Model of Text Comprehension
What it is
The Model of Text Comprehension explains how the mind converts a text into understanding through three levels of mental representation. The surface code holds the exact words and syntax, the textbase holds the propositions (the basic idea units) and the connections among them, and the situation model is the reader's mental picture of what the text is actually about, built by fusing the textbase with prior knowledge and inference.
The core idea
Comprehension is not passive decoding but active construction. Readers extract propositions from sentences, link them into a coherent textbase, and then integrate that textbase with what they already know to form a situation model. In Kintsch's later construction-integration account, a messy first pass loosely activates many candidate meanings (construction), after which a settling process strengthens what fits the context and suppresses the rest (integration).
How it is used
Researchers use the model to predict what readers will recall, summarize, and infer, and to explain why some texts are harder than others. It anchors readability and instructional-design work, the assessment of reading comprehension, and the design of educational texts. It also underwrites computational measures of text coherence and inference load, and informs how summaries, captions, and interfaces should be sequenced to keep the textbase coherent.
In practice
Reading "The thief fled. The police gave chase," a reader rarely remembers the precise wording (surface code) for long. She does retain the propositions that someone fled and was pursued (textbase), and she infers, though the text never says so, that the police were chasing the thief because of the theft. That inferred, integrated understanding, drawing on knowledge of crime, is her situation model.
Key studies & evidence
Walter Kintsch and Teun van Dijk introduced the model in their 1978 Psychological Review paper "Toward a Model of Text Comprehension and Production," formalizing propositions, the textbase, and a cyclical process limited by working memory. Their 1983 book "Strategies of Discourse Comprehension" added the situation model and a strategic, knowledge-driven view of processing. Kintsch's 1988 construction-integration model supplied a hybrid symbolic and connectionist mechanism: weakly constrained construction followed by integration through spreading activation. Empirical work showed that readers' memory degrades fastest for surface form, more slowly for the textbase, and is most durable for the situation model, and that reading times rise with the number of new arguments and required inferences, supporting the proposed architecture.
Critiques & limitations
The original model is computationally demanding and was criticized for treating inference and knowledge use somewhat vaguely before the 1988 revision. Hand-coding propositions is laborious and not fully reproducible across analysts. Rival accounts press different points: the structure-building framework of Gernsbacher emphasizes mapping and shifting, the landscape model of van den Broek stresses fluctuating activation over a whole reading, and constructionist "search after meaning" theories argue readers generate richer goal-driven inferences than the relatively passive construction phase implies. Critics also note the model centers on expository and narrative prose and on individual readers, saying less about multimodal, conversational, or socially situated comprehension.
Applications
The model is a workhorse in reading education, text design, and assessment, guiding how to write coherent instructional materials and how to test for deep rather than verbatim understanding. In communication teaching it clarifies why message design must respect a reader's prior knowledge and inference burden, not just word choice. The framework also extends to mediated and screen-based contexts studied in the AURA Lab: captions and on-screen text in streaming, the sequencing of information in social-media posts, and signage or narration in social VR all succeed or fail on whether viewers can build a coherent textbase and a usable situation model. It further informs natural-language-processing measures of text coherence and the evaluation of machine reading comprehension.