase holds the ideas the text expresses, usually described as propositions; and the situation model represents the state of affairs the text is about. The gap between form and meaning was an early target of experimental work, as in Sachs's (1967) study of recognition memory for the syntactic and semantic aspects of connected discourse.

Kintsch and van Dijk (1978) gave the text base and the gist a central place in their processing proposal, in which a reader condenses the stated content into a higher-level macrostructure. Later, a review of roughly 15 years of research surveyed the part situation models play in understanding language and in retrieving it from memory (Zwaan & Radvansky, 1998). A situation model is built from the text together with what the reader already knows, and that general knowledge lives in semantic memory. Context can therefore be a precondition for understanding at all, the theme of Bransford and Johnson's (1972) investigations of contextual prerequisites for comprehension and recall.

Figure 1

Three Levels of Representation in Comprehension

Surface code, textbase and situation model as three stacked levels Three stacked boxes. The bottom pale box, surface code, holds the exact wording of a sentence. The middle navy box, textbase, holds the ideas the sentence states. The top gold box, situation model, holds the reader's picture of the situation described, including details the text implies but does not state. Red arrows point upward from each level to the next. situation model the situation described, with what is implied textbase the ideas the sentences state surface code the exact words and syntax
Note. An original schematic, not data. It lays out the levels of representation discussed in this section, from the wording of a text to the reader's model of the situation the text describes.

Models of Text Comprehension

The first detailed processing proposal treated comprehension as a cycle constrained by memory. Kintsch and van Dijk (1978) proposed that a reader handles a text a portion at a time, with only a limited amount carried over in working memory between processing cycles. Of the three sets of operations they described, one works at a micro level, where the reader organises the stated content into a coherent text base, and another at a macro level, where macro-operators reduce that text base to a macrostructure that captures its gist. The authors were explicit about the limits of their proposal. Macroprocesses could be predicted only when the controlling schema, the reader's goal for the text, was made explicit, and the worked application covered a single paragraph (Kintsch & van Dijk, 1978).

The construction-integration model recast the role of knowledge. Kintsch later gave the model its book-length statement in the monograph Comprehension: A paradigm for cognition (Kintsch, 1998). In Kintsch's (1988) computational proposal, word identification, propositions and inferences are first constructed from the bottom up, and an integration phase then operates on what was constructed through spreading activation; one simulation addressed arithmetic word problems. The model stood in contrast to accounts in which a reader's expectations steer processing from above, and Kintsch judged that his simulation offered a plausible treatment of certain established findings in the area it simulated (Kintsch, 1988). Later work questioned whether a purely propositional representation is enough (Zwaan, 2016). Zwaan (2016), discussing the construction-integration and event-indexing models, argued that symbolic representations of the kind such models use and sensorimotor representations constrain each other, and he acknowledged that his proposals about abstract concepts still need to be fleshed out.

McNamara and Magliano (2009) wrote a chapter working toward a comprehensive model of comprehension.

A separate line of work tied comprehension to the moment-by-moment behaviour of the eyes. Just and Carpenter (1980) argued that the time a reader's eyes dwell on a word tracks the processing load it imposes, with gaze durations reflecting work at the level of the word, the clause and larger units of text. Their evidence came from college students reading scientific passages, so the analysis was grounded in one kind of reader and one genre (Just & Carpenter, 1980). Later computational models of eye-movement control linked how long the eyes stayed on a word, and when they moved on, to the progress of lexical access (Reichle et al., 1998).

Situation Models

A situation model represents what a text is about rather than what it says. In their review, Zwaan and Radvansky (1998) treated such models as multidimensional and as playing a part both in comprehension and in later memory retrieval, which links text understanding to episodic memory. They also observed that much of the research had relied on tasks assessing a single dimension, and they called for multidimensionality to receive serious attention (Zwaan & Radvansky, 1998).

The event-indexing model, published three years before that review, is one multidimensional account of this kind for narratives. It holds that readers monitor five dimensions of the story world as they read: the protagonist, the goals behind actions, causal links, time and space (Zwaan et al., 1995). Support came from a verb-clustering task, in which the way readers grouped events reflected these dimensions, but the materials were simple narratives only (Zwaan et al., 1995). Its first author later reported that the model's predictions had been supported in several empirical studies and that it had been extended to film understanding (Zwaan, 2016).

Situation models also affect what stays in the foreground of a reader's attention. Glenberg and colleagues examined how mental models contribute to foregrounding during text comprehension (Glenberg et al., 1987).

A more recent question concerns grounding: how the symbols of language connect to perception and action, and how abstract concepts fit a representation built from sensorimotor experience. Zwaan (2016) proposed two roles for abstract concepts, pointing back to earlier discourse (anaphoric) or forward to what follows (cataphoric), possibly at once, and used the image of a two-way clutch to describe symbolic and sensorimotor representations constraining each other. He allowed that other roles may exist and that the proposals need development (Zwaan, 2016).

The demonstration below steps through an invented story one sentence at a time. Each sentence either continues the current situation or shifts one or more dimensions, and the display marks the shifts and shows a simulated rise in reading time that grows with their number. The values are illustrative, not data.

Demo 1: which situational dimensions does each sentence shift?

Sentence 1 of 7. Maya unlocked the bakery just before dawn.
time continuesplace continuesprotagonist continuescause continuesgoal continues
simulated reading time: 1500 msbaseline 1500 ms plus 180 ms for each dimension that shifts
Nothing changes here, so the reader can keep updating the same situation model.

The story and the timing rule are invented for illustration: one fixed cost per shifted dimension. They show the idea that readers track several situational dimensions at once; they are not measured data.

Inference

Texts leave much unsaid, and readers fill the gaps by inference, a form of thinking in which the reader supplies what the text leaves implicit. A bridging inference supplies a link needed to connect a sentence to what came before; an elaborative inference adds detail that coherence does not strictly require. Which inferences readers routinely draw has been the subject of a long dispute between two positions.

The constructionist theory casts the reader as engaged in a search after meaning. Graesser, Singer and Trabasso proposed that readers build a referential situation model by drawing knowledge-based inferences that secure coherence both locally, between adjacent sentences, and globally, across the text, under the guidance of their goals (Graesser et al., 1994). The theory was developed for narrative text, set against alternative frameworks, and it does not claim that every possible inference is made: only some are constructed (Graesser et al., 1994).

The minimalist hypothesis takes a leaner view. McKoon and Ratcliff (1992) focused on which inferences arise automatically, on the inferences needed for local coherence, and on goal-directed inferences, and they reported new experiments that they took to favour their hypothesis over a constructionist view. Later work in the memory-based tradition reframed the question as a sequence: readers activate background knowledge through resonance, integrate it with the text and validate it, and the strength and timing of their slowdown at an inconsistency depended on how strongly the narrative was related to general world knowledge, which later experiments traced to the rates of those three processes (Cook & O'Brien, 2014).

Inference is also a common point of failure. A review of reading difficulties identified inference making, executive functions and the allocation of attention as sources of trouble for struggling readers (Kendeou et al., 2014). In a longitudinal study of children assessed at 8, 9 and 11 years, working memory, measured with sentence-span and digit tasks, and component skills both predicted reading comprehension, and the contributions of inference making and comprehension monitoring were not wholly explained by working memory (Cain et al., 2004).

The demonstration below presents invented two-sentence pairs whose link is explicit, needs a bridging inference, or needs an elaborative inference. After a probe question, the display explains which inference the probe required. It is illustrative, not a test.

Demo 2: what kind of inference does each question need?

Ana dropped the glass on the tiled floor. It broke into pieces.

What broke?

The sentence pairs are invented examples, not test items. They separate three cases: links the text states, gaps a reader must bridge for the text to cohere, and likely but optional elaborations.

Working Memory and Individual Differences

Readers differ widely in comprehension, and capacity limits are one of the oldest explanations. Working memory became a focus with Daneman and Carpenter's (1980) study of individual differences in working memory and reading.

Just and Carpenter (1992) developed the claim that working-memory capacity constrains comprehension, so that individual differences in capacity show up in how readers handle demanding material such as syntactically ambiguous sentences. The argument bore on whether syntactic processing is modular and gave activation a central role; their abstract reports no statistics, and the evidence concerned college-age adults (Just & Carpenter, 1992). Waters and Caplan argued that the evidence for this capacity theory was weak and that the theory itself was underdeveloped, and they sketched an alternative account, though they did not test it in that paper (Waters & Caplan, 1996). A 2018 meta-analysis pooling 197 studies put the association between working memory and reading at a moderate level (r = .29, 95% CI [.27, .31]) (Peng et al., 2018). Once decoding and vocabulary were taken into account, the effect of working memory on comprehension was no longer present, and from 4th grade onward verbal working memory was more strongly related to reading than other working-memory domains (Peng et al., 2018).

A meta-analysis gave the capacity idea a quantitative test. Pooling 77 studies with 6,179 participants, Daneman and Merikle (1996) found that measures combining processing with storage, such as reading span and listening span, predicted language comprehension better than storage-only measures such as word span and digit span. Processing-plus-storage measures built on mathematical operations also predicted comprehension well, so the predictive power did not depend on the task itself being verbal (Daneman & Merikle, 1996).

Capacity is not the whole story. Perfetti (2007) proposed that the quality of a reader's word knowledge, meaning how precisely spelling, sound and meaning are represented, affects reading skill, comprehension included. Drawing on six lines of research, including event-related potential work, he linked weak word representations to confusions between word forms and meanings; the evidence concerned adult readers only (Perfetti, 2007). Perfetti later restated this proposal inside a wider reading systems framework in which knowledge of words sits at the centre of comprehension (Perfetti & Stafura, 2014). A rival account, long-term working memory, held that skilled comprehenders stretched their effective working memory by maintaining retrieval cues that gave rapid access to knowledge stored in long-term memory, and its authors drew their case from both text comprehension and expert performance (Ericsson & Kintsch, 1995). A later theoretical account of discourse comprehension still cited Ericsson and Kintsch (1995), among other sources, when describing the active portion of a reader's developing situation model as held in working memory (Zwaan, 2016).

Reading Comprehension

The simple view of reading holds that reading comprehension is the product of two components, decoding and language comprehension. Gough and Tunmer (1986) proposed the multiplicative formulation and used it to distinguish three types of reading disability: dyslexia, hyperlexia, and a garden-variety reading disability. Hoover and Gough (1990) later published a paper titled The simple view of reading.

The simple view is supported, with qualification. In a cross-validation study of children whose first language was English, decoding and listening comprehension did predict reading comprehension, but the product model R = D × L was not supported, because the multiplicative term on its own fell short and a model joining the linear and product terms (R = D + L + D × L) was required (Chen & Vellutino, 1997). A later review of some 35 years of research judged the simple view incomplete, pointing out that decoding and comprehension are not cleanly separable and that further contributors, self-regulatory processes among them, shape reading, and it put forward an active view of reading that extends the simple view rather than replacing it (Duke & Cartwright, 2021). Across 757 children in Grades 3 through 5, a latent-variable analysis found that decoding and linguistic comprehension jointly explained nearly the entire variance in reading comprehension, although the two components overlapped heavily and their relative weight changed with development (Lonigan et al., 2018). A meta-analysis of beginning readers found that the relative weight of decoding and linguistic comprehension differed between English and more transparent alphabetic orthographies (Florit & Cain, 2011). In a study of 350 English-speaking second graders, Kim (2017) compared four variations of a structural specification and found that word reading and listening comprehension carried the influence of other skills, among them inference, vocabulary, comprehension monitoring, working memory and grammatical knowledge, onto reading comprehension. The effects of these component skills ranged from small to substantial, so the two-part framework held while its parts proved to rest on several contributing abilities (Kim, 2017).

Some children decode adequately yet understand poorly. A meta-analysis of 86 studies of such poor comprehenders found that they also show deficits in oral language, though milder than their deficit in reading comprehension (Spencer & Wagner, 2018). Against younger children matched on comprehension level, the poor comprehenders did not differ (d = 0.32, 95% CI [−0.49, 1.14]); this comparison bears on whether the profile reflects developmental delay or deviance, but its interval is wide (Spencer & Wagner, 2018).

How children should be taught to read is a related but distinct question within language development. A tutorial review of reading acquisition, from novice to expert and covering both word recognition and text comprehension, concluded that phonics instruction is central to learning to read English (Castles et al., 2018). Teaching comprehension itself is difficult: Elleman and Oslund (2019) described comprehension as hard both to teach and to measure.

Speed reading promises to sidestep these limits. A review of how reading works concluded that a trade-off between speed and accuracy governs reading and that speed-reading programmes and apps are unlikely to deliver what they promise: large gains in speed cost comprehension, and skimming yields only moderate understanding (Rayner et al., 2016).

The calculator below sets decoding and listening comprehension on a 0 to 1 scale and shows their product as predicted reading comprehension, contrasted with their average. It marks the four profiles and offers the readers from the Worked Example as presets. The arithmetic is illustrative.

Demo 3: reading comprehension as decoding times language comprehension

decodinglistening comprehension
Product (simple view)0.36
Average, for contrast0.65
Profiledecodes well but understands language poorly (a poor comprehender profile)
The product (0.36) falls below the average (0.65) whenever the two skills differ, because a weak component pulls the whole score down.

Illustrative arithmetic on a 0 to 1 scale, matching the worked example. It shows the logic of the original multiplicative formulation; it does not test whether a product fits real data better than a sum. Real assessments use standardised scores and regression weights, which differ across orthographies and ages.

Prediction and the Brain

Electrophysiology gave comprehension research a measure that unfolds in milliseconds. The N400 was first described as a late, negative-going brain potential elicited by words that did not make semantic sense within their sentence (Kutas & Hillyard, 1980). Kutas and Federmeier (2000) reviewed the N400 brain response as an index of how semantic memory is used during language comprehension, sensitive to similarity and association between concepts and compatible with pre-activation of upcoming words.

Whether comprehenders predict specific words became a test case. In an event-related potential study of sentence reading, DeLong and colleagues reported that brain responses showed graded pre-activation of upcoming words, scaled by their cloze probability, the proportion of people who complete a sentence frame with that word (DeLong et al., 2005). A pre-registered direct replication across 9 laboratories with 334 participants then tested that 2005 result. The N400 effect on the predicted noun replicated, but the effect on the preceding article did not; Bayesian and single-trial analyses left open that an article effect may be tiny but non-zero, detectable only with very large samples, and the authors concluded that routine pre-activation of a word's sound was not supported (Nieuwland et al., 2018).

Knowledge of the world enters comprehension quickly. Combining electroencephalography with functional magnetic resonance imaging, Hagoort and colleagues reported that world knowledge and word meaning are integrated rapidly and in parallel during sentence interpretation; the abstract reports no statistics (Hagoort et al., 2004). Spoken comprehension draws on the visual scene as well. Using a head-mounted eye tracker, Tanenhaus and colleagues showed that visual context shaped early syntactic parsing of spoken language, linking comprehension to speech perception and spoken word recognition and challenging the assumption of an encapsulated syntactic module (Tanenhaus et al., 1995).

Where language is processed in the brain depends partly on method. Across three fMRI experiments with 37 right-handed adults, Fedorenko and colleagues defined language regions functionally in each individual and found these subject-specific regions more selective (Fedorenko et al., 2010). This evidence concerns sentence-level language regions, not discourse comprehension as such: the sentences-versus-nonwords contrast did not target prosody, pragmatics or discourse, and each experiment used a small sample (Fedorenko et al., 2010).

Two broader proposals situate comprehension in cognition generally. Hasson and colleagues argued that memory is intrinsic to cortical processing, with areas integrating information over temporal receptive windows ranging from tens of milliseconds to minutes, a claim supported partly by responses to scrambled narratives; they left open the role of the medial temporal lobe at long timescales, the contribution of attention and the place of top-down feedback (Hasson et al., 2015). Pickering and Garrod (2013), in a target article published with open peer commentary, argued that production and comprehension are interwoven rather than distinct, giving covert imitation and forward models a role in prediction; the article reviewed existing evidence rather than presenting new data. A pre-registered direct replication with 334 participants across 9 laboratories reproduced the N400 effect on nouns but not the effect on the preceding articles, offering no support for routine phonological pre-activation and posing a serious challenge to the view that prediction runs through implicit production, although an article effect too tiny to detect without very large samples could not be excluded (Nieuwland et al., 2018).

Comprehension is also sometimes shallow. Under the good-enough approach, processing is sometimes only partial and the resulting interpretations are often incomplete, as studies of garden-path and passive sentences suggested; Ferreira and colleagues acknowledged that the conditions under which good-enough processing occurs remained unspecified (Ferreira et al., 2002). Ferreira and Lowder later returned to the approach in a book chapter that linked good-enough language processing to prediction and to information structure (Ferreira & Lowder, 2016).

Readability, Media and Applied Comprehension

Readability, the ease with which a text can be understood, is traditionally estimated with formulas such as the Flesch-Kincaid grade level. Crossley and colleagues reported that indices drawn from natural language processing tools, combined through machine learning and checked against crowd-sourced ratings, predicted adult readers' judgments of text comprehension, processing and familiarity better than traditional readability formulas; the abstract reports no effect sizes (Crossley et al., 2017).

A text that is easier is not always better. In two experiments with junior high school students, McNamara and colleagues varied the coherence of a text and found that whether added coherence helped depended on the reader's background knowledge and on the level of understanding being assessed; free recall mainly reflected superficial understanding, whereas inference and problem-solving questions reflected understanding of the situation (McNamara et al., 1996).

The medium matters too, though modestly. A meta-analysis of studies from 2000 to 2017, covering 171,055 participants, found a small advantage for reading on paper over screens (g = −0.21), with no paper advantage for texts that were narrative only; time frame, genre and publication year moderated the effect (Delgado et al., 2018). A second meta-analysis, restricted to randomised experiments from 2008 to 2018 (33 studies, 2,799 participants), also found reading performance slightly worse on screens, but the deficit may hold only for expository texts, narrative texts showed no difference, and the difference in reading time was not reliable (Clinton, 2019).

The MeSH scope note extends comprehension to a patient or research subject understanding disclosed information, and informed consent is the clearest case. A systematic review of 30 studies comprising 42 trials of interventions to improve participants' understanding found extended discussion the most promising approach, although that conclusion rested partly on statistical trends; five of the six enhanced-form trials that showed improvement were of limited quality, and gains from test-and-feedback methods may reflect rote memorisation (Flory & Emanuel, 2004). A later meta-analysis of 103 studies covering 135 cohorts found that participants' understanding varied considerably, with pooled rates across the aspects assessed ranging from 52.1% to 75.8%, and that it had not improved over 30 years; results were heterogeneous and most data sets came from middle- and high-income countries (Tam et al., 2015).

Worked Example

The simple view of reading expresses reading comprehension as decoding multiplied by language comprehension (Gough & Tunmer, 1986). The figures that follow are this article's own illustrative arithmetic; they come from no study.

Suppose four readers are scored on a 0 to 1 scale for decoding and for listening comprehension. Reader A scores 0.9 on both, so the predicted reading comprehension is 0.9 × 0.9 = 0.81. Reader B decodes at 0.9 but understands spoken language at only 0.4, giving 0.9 × 0.4 = 0.36. Reader C shows the reverse profile, decoding at 0.4 with listening comprehension at 0.9, and also reaches 0.4 × 0.9 = 0.36. Reader D scores 0.4 on both, for 0.4 × 0.4 = 0.16. Table 1 sets out the four profiles.

Readers B and C arrive at the same score by opposite routes. B resembles the poor comprehender, whose decoding is adequate but whose language comprehension lags; C resembles the reader with dyslexia, whose understanding is intact but whose decoding is weak. The score alone cannot tell them apart, which is why assessment measures both components. The product, rather than the sum, carries the claim that both components are necessary: a sum would credit a reader with zero decoding and perfect listening comprehension with half marks (0 + 1.0 = 1.0 out of a possible 2), whereas the product gives that reader 0 × 1.0 = 0, because a reader who cannot decode cannot read the text at all, however well he or she understands speech. The example follows the logic of the original multiplicative formulation (Gough & Tunmer, 1986); it illustrates that logic and is not a test of whether a product fits reading data better than a sum.

Table 1

Predicted Reading Comprehension Under the Simple View for Four Illustrative Readers

ReaderDecodingListening comprehensionProduct (reading comprehension)
A0.90.90.9 × 0.9 = 0.81
B0.90.40.9 × 0.4 = 0.36
C0.40.90.4 × 0.9 = 0.36
D0.40.40.4 × 0.4 = 0.16

Note. This article's own illustrative arithmetic on a 0 to 1 scale, not data. Readers B and C reach the same product by opposite routes, and a zero on either component would give a zero product.

Discussion

Several conclusions are well established. Comprehension operates at more than one level, and the situation model, the representation of what a text is about, plays a part in both understanding and memory (Zwaan & Radvansky, 1998). Processing-plus-storage span measures predict comprehension, although in one meta-analysis the effect of working memory on comprehension was partialed out once decoding and vocabulary were controlled, and the quality of word knowledge also matters. Reading comprehension requires decoding and language comprehension together, a claim that has held up even as the weights of the two components were shown to vary.

Other questions remain open. The dispute over inference set two positions against each other: the constructionist theory holds that readers build knowledge-based inferences in a search after meaning, though only some (Graesser et al., 1994), while the minimalist hypothesis gives automatic inference a narrower scope (McKoon & Ratcliff, 1992). How far prediction is routine is also unsettled, since a large replication confirmed the graded noun effect but not the effect on the preceding article (Nieuwland et al., 2018). Whether proposals about the roles of abstract concepts in discourse apply to scientific papers as well as to narratives is, by their author's own account, uncertain (Zwaan, 2016). And the practical question of how to teach and measure comprehension is, by the field's own account, difficult (Elleman & Oslund, 2019).

These open questions share a design problem. Each calls for studies that vary text genre, reader knowledge and task demands together.

Current Directions

Recent work has extended comprehension research in several directions. On the theoretical side, Zwaan (2016) has proposed ways for situation models to accommodate abstract concepts and grounding, while noting that the proposals need to be fleshed out further. In reading research, Kim (2017) has unpacked the components behind the two terms of the simple view in second graders, and a meta-analysis of poor comprehenders has shown that they also have milder oral-language deficits (Spencer & Wagner, 2018). A tutorial review has argued that phonics instruction is central to learning to read English (Castles et al., 2018), and a policy-oriented review has stressed that comprehension gains require sustained, collaborative effort (Elleman & Oslund, 2019).

On prediction, a large multi-laboratory replication has placed limits on how far comprehenders pre-activate specific word forms (Nieuwland et al., 2018). Applied research has quantified a small paper advantage over screens (Delgado et al., 2018) that may be confined to expository texts (Clinton, 2019), has shown that models built from natural language processing tools predicted adult readers' judgments of texts better than traditional readability formulas (Crossley et al., 2017), and has concluded that speed reading is unlikely to deliver fast reading without a cost to comprehension (Rayner et al., 2016).

Common Misconceptions

Reading comprehension is just a matter of decoding the words.
Decoding is necessary but not sufficient. The simple view of reading treats reading as the product of decoding and comprehension, so a reader who decodes well but understands poorly still reads poorly, and the same formulation was used to separate three types of reading disability (Gough & Tunmer, 1986).
Readers always predict the next word in full detail.
A pre-registered replication across 9 laboratories confirmed graded N400 effects on predictable nouns but not on the preceding article, and found no support for routine pre-activation of a word's sound (Nieuwland et al., 2018). The belief stretches the evidence: brain responses to nouns track how predictable they are, but the replication found no clear evidence that readers routinely pre-activate a word's form in advance.
Speed reading lets people read much faster with no loss of understanding.
Reading is governed by a trade-off between speed and accuracy; large speed gains cost comprehension, and skimming yields only moderate understanding (Rayner et al., 2016).
Reading on a screen is the same as reading on paper.
A meta-analysis covering 171,055 participants found a small advantage for paper, though not for texts that were narrative only (Delgado et al., 2018).

Glossary

Bridging inference.
An inference that supplies a connection the reader needs to link a sentence to earlier text, without which the passage would not hang together.
Construction-integration model.
A computational proposal in which word meanings, propositions and inferences are first generated from the bottom up, regardless of the surrounding discourse, and then integrated by spreading activation; later work questioned whether propositions alone suffice.
Elaborative inference.
An inference that adds detail beyond what is needed to make a text coherent.
Event-indexing model.
The proposal that readers of a story track several situational dimensions, including characters, goals, causes, time and place; first supported with simple narratives and later extended to film.
Good-enough processing.
Comprehension that settles for a partial or underspecified reading of a sentence rather than a fully worked-out one; the approach's original statement left the conditions that trigger it for future work.
Listening comprehension.
Understanding of spoken language, used in reading research as a measure of language comprehension separate from decoding.
Macrostructure.
The condensed, higher-level representation of a text's gist, derived from its detailed content.
Mental model.
An internal representation of a described state of affairs rather than of the text itself; in comprehension research the term is used interchangeably with situation model.
N400.
A brain potential, named for its negative polarity and its timing, used as an index of how meaning is accessed during comprehension.
Proposition.
A unit of meaning, typically a predicate with its arguments, used to describe the ideas a text expresses independent of its wording.
Readability.
How easily a text can be understood; a property of texts rather than readers, traditionally estimated with formulas.
Reading span.
A working memory measure in which a person reads sentences while holding material from them for later recall, combining processing with storage.
Simple view of reading.
The proposal that reading comprehension is the product of decoding and language comprehension; supported in that both components predict reading comprehension, though the product term alone has not always sufficed and component weights in beginning readers differ between English and more transparent alphabetic orthographies.
Situation model.
The reader's representation of the state of affairs a text describes, built from the text together with prior knowledge.
Surface code.
The representation of the exact words and syntax of a text.
Textbase.
The representation of the ideas a text explicitly states, organised into connected propositions.
Understanding.
The everyday term for comprehension, and a MeSH entry term for it: grasping what something means.

Key Researchers

Kate Cain

(Lancaster University). Developmental psychologist who studies children's reading comprehension, including the working-memory and component skills that predict it and how the simple view of reading holds across orthographies. ORCID · Faculty page

Arthur C. Graesser

(University of Memphis). Cognitive psychologist who set out, with Singer and Trabasso, a constructionist account of the inferences readers draw from narratives. Wikipedia · Wikidata · Google Scholar · Faculty page

Walter Kintsch

(University of Colorado Boulder; 1932–2023). Cognitive psychologist who, with van Dijk, modelled how readers build a text representation and later proposed the construction-integration model. Wikipedia · Wikidata · Faculty page

Marta Kutas

(University of California San Diego). Cognitive neuroscientist who studies the N400 brain response and what it shows about meaning and prediction during comprehension. Wikipedia · Wikidata · Faculty page

Danielle S. McNamara

(Arizona State University). Cognitive scientist who studies how text coherence, reader knowledge and text features shape comprehension and readability. ORCID · Wikipedia · Wikidata · Google Scholar · Faculty page

Charles A. Perfetti

(University of Pittsburgh). Cognitive psychologist who studies how the quality of a reader's word knowledge supports reading comprehension. Wikipedia · Wikidata

Rolf A. Zwaan

(Erasmus University Rotterdam). Cognitive psychologist who studies situation models and how readers track events in narratives. Google Scholar · Faculty page

Frequently Asked Questions

What does comprehension mean in cognitive psychology?

Cognitive psychologists treat comprehension as the building of a mental representation of what a text or utterance means. A major review surveyed the part the situation model, the representation of what the text is about, plays in both comprehension and later memory (Zwaan & Radvansky, 1998).

What do readers keep track of in a story?

According to one influential proposal, readers of a narrative monitor the protagonist, goals, causes, time and place as they read. Support for it first came from a verb-clustering study with simple narratives (Zwaan et al., 1995).

Do readers draw every possible inference while reading?

No. Even the constructionist position, which holds that readers search for meaning and draw inferences based on knowledge, maintains that only some inferences are constructed during reading (Graesser et al., 1994).

A meta-analysis of 77 studies with 6,179 participants found that tasks requiring people to process and store information at the same time predicted comprehension better than tasks requiring storage alone (Daneman & Merikle, 1996).

What is the simple view of reading?

The simple view holds that reading comprehension is the product of decoding and language comprehension, so that a reader weak in either one will read poorly. Its authors used it to describe three types of reading disability (Gough & Tunmer, 1986).

Can a child read words accurately but still fail to understand them?

Yes. A meta-analysis of 86 studies of children with poor comprehension despite adequate decoding found that these children also have difficulties with oral language, in milder form (Spencer & Wagner, 2018).

Does the brain predict upcoming words during reading?

To a degree. A replication across 9 laboratories with 334 participants confirmed that brain responses to nouns are graded by how predictable they are, but did not replicate an earlier effect on the preceding article, which places a limit on how detailed prediction is (Nieuwland et al., 2018).

Is reading on a screen worse for comprehension than reading on paper?

Slightly, and perhaps only for some texts. A meta-analysis of randomised experiments found reading performance somewhat worse on screens, but the deficit may hold only for expository texts, with no difference for narratives (Clinton, 2019).

References

Bransford, J. D., & Johnson, M. K. (1972). Contextual prerequisites for understanding: Some investigations of comprehension and recall. Journal of Verbal Learning and Verbal Behavior, 11(6), 717–726. https://doi.org/10.1016/S0022-5371(72)80006-9

Cain, K., Oakhill, J. V., & Bryant, P. (2004). Children's reading comprehension ability: Concurrent prediction by working memory, verbal ability, and component skills. Journal of Educational Psychology, 96(1), 31–42. https://doi.org/10.1037/0022-0663.96.1.31

Castles, A., Rastle, K., & Nation, K. (2018). Ending the reading wars: Reading acquisition from novice to expert. Psychological Science in the Public Interest, 19(1), 5–51. https://doi.org/10.1177/1529100618772271

Chen, R. S., & Vellutino, F. R. (1997). Prediction of reading ability: A cross-validation study of the simple view of reading. Journal of Literacy Research, 29(1), 1–24. https://doi.org/10.1080/10862969709547947

Clinton, V. (2019). Reading from paper compared to screens: A systematic review and meta-analysis. Journal of Research in Reading, 42(2), 288–325. https://doi.org/10.1111/1467-9817.12269

Cook, A. E., & O'Brien, E. J. (2014). Knowledge activation, integration, and validation during narrative text comprehension. Discourse Processes, 51(1–2), 26–49. https://doi.org/10.1080/0163853X.2013.855107

Crossley, S. A., Skalicky, S., Dascalu, M., McNamara, D. S., & Kyle, K. (2017). Predicting text comprehension, processing, and familiarity in adult readers: New approaches to readability formulas. Discourse Processes, 54(5–6), 340–359. https://doi.org/10.1080/0163853X.2017.1296264

Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19(4), 450–466. https://doi.org/10.1016/S0022-5371(80)90312-6

Daneman, M., & Merikle, P. M. (1996). Working memory and language comprehension: A meta-analysis. Psychonomic Bulletin & Review, 3(4), 422–433. https://doi.org/10.3758/BF03214546

Delgado, P., Vargas, C., Ackerman, R., & Salmerón, L. (2018). Don't throw away your printed books: A meta-analysis on the effects of reading media on reading comprehension. Educational Research Review, 25, 23–38. https://doi.org/10.1016/j.edurev.2018.09.003

DeLong, K. A., Urbach, T. P., & Kutas, M. (2005). Probabilistic word pre-activation during language comprehension inferred from electrical brain activity. Nature Neuroscience, 8(8), 1117–1121. https://doi.org/10.1038/nn1504

Duke, N. K., & Cartwright, K. B. (2021). The science of reading progresses: Communicating advances beyond the simple view of reading. Reading Research Quarterly, 56(S1), S25–S44. https://doi.org/10.1002/rrq.411

Elleman, A. M., & Oslund, E. L. (2019). Reading comprehension research: Implications for practice and policy. Policy Insights from the Behavioral and Brain Sciences, 6(1), 3–11. https://doi.org/10.1177/2372732218816339

Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102(2), 211–245. https://doi.org/10.1037/0033-295X.102.2.211

Fedorenko, E., Hsieh, P.-J., Nieto-Castañón, A., Whitfield-Gabrieli, S., & Kanwisher, N. (2010). New method for fMRI investigations of language: Defining ROIs functionally in individual subjects. Journal of Neurophysiology, 104(2), 1177–1194. https://doi.org/10.1152/jn.00032.2010

Ferreira, F., & Lowder, M. W. (2016). Prediction, information structure, and good-enough language processing. Psychology of Learning and Motivation, 65, 217–247. https://doi.org/10.1016/bs.plm.2016.04.002

Ferreira, F., Bailey, K. G. D., & Ferraro, V. (2002). Good-enough representations in language comprehension. Current Directions in Psychological Science, 11(1), 11–15. https://doi.org/10.1111/1467-8721.00158

Florit, E., & Cain, K. (2011). The simple view of reading: Is it valid for different types of alphabetic orthographies? Educational Psychology Review, 23(4), 553–576. https://doi.org/10.1007/s10648-011-9175-6

Flory, J., & Emanuel, E. (2004). Interventions to improve research participants' understanding in informed consent for research: A systematic review. JAMA, 292(13), 1593–1601. https://doi.org/10.1001/jama.292.13.1593

Glenberg, A. M., Meyer, M., & Lindem, K. (1987). Mental models contribute to foregrounding during text comprehension. Journal of Memory and Language, 26(1), 69–83. https://doi.org/10.1016/0749-596X(87)90063-5

Gough, P. B., & Tunmer, W. E. (1986). Decoding, reading, and reading disability. Remedial and Special Education, 7(1), 6–10. https://doi.org/10.1177/074193258600700104

Graesser, A. C., Singer, M., & 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

Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M. (2004). Integration of word meaning and world knowledge in language comprehension. Science, 304(5669), 438–441. https://doi.org/10.1126/science.1095455

Hasson, U., Chen, J., & Honey, C. J. (2015). Hierarchical process memory: Memory as an integral component of information processing. Trends in Cognitive Sciences, 19(6), 304–313. https://doi.org/10.1016/j.tics.2015.04.006

Hoover, W. A., & Gough, P. B. (1990). The simple view of reading. Reading and Writing, 2(2), 127–160. https://doi.org/10.1007/BF00401799

Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99(1), 122–149. https://doi.org/10.1037/0033-295X.99.1.122

Just, M. A., & Carpenter, P. A. (1980). A theory of reading: From eye fixations to comprehension. Psychological Review, 87(4), 329–354. https://doi.org/10.1037/0033-295X.87.4.329

Kendeou, P., van den Broek, P., Helder, A., & Karlsson, J. (2014). A cognitive view of reading comprehension: Implications for reading difficulties. Learning Disabilities Research & Practice, 29(1), 10–16. https://doi.org/10.1111/ldrp.12025

Kim, Y.-S. G. (2017). Why the simple view of reading is not simplistic: Unpacking component skills of reading using a direct and indirect effect model of reading (DIER). Scientific Studies of Reading, 21(4), 310–333. https://doi.org/10.1080/10888438.2017.1291643

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

Kintsch, W. (1998). Comprehension: A paradigm for cognition. Cambridge University Press.

Kintsch, W., & van Dijk, T. A. (1978). Toward a model of text comprehension and production. Psychological Review, 85(5), 363–394. https://doi.org/10.1037/0033-295X.85.5.363

Kintsch, W., Welsch, D., Schmalhofer, F., & Zimny, S. (1990). Sentence memory: A theoretical analysis. Journal of Memory and Language, 29(2), 133–159. https://doi.org/10.1016/0749-596X(90)90069-C

Kutas, M., & Federmeier, K. D. (2000). Electrophysiology reveals semantic memory use in language comprehension. Trends in Cognitive Sciences, 4(12), 463–470. https://doi.org/10.1016/S1364-6613(00)01560-6

Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity. Science, 207(4427), 203–205. https://doi.org/10.1126/science.7350657

Lonigan, C. J., Burgess, S. R., & Schatschneider, C. (2018). Examining the simple view of reading with elementary school children: Still simple after all these years. Remedial and Special Education, 39(5), 260–273. https://doi.org/10.1177/0741932518764833

McKoon, G., & Ratcliff, R. (1992). Inference during reading. Psychological Review, 99(3), 440–466. https://doi.org/10.1037/0033-295X.99.3.440

McNamara, D. S., Kintsch, E., Songer, N. B., & Kintsch, W. (1996). Are good texts always better? Interactions of text coherence, background knowledge, and levels of understanding in learning from text. Cognition and Instruction, 14(1), 1–43. https://doi.org/10.1207/s1532690xci1401_1

McNamara, D. S., & Magliano, J. (2009). Toward a comprehensive model of comprehension. Psychology of Learning and Motivation, 51, 297–384. https://doi.org/10.1016/S0079-7421(09)51009-2

Nieuwland, M. S., Politzer-Ahles, S., Heyselaar, E., Segaert, K., Darley, E., Kazanina, N., Von Grebmer Zu Wolfsthurn, S., Bartolozzi, F., Kogan, V., Ito, A., Mézière, D., Barr, D. J., Rousselet, G. A., Ferguson, H. J., Busch-Moreno, S., Fu, X., Tuomainen, J., Kulakova, E., Husband, E. M., … Huettig, F. (2018). Large-scale replication study reveals a limit on probabilistic prediction in language comprehension. eLife, 7, Article e33468. https://doi.org/10.7554/eLife.33468

Peng, P., Barnes, M., Wang, C., Wang, W., Li, S., Swanson, H. L., Dardick, W., & Tao, S. (2018). A meta-analysis on the relation between reading and working memory. Psychological Bulletin, 144(1), 48–76. https://doi.org/10.1037/bul0000124

Perfetti, C. (2007). Reading ability: Lexical quality to comprehension. Scientific Studies of Reading, 11(4), 357–383. https://doi.org/10.1080/10888430701530730

Perfetti, C., & Stafura, J. (2014). Word knowledge in a theory of reading comprehension. Scientific Studies of Reading, 18(1), 22–37. https://doi.org/10.1080/10888438.2013.827687

Pickering, M. J., & Garrod, S. (2013). An integrated theory of language production and comprehension. Behavioral and Brain Sciences, 36(4), 329–347. https://doi.org/10.1017/S0140525X12001495

Rayner, K., Schotter, E. R., Masson, M. E. J., Potter, M. C., & Treiman, R. (2016). So much to read, so little time: How do we read, and can speed reading help? Psychological Science in the Public Interest, 17(1), 4–34. https://doi.org/10.1177/1529100615623267

Reichle, E. D., Pollatsek, A., Fisher, D. L., & Rayner, K. (1998). Toward a model of eye movement control in reading. Psychological Review, 105(1), 125–157. https://doi.org/10.1037/0033-295X.105.1.125

Sachs, J. (1967). Recognition memory for syntactic and semantic aspects of connected discourse. Perception & Psychophysics, 2(9), 437–442. https://doi.org/10.3758/BF03208784

Spencer, M., & Wagner, R. K. (2018). The comprehension problems of children with poor reading comprehension despite adequate decoding: A meta-analysis. Review of Educational Research, 88(3), 366–400. https://doi.org/10.3102/0034654317749187

Tam, N. T., Huy, N. T., Thoa, L. T. B., Long, N. P., Trang, N. T. H., Hirayama, K., & Karbwang, J. (2015). Participants' understanding of informed consent in clinical trials over three decades: Systematic review and meta-analysis. Bulletin of the World Health Organization, 93(3), 186–198H. https://doi.org/10.2471/BLT.14.141390

Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M., & Sedivy, J. C. (1995). Integration of visual and linguistic information in spoken language comprehension. Science, 268(5217), 1632–1634. https://doi.org/10.1126/science.7777863

van Dijk, T. A., & Kintsch, W. (1983). Strategies of discourse comprehension. Academic Press.

Waters, G. S., & Caplan, D. (1996). The capacity theory of sentence comprehension: Critique of Just and Carpenter (1992). Psychological Review, 103(4), 761–772. https://doi.org/10.1037/0033-295X.103.4.761

Zwaan, R. A., Langston, M. C., & 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., & 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

Zwaan, R. A. (2016). Situation models, mental simulations, and abstract concepts in discourse comprehension. Psychonomic Bulletin & Review, 23(4), 1028–1034. https://doi.org/10.3758/s13423-015-0864-x