HSW-0212 · How Studying Works
A diagram can disagree with the paragraph beside it, and your reading system can notice enough to slow down without noticing enough to tell you that something is wrong.
This is one of the quiet dangers of modern studying. Textbooks, slides, websites and AI-assisted notes are increasingly multimodal. A sentence explains the graph. A caption explains the photograph. A diagram names a mechanism. A worked solution sits beside a geometric figure. Learners often assume that because the pieces are on the same page, somebody has already checked that they agree.
That assumption is convenient. It is also unsafe.
Quick answer
Recent research on illustrated text found that readers often failed to consciously detect contradictions even when their reading behaviour showed some sensitivity to the inconsistency. More than 40% of tested inconsistencies were overlooked across the reported experiments. Adding a picture did not reliably improve conscious detection when the contradiction was already present within the text, and contradictions that had to be checked across text and picture were missed even more often.
The studying lesson is not that pictures are bad for learning. Pictures can be enormously useful. It is that integration is not the same as verification. A learner can process several representations without deliberately checking whether they make the same claim.
This article owns one narrow job: how to detect and repair contradictions across words, diagrams, graphs, images and other representations before the inconsistency becomes part of what you learn.
Passive reading contains a weak alarm system
Readers do not simply accept every sentence. Comprehension involves a form of validation: incoming information is compared with what is already active in memory and with the developing representation of the text. When something conflicts, processing can become more difficult.
The important problem is that a processing disturbance does not always become a conscious judgment such as, “These two claims contradict each other.” The system may register friction without producing a usable alarm.
That distinction is easy to recognise in ordinary life. You reread a sentence because something feels odd. You pause at a graph. A caption takes longer than expected. You know the page is not flowing, but you cannot immediately say why. Sometimes the reason is merely difficult prose. Sometimes the representation conflicts with something else on the page.
What the 2026 experiments found
Frick and Schüler examined consistency validation in illustrated texts in a 2026 Memory & Cognition paper. Across three experiments, participants read short materials containing consistent or inconsistent information and judged whether sentences fit with what they had already read or seen.
The researchers found a striking gap between processing sensitivity and explicit detection. Contradictory information could affect reading time, yet readers still overlooked a substantial proportion of inconsistencies. Across the experiments, more than 40% of inconsistencies were missed.
Pictures did not provide a simple solution. When a contradiction could already be identified within the text, adding a related picture did not reliably improve conscious detection. When the inconsistency had to be identified by comparing a picture with later text — a cross-modal check — readers overlooked even more contradictions. The authors argue that cross-modal validation may be more resource-demanding than checking information within a single representation.
The responsible conclusion is narrow. The study does not show that diagrams reduce learning, or that readers never integrate text and pictures. It shows that passive validation is not reliable enough to guarantee conscious consistency checking, especially when evidence is distributed across different forms.
A page can contain two models at once
Suppose a physics note says that the velocity of an object is increasing while the graph beside it shows a line with constant slope on a displacement–time graph. Depending on the exact graph, the visual representation may imply constant velocity rather than increasing velocity.
A student can read the sentence, glance at the graph and continue. If the verbal statement is more familiar, it may dominate. If the graph is visually salient, the student may remember the picture instead. Either way, the page contains incompatible models.
Learning from that page without checking consistency creates a dangerous state: the learner may possess both representations without realising they disagree. Later, the contradiction appears as hesitation, mixed methods or an answer that changes depending on which cue is present.
The consistency problem is bigger than text versus picture
Students regularly study from paired representations:
- equation and graph;
- map and written description;
- table and conclusion;
- diagram and label;
- photograph and caption;
- worked solution and geometric figure;
- timeline and historical narrative;
- audio explanation and slide;
- AI summary and original source.
Each pair creates a potential verification job. The job is not merely to understand both representations. It is to ask whether they preserve the same quantities, directions, conditions, categories and causal relationships.
Worked example: the graph that quietly contradicts the paragraph
Illustrative case. A revision sheet states: “As temperature rises from 20°C to 60°C, the measured rate doubles every 10°C.” The graph beside it shows values of 2, 3, 5, 7 and 9 units across those temperatures.
A learner who reads only the prose stores a doubling rule. A learner who looks only at the graph stores a rising but non-doubling pattern. A learner who passively scans both may retain a vague idea that “rate increases with temperature” and never notice the mismatch.
An active consistency check asks:
- What numerical relation does “doubles every 10°C” predict?
- Do the plotted values satisfy that relation?
- If not, is the text wrong, the graph wrong, or are they describing different quantities?
- What claim is actually safe to study before the source is corrected?
That small audit turns a passive page into an evidence problem.
Why pictures can make checking harder
It is tempting to assume that a picture always makes a claim easier to inspect. Sometimes it does. But cross-modal checking requires translation. The learner must map a verbal proposition onto a visual relation or map a visual feature back into language.
For example, the sentence “the larger angle is opposite the longer side” is verbal. A triangle diagram represents the same relation spatially. To verify consistency, the learner must identify the relevant side and angle, preserve correspondence, and compare magnitude across forms. That is additional work.
If working memory is already occupied by unfamiliar terminology, calculations or navigation, the learner may integrate enough to continue reading but not enough to produce an explicit contradiction judgment.
The five-point consistency audit
When two representations are supposed to describe the same thing, compare five features deliberately.
- Object: Are both representations actually about the same variable, group, process or time period?
- Direction: Do both say increase, decrease, remain stable, move clockwise, become more concentrated or change in the same direction?
- Magnitude: Do words such as double, half, largest or proportional match the numbers or geometry?
- Condition: Are both claims operating under the same assumptions, units, domain and boundary conditions?
- Timeline: Are sequence, before/after relations and durations aligned?
Most contradictions become easier to locate when the learner stops asking, “Do these look related?” and starts asking, “Do these make the same claim about the same object under the same conditions?”
Reading time is not a sufficient warning
One of the most useful ideas from the research is that slower processing does not guarantee conscious correction. A learner can hesitate at an inconsistency and still accept it.
This matters because educational technology increasingly measures process traces: time on a page, cursor movement, rereads, clicks and pauses. Those signals can indicate difficulty, but they do not reveal the learner’s interpretation directly. A longer look at a contradiction might mean successful detection, confusion, distraction or merely difficult integration.
If accuracy matters, ask for an explicit judgment. “Do these two representations agree? Show me where.” The response is better evidence than dwell time alone.
The representation handoff test
A strong way to check learning is to force the idea to cross the representation boundary.
- Turn the paragraph into a sketch.
- Turn the graph into one precise sentence.
- Turn the equation into a verbal relationship.
- Turn the map into a route description.
- Turn the diagram into a prediction about what changes when one part changes.
If the translation produces a contradiction, investigate before continuing. This method does more than detect errors in the source. It tests whether the learner actually understands what each representation means.
Worked example: Mathematics
Illustrative case. A student is told that a quadratic has two distinct real roots. The accompanying graph touches the x-axis once and turns around. The prose and graph cannot both represent the same quadratic under the usual interpretation.
A passive learner may memorise “two distinct roots” and copy the picture separately. An active learner translates both into the same representation:
- two distinct real roots → two different x-intercepts;
- touches once and turns → one repeated real root.
The contradiction becomes visible because the learner knows what counts as the same mathematical statement across words and graph.
Worked example: Science
Illustrative case. A labelled diagram shows heat flowing from object A to object B. The caption says B is hotter than A and that the system is isolated. Under ordinary spontaneous thermal transfer, those claims raise a consistency problem. Instead of memorising both, the learner asks whether the arrow direction, temperature labels or stated conditions are wrong.
The habit matters because many science misconceptions survive as representational contradictions. A student can recite the correct sentence while carrying an incorrect diagram, or draw a correct graph while explaining it with the wrong causal story.
AI-generated notes need cross-modal verification too
AI can produce polished summaries, diagrams and tables quickly. The speed makes consistency checking more important, not less. A generated explanation may use correct terminology while a generated diagram reverses a direction, mislabels an axis or depicts an impossible geometry.
Do not treat visual polish as evidence of internal agreement. If the note matters, translate each representation back into claims and compare them. For technical topics, verify important claims against authoritative sources rather than allowing one generated representation to validate another generated representation.
When checking every detail becomes too expensive
Active verification has a cost. A student cannot audit every icon, sentence and decorative image. The goal is selective checking.
Prioritise consistency checks when:
- the representation carries a central concept;
- the claim will be used in a calculation or argument;
- the source is unfamiliar or automatically generated;
- units, scale, direction or causal interpretation matter;
- two representations feel difficult to reconcile;
- the learner has previously made errors at that handoff.
Skip exhaustive checking when the image is decorative or the relationship is already well established and low-risk. Study time has opportunity cost.
Delayed and independent performance check
After a day, remove one representation and ask the learner to reconstruct the other. Give the graph and ask for the paragraph. Give the paragraph and ask for the graph. Then introduce a deliberately inconsistent alternative and ask the learner to locate the mismatch.
This tests three different capabilities: understanding each representation, translating between them and actively validating agreement. A learner who can do only the first two may still miss contradictions.
For parents and tutors
When reviewing a child’s notes, do not ask only, “Is the sentence correct?” Occasionally point to the diagram and ask, “Does this picture say the same thing?” Then ask the child to prove the match.
If the representations disagree, resist fixing it immediately. Let the learner identify the exact feature that breaks consistency: wrong direction, wrong scale, missing condition, wrong label, changed time period or different object. That diagnostic act is more transferable than simply replacing the note.
What not to conclude
The 2026 experiments do not show that students should distrust all pictures or read every textbook as though it contains errors. Nor do they establish that a particular percentage of contradictions will be missed in every classroom. Detection rates depend on materials, tasks, knowledge and conditions.
The useful conclusion is smaller and stronger: passive comprehension can leave contradictions unresolved, especially when checking must cross representational formats. If consistency matters, build an explicit verification step.
The deeper studying principle
Multimodal learning is powerful because different representations can reveal different structure. The same strength creates a new responsibility: the learner must know when the representations are supposed to converge.
Do not assume a graph confirms the paragraph because they share a page. Do not assume a diagram proves the caption because they use the same labels. Translate, compare and verify. When two representations disagree, the contradiction is not an inconvenience to skip. It is information about where understanding — or the source itself — needs repair.
Research basis and routes
Primary research: Frick, J. & Schüler, A. (2026), “Consistency validation in illustrated texts”, Memory & Cognition. The paper was published online on 25 June 2026.
Continue through the How Studying Works Numbered Series Reading Index or return to the How X Works Hub.