{
  "id": "reading-comprehension",
  "title": "Reading Comprehension",
  "lang": "en",
  "source_lang": "en",
  "summary": "The cognitive processes and learned abilities involved in constructing, interpreting, and evaluating meaning from written texts.",
  "category": "society",
  "subcategory": "education",
  "type": "concept",
  "aliases": [
    "Written-text comprehension",
    "Text comprehension"
  ],
  "generated_at": "2026-10-22",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/reading-comprehension/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/reading-comprehension/",
      "json": "https://aiwiki.page/zh/reading-comprehension.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Reading comprehension is the process of constructing meaning from written text by connecting its words and ideas with a reader’s knowledge, purposes, and understanding of language. It involves more than recognizing words: readers must establish relationships between ideas, make inferences, and develop a coherent understanding of what a text communicates. Comprehension emerges through interactions among the reader, the text, and the reading activity within a broader social and cultural context. (rand.org)\n\nReading comprehension is central to literacy, learning across school subjects, and participation in everyday and public life. Its study brings together cognitive psychology, linguistics, and education, encompassing the development of reading abilities, differences among readers, instructional approaches, and assessment. Digital environments extend these concerns to navigating interconnected documents, integrating information from multiple sources, and evaluating credibility; computational research uses the related term “machine reading comprehension” for systems that answer questions about texts. (nichd.nih.gov)"
    },
    {
      "heading": "Definition, Scope, and Reading Purposes",
      "text": "Reading comprehension denotes both the ongoing process of making meaning from written language and the understanding that this process produces. In its 2002 report Reading for Understanding, the RAND Reading Study Group, chaired by Catherine Snow, characterized comprehension as the joint extraction and construction of meaning: readers interpret textual information using their knowledge and experience. The resulting understanding can support an explanation, a decision, or an action; it is not simply a record of the words encountered. (rand.org)\n\nComprehension is therefore distinct from word recognition, accurate oral reading, and verbatim recall. Recognizing or pronouncing every word does not establish that a reader understands the relationships among them. Likewise, remembering exact wording is different from preserving its meaning in a paraphrase or explanation. For example, someone might repeat “Applications received after Friday will not be considered” without correctly identifying whether an application submitted on Saturday is eligible. The distinction concerns access to meaning, rather than the accuracy with which written forms are reproduced. (oecd.org)\n\nReading comprehension shares linguistic resources with listening comprehension, but written language introduces an additional requirement: identifying words from their written forms. Both activities involve interpreting vocabulary, grammatical relationships, and connected discourse. In their 1990 study, Wesley A. Hoover and Philip B. Gough examined English–Spanish bilingual children from first through fourth grade and found that decoding and listening comprehension jointly explained variation in reading comprehension, with their product improving the explanation. This relationship distinguishes understanding language from gaining access to it through print: understanding a passage when it is spoken does not, by itself, establish that the same passage can be read independently. (doi.org)\n\nReading literacy is a broader construct that places comprehension within the purposeful use of texts. The OECD’s PISA 2018 framework encompasses understanding, application, evaluation, reflection, and engagement, relating these activities to personal goals, the development of knowledge, and participation in society. In this sense, literacy includes not only grasping what a text says but also using it in a particular setting. Its scope extends beyond continuous printed prose to digital texts and noncontinuous materials such as tables and forms. Comprehension is central to this broader construct, but is not synonymous with all its components. (oecd.org)\n\nReading purposes determine which information matters and how thoroughly it must be understood. Following assembly instructions requires identifying components, conditions, and the order of operations; locating a departure time may require only the relevant timetable entry. Studying an explanation calls for retaining relationships among ideas, whereas reading literature for enjoyment may emphasize characters, events, and emotional engagement. These are illustrative differences in task demands, not fixed boundaries between kinds of reading. RAND distinguishes purposes, processing, and outcomes, and notes that purposes can change during reading: an unexpected statement may prompt closer examination, while recreational reading may also produce new knowledge. (rand.org)"
    },
    {
      "heading": "Research History and Explanatory Models",
      "text": "Philip B. Gough and William E. Tunmer introduced the Simple View of Reading in 1986 to explain the joint contributions of decoding and comprehension. Wesley A. Hoover and Gough developed and tested this account in 1990, expressing reading comprehension as \\(R = D \\times C\\), where \\(D\\) represents decoding and \\(C\\) linguistic comprehension. Decoding here encompasses efficient word recognition, rather than only laborious sounding out; linguistic comprehension concerns interpreting words, sentences, and discourse, commonly assessed through listening. The multiplicative formulation makes both components necessary: recognizing written words cannot produce understanding without access to their meanings, while understanding spoken language cannot compensate completely for an inability to identify printed words. (cir.nii.ac.jp)\n\nWith component abilities represented on a scale from zero to one, the equation predicts zero reading comprehension when either component is zero. As a hypothetical illustration, \\(D = 0.9\\) and \\(C = 0.2\\) yield \\(R = 0.18\\), showing why strength in one component does not simply cancel weakness in the other. These numbers illustrate the model, not a conversion rule for ordinary test scores. Hoover and Gough’s 1990 longitudinal study of English–Spanish bilingual children in grades 1–4 found that including the product of decoding and listening comprehension significantly improved the explanation of reading-comprehension variation beyond their linear combination. The Simple View is therefore a component-level account with testable predictions, not a specification of every mental operation occurring during reading. (doi.org)\n\nWalter Kintsch’s 1988 construction–integration model addressed the processing question more directly: how does a reader turn successive words and statements into a coherent representation? During construction, word meanings become activated, propositions—representations of claims or relations—are formed, and associated knowledge, inferences, and elaborations enter a network. This initial generation is relatively permissive rather than restricted to meanings already known to fit the discourse. During integration, activation spreads through the network, strengthening mutually supported elements and reducing the influence of poorly connected or incompatible alternatives. Context thus constrains the resulting interpretation without requiring a fully correct interpretation at the outset. (condor.depaul.edu)\n\nKintsch’s proposal contrasted with strongly expectation-driven accounts in which prior knowledge guides interpretation from the beginning. Its distinctive mechanism was the combination of comparatively indiscriminate construction with coherence-sensitive integration. The 1988 paper examined evidence about the time course of word identification in discourse and presented a computational simulation of arithmetic word-problem understanding. Such problems provided a concrete case in which identifying the words and numbers was insufficient: the representation had to support the appropriate relationships among quantities. Unlike the Simple View’s broad decomposition of ability, this model specified processes through which textual information and knowledge could become organized during comprehension. (condor.depaul.edu)\n\nThe RAND Reading Study Group, chaired by Catherine Snow, shifted the explanatory scope in its 2002 report Reading for Understanding. Its framework placed three interacting elements—reader, text, and activity—within a sociocultural context. The reader contributes capabilities, knowledge, motivation, and experience; the text supplies particular linguistic and conceptual demands; and the activity includes purposes, processes, and consequences of reading. Home, classroom, and community practices shape all three, including which texts are available and which interpretations or uses are valued. The same passage read to locate a fact or to prepare a comparative analysis therefore constitutes different reading activities. RAND presented this framework as a heuristic for organizing research and development, rather than a component equation or a computational processing mechanism. It made interactions among readers, materials, purposes, and learning environments explicit objects of investigation. (rand.org)"
    },
    {
      "heading": "Word Recognition and Linguistic Foundations",
      "text": "Decoding connects written forms with spoken language. In an alphabetic writing system, readers use relationships between graphemes and phonemes to identify unfamiliar words: the letters sh in ship, for example, represent one speech sound rather than two. Blending the sounds provides access to a pronunciation, which can activate a meaning already known from speech. Readers also use familiar spelling patterns and meaningful word parts, such as roots and affixes. Successful pronunciation does not, however, establish understanding: a reader may decode isotope accurately without knowing what it denotes. The 2016 US Institute of Education Sciences foundational-skills guide, developed by a panel chaired by Barbara Foorman, distinguishes these word-reading foundations from the broader work of understanding text. (nichd.nih.gov)\n\nWith increasing proficiency, familiar words are recognized rapidly and with less conscious effort, rather than being laboriously sounded out on every encounter. Connected-text fluency extends beyond recognizing isolated words: it includes accuracy, automaticity, an appropriate pace, and expression and phrasing in oral reading. The psychological explanation emphasized in the OECD’s PISA 2018 reading framework is that efficient word processing leaves more attention and memory resources available for comprehension. Fluency is therefore not simply speed, nor does smooth oral delivery guarantee understanding of a passage’s vocabulary or grammatical relationships. (ies.ed.gov)\n\nVocabulary breadth concerns how many words a reader knows; depth concerns how fully those words are understood, including their senses, associations, and uses. Knowing bank only as a financial institution is insufficient for interpreting “The stream overflowed its bank.” Similarly, everyday knowledge of volume as loudness does not supply its mathematical meaning. Such examples distinguish recognition of a word’s form from access to the meaning appropriate to a passage. Knowledge of morphology can contribute further: recognizing re- and -able helps a reader analyze reusable, although word-part analysis does not replace knowledge of usage and context. The US National Reading Panel’s 2000 review examined 50 vocabulary studies representing 21 instructional methods; its findings emphasized repeated encounters and varied contexts rather than treating one exposure as complete word knowledge. (ies.ed.gov)\n\nSentence comprehension also requires parsing: identifying how words form phrases and how those phrases are grammatically related. Syntax helps determine who performs an action and who receives it. For example, “The dog chased the boy” and “The dog was chased by the boy” contain the same principal nouns and verb but assign different roles. Comprehending the second sentence requires interpreting the passive construction rather than assuming that the first noun names the actor. Recognizing individual words is thus distinct from assembling their sentence-level meaning. (oecd.org)\n\nAcross clauses and sentences, connectives and referring expressions guide integration. In the illustrative contrast “The road flooded because it rained” versus “The road flooded although it had not rained,” the connective changes the stated relationship between events. In “Maya placed the seedlings beside the window. They needed more light,” interpreting they requires connecting the pronoun with the seedlings. Readers must retain the relevant referent while processing the next sentence; recognizing every word separately does not itself establish this connection. These grammatical links support local understanding before the reader constructs a representation of the passage as a whole. (oecd.org)"
    },
    {
      "heading": "Inference, Coherence, and Mental Representation",
      "text": "Accounts in cognitive psychology distinguish three levels of representation. The surface representation preserves a text’s wording and grammatical form. The textbase represents its propositional content: who did what, to whom, and under which conditions. Thus, “Maya opened the gate” and “The gate was opened by Maya” differ in surface form but express the same central proposition. A situation model represents the events or circumstances described, integrating textual information with relevant prior knowledge. It can include relations among people, objects, locations, times, and causes that are not stated in a single sentence. These are distinguishable representations, not three successive stages in which each replaces the preceding one. (link.springer.com)\n\nLocal coherence concerns connections among nearby statements, such as identifying a pronoun’s referent or understanding how one event explains the next. Global coherence concerns how information across a passage contributes to an overarching topic, explanation, or sequence of goals and outcomes. A reader may understand every sentence separately yet fail to identify the passage’s main claim. Conversely, recognizing a general topic does not establish how its individual claims fit together. The OECD’s PISA 2018 reading framework describes integration as extending from individual sentences to the entire passage, through spatial, temporal, causal, and claim–argument relationships; identifying an unstated main idea also requires inference. (oecd.org)\n\nInference supplies connections that wording alone leaves unspecified. Bridging inferences establish links needed for coherence, whereas elaborative inferences add information beyond those links. Consider this illustrative passage: “Lena set a glass on the edge of the table. A moment later, she swept up the broken pieces.” The text explicitly states where Lena put the glass and what she subsequently did. Connecting “the broken pieces” with the glass supplies a referential bridge; understanding that the glass broke supplies an unstated event connecting the sentences. That it fell from the table is a plausible causal explanation, but not a logical certainty. That a cat knocked it down, or that Lena was upset, is merely possible elaboration. The distinction is therefore not simply between explicit and implicit information: implicit interpretations differ in how strongly the text constrains them. (condor.depaul.edu)\n\nWalter Kintsch’s 1988 construction–integration account explains how relevant knowledge can support these connections without making every activated association part of the final interpretation. Construction activates word meanings, propositions, and associated knowledge, including candidates that may be inappropriate in context. Integration strengthens mutually supporting elements within the resulting network, while poorly supported alternatives lose activation. On this account, coherence emerges through selection among activated possibilities rather than through indiscriminate accumulation of everything a reader knows about the topic. (condor.depaul.edu)\n\nBackground knowledge consequently supplies both useful connections and expectations that remain revisable. Knowing that fragile objects can break helps connect the two sentences about Lena; a continuation stating that the pieces came from a previously broken plate would require replacing that interpretation. The situation model must accommodate the additional textual evidence rather than preserve the first plausible explanation. This illustrates the interaction identified in the PISA framework between representing literal meaning and integrating it with prior knowledge through mapping and inference: understanding depends on knowledge, but the representation being constructed remains accountable to what the passage actually says. (oecd.org)"
    },
    {
      "heading": "Text Structure, Genre, and Disciplinary Reading",
      "text": "Narratives typically organize information around characters in a setting, their goals, the events that help or obstruct those goals, and eventual outcomes. Understanding a plot therefore involves explaining why events matter to characters, not merely remembering their order. Informational texts organize relationships differently: sequence establishes stages or chronological order; comparison aligns subjects along shared dimensions; cause and effect explains how one condition produces another; and problem–solution connects a difficulty with proposed responses. These structures provide frameworks for distinguishing central information from supporting detail. The September 2010 Institute of Education Sciences practice guide, developed by a panel chaired by Timothy Shanahan, identified recognizing and using textual organization as a comprehension practice supported by moderate evidence for kindergarten through third grade. (thecorestandards.org)\n\nText format is distinct from organizational structure. The OECD’s PISA 2018 reading framework distinguishes continuous texts, composed principally of sentences and paragraphs, from noncontinuous texts, including tables, schedules, diagrams, and forms. A continuous explanation may express a comparison verbally; a table may encode the same relationship through aligned rows and columns. Mixed texts combine formats within one coherent document, such as a scientific report containing prose, a graph, and its legend. Reading them requires coordinating representations rather than treating graphics as decoration. For example, interpreting a report’s claim that rainfall increased requires relating the sentence to the graph’s dates, measurement units, and plotted values. (oecd.org)\n\nDisciplinary reading concerns how a field constructs and communicates knowledge, not simply the topic named in a passage. In their 2008 study, Timothy Shanahan and Cynthia Shanahan reported contrasting approaches among historians, chemists, and mathematicians. Historians attended to authorship, perspective, and the circumstances in which an account was produced; chemists coordinated prose with diagrams, formulas, and descriptions of experimental processes. Specialized terminology also changes the interpretive task: an everyday meaning may be inadequate for a disciplinary concept. Their discussion of dissolution, for instance, shows how scientific language names a general process rather than merely recounting a particular event such as salt dissolving. These findings describe the experts studied, rather than a universal procedure followed by every reader in each field. (shanahanonliteracy.com)\n\nLiterary reading may require attention to narrative perspective, figurative language, and discrepancies between literal statements and implied meanings. A narrator’s description can reveal that narrator’s limitations rather than establish an uncontested fact within the story; irony can make a literal paraphrase inadequate. Reading an argument instead requires identifying claims, reasons, supporting evidence, and the connections among them. A historical account may combine narrative sequence with an argument about causes, while a scientific explanation may combine causal reasoning with a comparison of competing explanations. Genre expectations therefore guide interpretation without assigning every document a single, exclusive structure. (thecorestandards.org)\n\nText complexity involves vocabulary, sentence structure, organization, levels of meaning, conceptual density, and assumed knowledge. Closely packed technical concepts or unstated background assumptions can make a short passage demanding, even when its sentences are brief. The Common Core framework accordingly distinguishes quantitative measures from qualitative judgments about meaning, structure, language, and knowledge demands, and also considers the reader and assigned task. Complexity is not identical to question difficulty: locating one explicitly stated date in a demanding historical account can be easier than explaining an unstated motive in a linguistically simple story. PISA likewise treats difficulty as dependent both on the text and on the operations required, including the amount of information to integrate and the inferences to make. (thecorestandards.org)"
    },
    {
      "heading": "Development and Differences Among Readers",
      "text": "Early oral-language development contributes to later understanding of print: children draw on knowledge of word meanings, grammar, and connected discourse when interpreting written sentences and passages. Longitudinal evidence shows that some comprehension difficulties precede formal reading instruction rather than arising only after unsuccessful reading. In a 2010 study, Kate Nation, Joanne Cocksey, Jo S. H. Taylor, and Dorothy V. M. Bishop followed 242 children from age 5, assessing them again at ages 5½, 6, 7, and 8. Fifteen children identified at age 8 as having poor comprehension despite age-appropriate reading accuracy and fluency were compared with 15 controls. Their weaknesses in expressive and receptive language, listening comprehension, and grammatical understanding were already evident at school entry, whereas their phonological skills were generally normal. These findings distinguish early language vulnerabilities from difficulties confined to recognizing printed words. (doi.org)\n\nReader differences can therefore be described through contrasting educational profiles, rather than inferred from a single comprehension score. One profile involves weak word recognition alongside relatively stronger understanding of spoken language: access to the printed words constrains understanding. A second involves adequate word recognition but weak linguistic comprehension: accurate oral reading can coexist with difficulty understanding sentences, relationships, or the passage as a whole. A third combines weaknesses in both components. These are descriptions of performance, not diagnoses or mutually exclusive explanations of every reading problem. Hugh W. Catts, Suzanne M. Adlof, and Susan Ellis Weismer’s 2006 study compared poor comprehenders, poor decoders, and typical readers; it documented language weaknesses among poor comprehenders extending back to kindergarten. The distinction makes clear why similar levels of reading comprehension can accompany different underlying strengths and weaknesses. (pubmed.ncbi.nlm.nih.gov)\n\nDevelopment also changes the demands against which those capacities operate. Longer texts require readers to maintain connections across paragraphs, while specialized material presupposes increasingly specific vocabulary and subject knowledge. Success with short, familiar passages consequently does not establish readiness for extended explanations or unfamiliar disciplinary content. The RAND Reading Study Group’s 2002 report treated comprehension development as continuing beyond the acquisition of accurate word reading, emphasizing the interaction of linguistic knowledge, topic knowledge, reading processes, and engagement. (rand.org)\n\nReading experience, interests, motivation, and learning opportunities are associated with differences among readers, but their relationships cannot be reduced to a one-way causal sequence. A child who understands reading material readily may choose to read more; reading may, in turn, provide further encounters with language and information. Ulrich Schiefele, Franziska Stutz, and Ellen Schaffner’s 2016 longitudinal study examined reciprocal relations between motivation and comprehension, finding reciprocal associations involving intrinsic involvement and word- and sentence-level comprehension. Such observational relationships do not by themselves establish causation. Access to books, classroom learning, and opportunities for discussion also forms part of the context in which reading experience accumulates; differences in performance cannot automatically be attributed to individual effort or preference. (sciencedirect.com)\n\nVariation occurs within readers as well as between them. The same person may understand a familiar topic more fully than an unfamiliar one, or read differently when locating a fact and when studying an explanation. Interest, attention, available knowledge, and the demands of the particular text and activity alter performance across occasions. RAND’s framework explicitly distinguishes these intra-individual differences from differences among people. Comprehension is thus not a wholly context-independent capacity: a reader’s relatively enduring skills interact with what is being read, why it is being read, and the conditions under which reading occurs. (rand.org)"
    },
    {
      "heading": "Multilingual and Additional-Language Reading",
      "text": "Learning to read and learning the language of a text are related but distinct processes. A learner may already understand how written texts work while still acquiring the vocabulary and syntax needed to read in an additional language; another may be developing both abilities simultaneously. Existing literacy therefore matters independently of conversational proficiency. The National Academies’ 2017 review, Promoting the Educational Success of Children and Youth Learning English, found positive relationships between first-language literacy and English literacy development. Such relationships do not imply that every skill transfers unchanged: readers must also learn the conventions of the new writing system and the linguistic forms through which familiar concepts are expressed. (nationalacademies.org)\n\nAcademic reading adds demands that everyday conversation may not reveal. Readers encounter abstract vocabulary, complex grammatical relationships, and familiar words with unfamiliar disciplinary meanings. The 2014 Institute of Education Sciences guide distinguishes general academic words, used across subjects, from domain-specific terminology. For example, factor denotes a multiplicative relationship in mathematics but a contributing circumstance in historical explanation. Recognizing the printed word does not establish which meaning applies. Understanding a passage consequently requires coordinating word meanings with sentence structure and subject knowledge, rather than translating isolated words or relying on conversational fluency alone. (ies.ed.gov)\n\nResearch-supported instruction has addressed language and subject learning together rather than treating English proficiency as a prerequisite for access to academic content. The April 2014 guide, developed by a panel chaired by Scott Baker for elementary and middle schools, rated intensive academic-vocabulary teaching and integration of oral and written English into content-area instruction as supported by strong evidence. It rated small-group intervention as supported by moderate evidence, while the separate recommendation for regular structured writing opportunities received a minimal-evidence rating. These ratings concern the supporting research base, not a claim that writing is unimportant. The guide’s intended setting is school instruction for English learners through grade 8, rather than adult additional-language learning generally. (ies.ed.gov)\n\nThe instructional mechanism combines meaningful encounters with content and repeated opportunities to use its language: students read informational material, discuss ideas in pairs, examine visual representations, and write explanations using selected vocabulary. One study summarized in the guide was Diane August and colleagues’ 2009 randomized controlled trial of integrated science and vocabulary instruction with 562 sixth-grade English learners. It provided approximately 30 hours of instruction over nine weeks and produced a statistically significant effect of 0.26 standard deviations on a researcher-developed vocabulary measure. This outcome documents vocabulary learning in that setting; it is not itself evidence of an equivalent gain in general reading comprehension. (ies.ed.gov)\n\nAssessment interpretation depends on which language was tested, the learner’s proficiency in that language, and opportunities to acquire literacy through schooling. ETS’s 2009 Guidelines for the Assessment of English Language Learners, by Mary Pitoniak and colleagues, emphasizes variation in home-language literacy, English proficiency, and educational background among United States K–12 learners. Speaking a home language does not necessarily entail reading it, so translated directions are not automatically accessible. Conversely, difficulty answering an English-language question may reflect unfamiliar vocabulary or response-language demands as well as difficulty understanding the passage. A score in one language therefore describes performance under particular linguistic conditions, not the learner’s complete literacy or knowledge across languages. (ets.org)"
    },
    {
      "heading": "Comprehension Monitoring and Reading Strategies",
      "text": "Comprehension monitoring is the metacognitive activity of judging whether a text makes sense and recognizing gaps, contradictions, or a loss of understanding. It concerns the reader’s understanding, rather than simply whether words have been pronounced correctly or a page completed. Monitoring can involve checking whether a paragraph’s main point can be restated, whether a new statement fits earlier information, or whether an unfamiliar expression prevents interpretation. The 2022 What Works Clearinghouse guide for reading interventions in grades 4–9 distinguishes recognizing a breakdown from taking action to resolve it: readers first identify what is unclear, then select a response. Its examples begin with detecting nonsensical sentences and extend to checking understanding across longer passages. (ies.ed.gov)\n\nRepair strategies address the particular source of difficulty. A reader may reread a sentence more slowly, return to an earlier paragraph, check a word’s meaning, or reconsider how two ideas connect. Rereading is therefore not merely repeating the same words: it can be directed toward a specific unresolved relationship. For example, in the invented passage “Lena lent Priya a notebook. She needed it for the experiment,” a reader may notice that she has no unambiguous referent and inspect surrounding sentences rather than assume an interpretation. Repair also includes seeking additional information when the passage alone is insufficient. The grades 4–9 guide illustrates this process with a biography of Mary Winston Jackson: difficulty explaining what Jackson did prompts a slower rereading of the relevant section. (ies.ed.gov)\n\nA reading strategy is a deliberate cognitive action selected to support understanding or memory. The 2010 What Works Clearinghouse guide for kindergarten through grade 3 defines strategies as intentional routines used before, during, or after reading. This distinguishes consciously checking a pronoun’s referent from resolving it without noticeable effort during increasingly fluent processing. Familiar routines can become easier to carry out, but strategic regulation still involves deciding when extra attention is needed. Completing a worksheet or highlighting sentences does not, by itself, constitute a comprehension strategy: the relevant activity is the thinking that selects, connects, or checks information. Strategies may be used separately or together, rather than as a compulsory sequence for every text. (ies.ed.gov)\n\nQuestioning and prediction direct attention toward information that matters for the reading purpose. Content questions, such as “Why did the character change plans?”, encourage examination of motives and evidence; monitoring questions, such as “Can this change of plan be explained from the passage?”, check the adequacy of understanding. Prediction uses textual clues and relevant prior knowledge to form a provisional expectation, which subsequent reading can confirm or revise. For example, a character’s discovery of dark clouds may support predicting a postponed journey, but that prediction remains tentative until later events establish the outcome. The Institute of Education Sciences’ K–3 comprehension toolkit treats prediction as connected to prior knowledge and continued reading, not as guessing detached from the text. (ies.ed.gov)\n\nSummarizing and graphic organization make relationships among ideas available for inspection. A summary selects central information and expresses it compactly rather than reproducing every sentence; an inability to construct one can expose incomplete understanding. A graphic organizer can arrange events in sequence, place similarities and differences in parallel columns, or connect causes with consequences. For an invented passage comparing two transport systems, a comparison chart might separate cost, capacity, and travel time, making an omitted comparison visible. These activities can support preparation, ongoing interpretation, or review after reading. Their use depends on the task: a sequence diagram may clarify a multistage process, whereas a brief factual lookup may require neither a diagram nor a written summary. (ies.ed.gov)"
    },
    {
      "heading": "Instruction and Evidence of Effectiveness",
      "text": "Explicit comprehension instruction makes the reasoning involved in understanding a text visible and gives learners supported opportunities to practise it. A teacher explains a strategy’s purpose and when it is useful, models its application through a think-aloud, and guides students before transferring responsibility to them. For example, when teaching inference, the teacher can identify a textual clue, connect it with relevant knowledge, and explain why that combination supports an interpretation. During guided practice, students supply the connections while the teacher prompts or corrects their reasoning; independent practice requires them to make and justify connections without those prompts. This gradual transfer concerns control over meaning-making, not merely completion of increasingly difficult worksheets. The Institute of Education Sciences’ kindergarten–grade 3 toolkit describes this progression as explicit explanation, modeling, collaborative use, guided practice, and independent use. (ies.ed.gov)\n\nStructured discussion extends instruction beyond checking whether answers are correct. Teachers pose questions about important ideas, ask students to explain their reasoning with textual evidence, and encourage them to consider classmates’ interpretations. Instruction in text organization similarly connects form with meaning: a cause-and-effect organizer, for example, can help students distinguish an event’s causes from its consequences rather than simply locate sentences containing particular words. Purposeful text selection provides material suited to these goals, considering the text’s quality, difficulty, content, and opportunities for applying the intended strategy. A narrative with conflicting character goals offers different instructional opportunities from an explanation organized around a scientific process. These practices teach readers how to construct understanding; administering comprehension questions alone does not supply the same modeling and feedback. (ies.ed.gov)\n\nVocabulary, foundational reading skills, and comprehension instruction address complementary demands. Instruction connecting phonemes with letters and teaching decoding supports access to written words; connected-text reading develops accuracy and fluency. Vocabulary teaching develops the meanings needed to interpret those words, including academic terms used across subjects. Comprehension instruction addresses relationships among ideas and the use of information across a passage. The 2016 IES foundational-skills guide therefore includes academic language alongside sound–letter relationships, decoding, and daily connected-text reading. Its vocabulary guidance emphasizes repeated encounters and use across activities rather than isolated definition memorization. These components are not interchangeable: successfully decoding an unfamiliar academic word does not establish its meaning, while knowing a word orally does not establish the ability to recognize its printed form. (ies.ed.gov)\n\nEvidence ratings distinguish recommended practice from the strength of the available causal research. The September 2010 What Works Clearinghouse guide, developed by a panel chaired by Timothy Shanahan, assigned strong evidence to comprehension-strategy instruction and moderate evidence to text-structure instruction. Structured discussion and purposeful text selection received minimal evidence ratings. These ratings concerned the kindergarten–grade 3 research reviewed for that guide, not every age group or subsequent study. For discussion, few studies had examined the target grades and none met the review’s evidence standards; for text selection, one qualifying study addressed only part of the recommendation. Minimal evidence thus indicated limitations in the qualifying research, rather than a demonstrated absence of benefit. (ies.ed.gov)\n\nIntervention findings depend on experimental design, the comparison instruction, and the learners studied. A gain relative to ordinary classroom provision answers a different question from a gain relative to another intensive intervention. Evaluation also considers initial reading and language skills, implementation, and whether an instructional package permits effects to be attributed to individual components. Outcome selection is especially consequential: success on taught passages or closely aligned questions provides narrower evidence than improvement on unfamiliar texts. The RAND Reading Study Group’s 2002 report, chaired by Catherine Snow, distinguished immediate understanding from lasting improvement in comprehension capacity. It noted that explicit instruction could produce immediate strategy use while evidence for continued use after instruction and transfer to new situations was less secure. Delayed follow-up and unfamiliar-text measures therefore address claims that immediate post-tests alone cannot establish. (ies.ed.gov)"
    },
    {
      "heading": "Assessment and Interpretation of Performance",
      "text": "Selected-response questions, including multiple-choice items, ask readers to recognize an answer among alternatives; depending on their design, they can assess retrieval, inference, interpretation, or evaluation. Constructed responses require readers to supply an answer, explain a relationship, or support an interpretation with textual evidence. They make aspects of reasoning available for scoring but also introduce demands on written expression. Neither format corresponds automatically to a particular depth of understanding: a multiple-choice question can require integration across a passage, whereas a short written answer may require only copying a detail. The PISA 2018 framework therefore treats response format separately from reading process and specifies that spelling and grammar should not receive undue weight in reading scores. (oecd.org)\n\nRetellings, summaries, and cloze tasks elicit different evidence. A retelling asks readers to reconstruct a text, revealing which events, ideas, and relationships they retain; the National Reading Panel’s 2000 review included story recall and retelling among comprehension measures. A summary places a stronger requirement on selecting central information and compressing it into a coherent account rather than reproducing every remembered detail. Cloze tasks remove words from sentences or passages and require readers to supply replacements. These test the use of linguistic and contextual constraints, but success on a short item need not demonstrate understanding of an extended argument. For example, completing “The puppy wagged its ___” provides narrower evidence than explaining how two paragraphs support an author’s conclusion. (nichd.nih.gov)\n\nValidity concerns whether evidence supports the intended interpretation and use of a score; reliability concerns its consistency and precision. Passage selection matters because performance reflects interactions between readers and texts, including prior topic knowledge, vocabulary, and familiarity with a genre. A science passage can therefore yield a different picture from a narrative passage for the same reader. Response requirements can introduce additional variation through writing, oral expression, or memory demands. Sampling several topics and texts helps distinguish broadly recurring difficulties from passage-specific performance; many questions about one passage do not provide the same coverage as questions distributed across different passages. The 2002 RAND Reading Study Group report, chaired by Catherine Snow, criticized assessments that reduced comprehension to a single, method-dependent dimension and called for measures sensitive to reader, text, and activity. (rand.org)\n\nThe Progress in International Reading Literacy Study (PIRLS), first administered in 2001, assesses reading principally in the fourth grade. Its two reading purposes are literary experience and acquiring and using information. PIRLS reports overall achievement, purpose-related performance, and percentages reaching international benchmarks: Low at 400, Intermediate at 475, High at 550, and Advanced at 625. These benchmarks describe progressively more demanding performances rather than percentages of questions answered correctly. The Programme for International Student Assessment (PISA) instead assesses 15-year-old students across grades. Its reading framework addresses using texts in personal, public, educational, and occupational situations, rather than mastery of a particular school curriculum. PISA 2018 reported reading proficiency from Level 1c through Level 6, alongside performance below Level 1c. PIRLS and PISA scores are consequently not interchangeable measures of the same population or reading demands. (nces.ed.gov)\n\nIndividual educational assessment seeks evidence about a learner’s strengths, difficulties, and instructional needs; population assessments estimate distributions and differences across education systems or groups. A population mean does not identify the particular processes limiting an individual’s understanding. Nor does an achievement gap by itself identify its cause. Associations between scores and teaching practices, family resources, or school characteristics can reflect selection, reciprocal influences, and other measured or unmeasured differences. OECD methodological guidance explicitly identifies PISA’s cross-sectional design as a limitation on causal interpretation: establishing causation requires evidence beyond the observed correlation between a characteristic and reading performance. (rand.org)"
    },
    {
      "heading": "Digital Reading and Evaluation Across Sources",
      "text": "Reading on the World Wide Web requires decisions about which texts to access as well as interpretation of the texts selected. Readers use search engines, menus, hyperlinks, tabs, and scrolling to locate material rather than necessarily following a predetermined sequence. The PISA 2018 reading framework distinguishes retrieving information within a document from searching for and selecting relevant documents. In the latter activity, headings, search-result descriptions, authorship, and publication dates provide preliminary clues to relevance. Navigation therefore involves judgments about meaning: whether a link is likely to answer the question, whether a passage warrants close reading, and when to leave irrelevant material. These activities connect comprehension with information retrieval. (oecd.org)\n\nAcross documents, integration requires preserving relationships between information and its sources. Remembering that one author advances a claim while another disputes it differs from combining both statements into an undifferentiated account. Readers must connect claims with supporting evidence while retaining who said what and where disagreement occurs. The PISA framework explicitly treats assigning conflicting claims to their respective sources as part of handling disagreement. A coherent representation of several texts can therefore include unresolved alternatives rather than a single reconciled narrative. (oecd.org)\n\nCredibility evaluation goes beyond identifying what a text says. It concerns the quality and validity of the information, the publication context, and the relationship between presentation and authorial purpose. The OECD’s account of PISA 2018 distinguishes understanding literal or inferred meaning from evaluating content and form; it also emphasizes triangulating sources and distinguishing factual statements from opinions. Understanding a claim consequently does not entail accepting it: a reader may accurately reconstruct an argument while judging its evidence insufficient. Relevance and credibility are likewise separate judgments—a text can address the question directly without providing a dependable answer. (oecd.org)\n\nLateral reading investigates a website by leaving it temporarily to examine information elsewhere. In a study published in 2019, Sam Wineburg and Sarah McGrew observed 45 participants: 10 professional fact-checkers, 10 historians with doctorates, and 25 Stanford undergraduates. Participants thought aloud while evaluating live websites. Fact-checkers commonly opened additional tabs to investigate a source’s credibility, whereas historians and students more often remained within the original site. The latter groups were sometimes influenced by official-looking logos or domain names. Fact-checkers reached better-supported judgments more quickly in these tasks; the study compared observed practices rather than randomly assigning participants to reading strategies. (doi.org)\n\nA concrete example of cross-source reasoning appears in the PISA 2018 “Rapa Nui” unit. Readers encountered three texts: a professor’s blog, a review of Jared Diamond’s Collapse, and an article presenting an alternative explanation for the disappearance of the island’s large trees. One question required an explanation supported by specific information from these sources. Scoring did not prescribe a single historical conclusion: both a supported interpretation and a source-grounded judgment that further information was needed could receive full credit. Vague expressions of uncertainty did not suffice. The task thus distinguished comparison of competing accounts from either uncritical acceptance or unsupported rejection. (oecd.org)"
    },
    {
      "heading": "Machine Reading Comprehension and Its Limits",
      "text": "Machine reading comprehension is a task in natural language processing in which a system receives a passage and a question and predicts an answer supported by that passage. A benchmark typically supplies training examples and evaluates predictions against reference answers on held-out examples. This operational definition concerns observable question-answering performance; it does not make the task equivalent to the human educational construct of comprehension. In particular, producing a correct answer does not, by itself, establish how a system used the question or represented the passage. (aclanthology.org)\n\nIn extractive reading comprehension, the answer is a contiguous sequence of words already present in the passage. The Stanford Question Answering Dataset (SQuAD), introduced by Pranav Rajpurkar and colleagues in 2016, contained more than 100,000 crowdworker-written questions about Wikipedia passages, with answers annotated as text spans. A system can therefore formulate answering as selecting the beginning and end of a span rather than generating a new explanation. This format supports standardized evaluation but excludes questions whose answers cannot be expressed as a single passage span, including some questions requiring synthesis across separated statements. (arxiv.org)\n\nNarrativeQA, introduced by Tomáš Kočiský and colleagues in a December 2017 preprint and published in 2018, was designed to emphasize integration across extended narratives. It contains 46,765 questions associated with 1,572 books and movie scripts. Annotators wrote questions and answers from independently written summaries; systems could then be evaluated using either summaries or full stories. In the full-story setting, differences between a summary’s wording and the original text reduce the usefulness of direct lexical matching. Questions target characters, events, and relationships: one published example asks how Oscar is related to Dana in Ghostbusters II, with the answer “her son.” The task design encourages locating relevant evidence and interpreting relationships rather than copying a matching sentence. (arxiv.org)\n\nDROP, introduced by Dheeru Dua and colleagues in 2019, targets discrete reasoning over paragraphs. Its 96,567 questions require operations such as addition, subtraction, counting, comparison, and sorting, combined with identifying the relevant entities and resolving references. Answers can be text spans, dates, or numbers. As an illustrative operation, a question asking the difference between two teams’ scores requires identifying which numbers denote their scores and subtracting them; the result need not occur anywhere in the passage. DROP thus separates locating textual evidence from computing an answer over that evidence. (ucinlp.github.io)\n\nBenchmark shortcuts arise when statistical regularities permit successful prediction without using all the information the task ostensibly requires. In 2018, Divyansh Kaushik and Zachary C. Lipton tested models given only questions or only passages. Passage-only models exceeded 50% accuracy on 14 of the 20 bAbI tasks, sometimes matching models with both inputs. For the Children’s Book Test, retaining only the last sentence of a 20-sentence context could yield comparably accurate predictions. Such ablation experiments test whether performance actually depends on the supposedly necessary input, rather than assuming that a reading-themed dataset necessarily measures integration. (aclanthology.org)\n\nSensitivity to irrelevant material provides another diagnostic. Robin Jia and Percy Liang’s 2017 adversarial evaluation appended distracting sentences to SQuAD passages without changing the correct answers. Across 16 published models, average F1 fell from 75% to 36%. The result demonstrated that strong ordinary-test performance could coexist with vulnerability to superficial cues. Accordingly, matching a human benchmark score establishes performance under that benchmark’s particular questions and scoring procedure, not equivalence to human reading processes; input ablations and controlled distractors probe dependencies that aggregate scores leave unspecified. (aclanthology.org)"
    }
  ],
  "links": [
    "language",
    "literacy",
    "cognitive-psychology",
    "linguistics",
    "education",
    "literature",
    "knowledge",
    "writing-system",
    "grapheme",
    "phoneme",
    "morphology",
    "syntax",
    "narrative",
    "plot-literature",
    "narrative-perspective",
    "irony",
    "language-acquisition",
    "grammar",
    "causation",
    "randomized-controlled-trial",
    "experimental-design",
    "correlation",
    "world-wide-web",
    "information-retrieval",
    "natural-language-processing"
  ],
  "sources": [
    {
      "title": "Reading for Understanding: Toward an R&D Program in Reading Comprehension",
      "url": "https://www.rand.org/content/dam/rand/pubs/monograph_reports/2005/MR1465.pdf"
    },
    {
      "title": "The simple view of reading",
      "url": "https://doi.org/10.1007/bf00401799"
    },
    {
      "title": "Decoding, Reading, and Reading Disability",
      "url": "https://cir.nii.ac.jp/crid/1363951795036118784"
    },
    {
      "title": "Foundational Skills to Support Reading for Understanding in Kindergarten Through 3rd Grade",
      "url": "https://ies.ed.gov/ncee/wwc/PracticeGuide/21/Published"
    },
    {
      "title": "Foundational Skills to Support Reading for Understanding in Kindergarten Through 3rd Grade",
      "url": "https://ies.ed.gov/ncee/WWC/Docs/PracticeGuide/wwc_foundationalreading_040717.pdf"
    },
    {
      "title": "Report of the National Reading Panel",
      "url": "https://www.nichd.nih.gov/publications/pubs/nrp/findings"
    },
    {
      "title": "Children’s surface, textbase, and situation model representations of written and illustrated written narrative text",
      "url": "https://link.springer.com/article/10.1007/s11145-020-10118-1"
    },
    {
      "title": "Improving Reading Comprehension in Kindergarten Through 3rd Grade",
      "url": "https://ies.ed.gov/ncee/wwc/PracticeGuide/14"
    },
    {
      "title": "Teaching Disciplinary Literacy to Adolescents: Rethinking Content-Area Literacy",
      "url": "https://www.shanahanonliteracy.com/upload/publications/38/pdf/ShanahanArticle061909.pdf"
    },
    {
      "title": "Common Core State Standards for English Language Arts & Literacy in History/Social Studies, Science, and Technical Subjects",
      "url": "https://www.thecorestandards.org/assets/Appendix_A.pdf"
    },
    {
      "title": "Language deficits in poor comprehenders: a case for the simple view of reading",
      "url": "https://pubmed.ncbi.nlm.nih.gov/16671844/"
    },
    {
      "title": "Longitudinal Stability and Predictors of Poor Oral Comprehenders and Poor Decoders",
      "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC3661702/"
    },
    {
      "title": "Promoting the Educational Success of Children and Youth Learning English: Promising Futures—The Development of English Language Proficiency in Grades K–12",
      "url": "https://www.nationalacademies.org/read/24677/chapter/8"
    },
    {
      "title": "Teaching Academic Content and Literacy to English Learners in Elementary and Middle School",
      "url": "https://ies.ed.gov/ncee/wwc/Docs/practiceguide/english_learners_pg_040114.pdf"
    },
    {
      "title": "WWC: Teaching Academic Content and Literacy to English Learners in Elementary and Middle School",
      "url": "https://ies.ed.gov/ncee/WWC/PracticeGuide/19/Published"
    },
    {
      "title": "Guidelines for the Assessment of English Language Learners",
      "url": "https://www.ets.org/pdfs/about/ell-guidelines.pdf"
    },
    {
      "title": "Providing Reading Interventions for Students in Grades 4–9",
      "url": "https://ies.ed.gov/ncee/WWC/Docs/PracticeGuide/WWC-practice-guide-reading-intervention-full-text.pdf"
    },
    {
      "title": "Improving Reading Comprehension in Kindergarten Through 3rd Grade",
      "url": "https://ies.ed.gov/ncee/wwc/Docs/PracticeGuide/readingcomp_pg_092810.pdf"
    },
    {
      "title": "WWC: Improving Reading Comprehension in Kindergarten Through 3rd Grade",
      "url": "https://ies.ed.gov/ncee/WWC/PracticeGuide/14/Published"
    },
    {
      "title": "Foundational Skills to Support Reading for Understanding in Kindergarten Through 3rd Grade",
      "url": "https://ies.ed.gov/ncee/wwc/Docs/practiceGuide/wwc_foundationalreading_040717.pdf"
    },
    {
      "title": "National Reading Panel—Reports of the Subgroups: Chapter 4, Part II, Text Comprehension Instruction",
      "url": "https://www.nichd.nih.gov/sites/default/files/publications/pubs/nrp/Documents/ch4-II.pdf"
    },
    {
      "title": "Individual differences in fifth graders’ reading and language predict their comprehension monitoring development: An eye-movement study",
      "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC4824948/"
    },
    {
      "title": "NCES Handbook of Survey Methods—Progress in International Reading Literacy Study (PIRLS)",
      "url": "https://nces.ed.gov/statprog/handbook/pirls.asp"
    },
    {
      "title": "PIRLS 2021—Methodology and Technical Notes",
      "url": "https://nces.ed.gov/surveys/pirls/pirls2021/technotes.asp"
    },
    {
      "title": "Description of PIRLS international reading benchmarks: 2016",
      "url": "https://nces.ed.gov/surveys/pirls/pirls2016/tables/pirls2016_exhibit02.asp"
    },
    {
      "title": "Average reading literacy scores of 15-year-old students and percentage attaining reading literacy proficiency levels, by country or other education system: 2018",
      "url": "https://nces.ed.gov/programs/digest/d22/tables/dt22_602.50.asp?current=yes"
    },
    {
      "title": "School Factors Related to Quality and Equity",
      "url": "https://www.oecd.org/content/dam/oecd/en/publications/reports/2005/04/school-factors-related-to-quality-and-equity_g1gh5627/9789264008199-en.pdf"
    },
    {
      "title": "How Much Reading Does Reading Comprehension Require? A Critical Investigation of Popular Benchmarks",
      "url": "https://aclanthology.org/D18-1546/"
    },
    {
      "title": "SQuAD: 100,000+ Questions for Machine Comprehension of Text",
      "url": "https://arxiv.org/abs/1606.05250"
    },
    {
      "title": "The NarrativeQA Reading Comprehension Challenge",
      "url": "https://arxiv.org/abs/1712.07040"
    },
    {
      "title": "The NarrativeQA Reading Comprehension Challenge",
      "url": "https://aclanthology.org/Q18-1023.pdf"
    },
    {
      "title": "DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs",
      "url": "https://ucinlp.github.io/files/papers/drop-naacl19.pdf"
    },
    {
      "title": "DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs",
      "url": "https://aclanthology.org/N19-1246.pdf"
    },
    {
      "title": "Adversarial Examples for Evaluating Reading Comprehension Systems",
      "url": "https://aclanthology.org/D17-1215/"
    }
  ],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}