A Science video looks polished. The teacher speaks clearly, subtitles appear at the bottom, and the viewer can pause whenever needed. Yet halfway through, a diagram appears without narration. A green arrow moves from one chamber to another while the teacher says, “Notice what happens here.” A student reading the captions sees only those words. A student who cannot see the diagram receives no account of the direction or the relationship. The video’s most important explanatory moment may be missing from the accessible information stream.
The common assumption is that captions solve every access problem in video. They do not. Captions convey speech and relevant sound information. They are valuable for deaf and hard-of-hearing viewers, for noisy environments and for many learners who benefit from reading along. But if the essential information is visual and never spoken, accurately captioning the audio cannot supply that missing fact. The caption “Now this moves across” is faithful to the speaker and still inadequate as a standalone explanation.
This article concerns the visual-only information gap in instructional video. It asks how a teacher can recognise diagrams whose meaning is not available through speech, how to describe those visuals accurately and how to provide a meaningful alternative without drowning the lesson in narration. It does not own automatic captions that mishear technical vocabulary, translated subtitles, captions as a general accessibility category or the broader interpretation of scientific models.
The 50-second answer
Captions usually represent what can be heard: spoken words, meaningful non-speech sounds and sometimes speaker identification. They do not automatically describe an unspoken diagram, chart, on-screen label or moving arrow. A learner who depends on audio or a text alternative may therefore miss information the video shows but never says.
The repair is to identify every essential visual fact and make it available in another appropriate mode. In many new educational videos, the simplest method is integrated description: the teacher says, “The arrow moves from the higher-pressure chamber on the left to the lower-pressure chamber on the right,” rather than “It goes over there.” For existing videos, a separate audio-description track, described version or descriptive transcript may be appropriate depending on the medium, viewer needs and applicable standard. Captions remain necessary for audio information; visual description complements them.
First, distinguish captions, subtitles, transcripts and description
Every media support method has a job. Captions give a text counterpart to speech and important sounds, generally synchronised with the video. Subtitles may emphasise spoken dialogue, often for translation, though usage differs among platforms. A basic transcript presents the audio information in a reading format. A descriptive transcript adds the significant visual events and text that matter to understanding. Audio description makes those visual facts available through speech or another synchronised accessible channel.
The distinction is not academic bureaucracy. Imagine a video of a graph with a rising blue line. The speaker says only “Look at the trend.” The captions accurately display “Look at the trend.” A basic transcript repeats that sentence. Neither tells a reader unable to see the graph whether the line rises, what its axes measure or what the teacher wants pupils to infer. A descriptive transcript could write: “The graph plots time along the horizontal axis and temperature on the vertical axis; the blue line rises steadily from the first to the final measurement.” That is new information, not a better spelling of the spoken words.
The W3C Web Accessibility Initiative explains these roles explicitly in Making Audio and Video Media Accessible. Its separate guidance on description of visual information says that relevant charts, graphs, text and actions need to be conveyed to people who cannot adequately see the video. We use those official sources as the accessibility foundation; exact compliance obligations depend on the media format, jurisdiction and standard version.
Mechanism one: hearing and seeing deliver different evidence
A diagram can encode relationships through position, colour, direction, size, proximity, shape, labels and motion. Some of those relationships can be communicated through speech, but they do not automatically become audible when the picture appears. If a teacher says, “Here is the answer,” the sentence conveys a social event—the answer is being shown—but not the answer itself.
Consider a water-cycle animation. A blue arrow rises from a lake toward a cloud, then another arrow points downward toward the ground. The narrator says, “This happens and then that happens.” Accurate captions reproduce the words but not the evaporation and precipitation stages. A learner may have to guess which movement corresponds to which process. Even a sighted child on a small phone may miss a tiny arrow or legend.
The repair is to name the relationships. “Water evaporates from the lake into the air; later, condensed water falls as precipitation” supplies conceptual information. It should be checked for scientific accuracy and made specific enough to match the animation. If the video intentionally simplifies the cycle, describe the model’s limits rather than implying that every real atmospheric path follows exactly that one cartoon sequence.
The principle extends far beyond Science. In Mathematics, an animation may reveal a newly constructed line; in Geography, shading may spread across a map; in English, a sentence may be rewritten while the teacher discusses the new meaning. If the change is essential and not narrated, captions of the teacher’s speech alone leave a gap.
Mechanism two: deictic language points without describing
Words such as “this,” “that,” “here,” “there,” “the upper one” and “like this” often work beautifully in a classroom where everyone can see the same board. In a video, their meaning depends on a visible gesture or object. An automatic caption system cannot legitimately invent which arrow or variable the speaker meant. It should transcribe the utterance accurately, not guess the hidden referent.
This is why a teacher who speaks excellent English can still produce an inaccessible explanation. The problem is not pronunciation. It is reference without identification. “This increases” may refer to pressure, temperature or surface area. “Move it to the other side” may refer to a term in an equation or a piece of apparatus. Without the diagram, the phrase may have no recoverable object.
A useful script revision identifies the object in the spoken sentence: “As temperature rises, the measured evaporation rate increases,” or “Move the constant term from the left-hand side to the right-hand side by subtracting it from both sides.” The spoken words become clearer for everyone while the captions gain meaningful content automatically. This is called integrated description because the information is woven into the main explanation instead of added as a separate commentary track.
W3C specifically notes that integrated description often suits presentations and training material when it is planned in advance. Its guidance on creating accessible audio and video provides examples where an ambiguous phrase is replaced with a factual verbal description. Our examples are independently written for classroom subjects, not copied from those examples.
Mechanism three: on-screen text is also visual information
Teachers frequently show definitions, variable names, source citations or instructions in the corner of a video without speaking them. A caption file generated from the audio can be perfect and still omit all those textual details. This matters when on-screen wording provides conditions for an experiment, the scale of a graph or the precise wording of a conclusion.
For example, a video says, “Use the equation at the top,” while the screen displays a formula and a note that the calculation assumes constant speed. If the formula and condition are not spoken, a listener has no way to reconstruct the method from the captions. A basic transcript that repeats only “Use the equation at the top” has the same defect. The solution may be to read the formula and its conditions aloud, provide a properly accessible text copy near the video, or supply descriptive text at the relevant time.
Not every incidental title or decorative label requires a long speech. The standard educational question is whether the text is needed to understand the lesson. A school logo that appears briefly may not matter to the specific learning objective; a red warning that a demonstration is not safe to repeat at home may matter greatly. Teachers should make this determination thoughtfully rather than reciting every visual pixel.
Mechanism four: the timing of description must match the event
Visual description is not simply a paragraph added somewhere at the end. If a learner hears the direction of an arrow thirty seconds after the teacher has moved to a different graph, the description may be too late to make sense. A described audio track must align with the meaningful visual changes and avoid masking essential speech.
When planning a new video, script the important visual facts into the teacher’s main narration. That often produces the simplest synchronisation because the explanation and the visual were designed together. If a separate description is needed, reserve spaces in the narration or create a version that allows descriptions to be heard. Some media players support separate description tracks or timed text descriptions, but support varies. A descriptive transcript can complement the video and give readers a stable independent reference.
The W3C description guidance describes integrated, alternative-video, separate-audio and text-file approaches, each with technical conditions. It also explains that some prerecorded content needs audio description under WCAG Level AA, while Level A and other cases have different alternatives and details. Do not flatten these distinctions into a claim that every video in every school legally requires exactly one method.
Mechanism five: too much description can also harm clarity
A well-intentioned teacher might respond by narrating everything: “There is a white background, a blue rectangle, a green rectangle, a small border, a title, three icons…” The learner can lose the relevant relationship inside a flood of decoration. Description needs prioritisation. What visual facts explain the task, the model, the evidence or the next action? What can safely be omitted because it is already spoken or is merely ornamental?
Think of a graph. The essential facts might be the variables on both axes, units, the range of values and the main pattern. The precise shade of the gridlines may be irrelevant. An explanation that spends ten seconds describing the frame and fails to name the trend has preserved appearance rather than meaning. In a chemistry animation, the number and position of particles could matter; in a decorative title sequence, the colour of a moving dot may not.
Description should be factual and not smuggle in an interpretation that the picture cannot support. “The line rises from 10 to 15” is a report of a displayed trend. “This proves the intervention works” is a causal conclusion requiring more evidence. Keep observation, inference and judgement separate, just as a thoughtful science teacher would in a laboratory report.
Mechanism six: a transcript is useful only if it contains the missing modality
Teachers sometimes solve an accessibility complaint by posting a transcript of the narration. That can be useful for many students, including those who prefer reading, but a basic transcript may repeat the same missing references as the captions. If the speaker says “look at the right-hand shape,” the text says exactly that. The missing shape remains missing.
A descriptive transcript should include concise descriptions of necessary visual changes in their proper order. For example: “At 00:21, a second line appears above the first graph; both lines begin together, but the new line rises faster.” Then the teacher’s explanatory dialogue continues. The description allows someone using text, assistive technology or a screen reader to follow the significant information without seeing the original pixels.
For some audiences, accessible diagrams, data tables or text equations are a better complement than prose alone. A complex graph may require a data table or structured description. A geometrical proof may need a labelled diagram plus a verbal explanation that identifies relationships in a way assistive technology can use. The objective is not perfect sensory imitation but meaningful access to the subject matter.
The W3C’s transcript guidance explains why descriptive transcripts cover a wider range of access needs than audio-only transcripts. Teachers should test whether the chosen alternative is actually usable rather than assuming that posting more text automatically resolves all barriers.
A worked Mathematics example: the arrow that explains the operation
A video shows 3x + 5 = 20. The teacher draws an arrow from +5 to the opposite side while saying, “This goes over there.” A learner might mimic a dangerous shortcut: terms mysteriously cross an equals sign and change signs. The actual algebraic operation is subtracting 5 from both sides of an equation, preserving equality. If the video only shows the arrow and captions the spoken phrase, the reasoning is unavailable to someone who cannot see the visual cue—and may be misleading even to viewers who can.
A stronger narration says, “Subtract 5 from each side, giving 3x = 15. Then divide both sides by 3, so x = 5.” A visual can still support that explanation. Captions now convey the mathematics because the spoken reasoning is complete. A descriptive transcript can also write the sequence of equations explicitly.
A student should then solve 4x + 7 = 31 without being shown the same arrows. If the child can explain why equivalent operations preserve equality, the lesson has achieved more than recognition of an animated trick.
A worked Science example: an arrow with two possible meanings
In a Science model, an arrow can represent physical movement, energy transfer, a force, a comparison or a causal relationship. A video displays an arrow between two shapes while a narrator says, “This explains the change.” Without a verbal account of what the arrow represents, a listener cannot know whether something travelled, was converted or merely influenced another variable.
The repair must explain both the referent and the arrow’s function: “The arrow indicates energy transferred from the warmer body to the cooler body,” if that accurately describes the diagram. If the arrow is actually a force vector, its direction and magnitude convention need a different explanation. Do not read every arrow as motion. The existing guide on arrows in Science diagrams explores that subject-specific interpretation; the present article concerns access when such interpretations are shown but never said.
A good script can deliberately identify model limits: “The arrow shows the direction of net transfer in this simplified model; it is not a literal picture of an individual molecule’s path.” That single sentence helps a blind listener and a sighted learner who might otherwise overread the picture.
A worked Geography example: a map whose colours change silently
A Geography video animates land use over successive years. The narrator says, “Watch the area spread across the island.” If the video captions only that sentence, a text-only reader has no idea which area, what years or how much the mapped category changes. A useful description names the category, date sequence, legend and observed pattern without claiming more precision than the map provides.
For instance: “The maps compare 2010 and 2020. The category marked as built-up land covers more of the eastern region in the second map; the two panels use the same legend and scale.” If the scale changes, the comparison must be qualified. If the animated area represents a forecast rather than observed data, the narration should say that explicitly. Visual description is not merely accessibility decoration; it is also a discipline of honest evidence communication.
A descriptive table of map categories may be suitable when precise values matter. A line such as “green areas shrink” may be too vague if the legend distinguishes several kinds of vegetation. Careful naming is as important as technical caption quality.
Evidence and sources
The accessibility mechanism is directly supported by the W3C Web Accessibility Initiative’s media guide, its description of visual information, its planning guide and transcript guidance. For wider instructional design, CAST’s Universal Design for Learning resources identify multiple means of representation among the UDL principles. UDL is a framework, not evidence that any specific narration automatically improves examination scores.
These sources justify making critical visual information available; they do not prove that all silent diagrams are inaccessible or that all videos need equally extensive description. A video with no essential visual-only information may not require an additional description track, and different compliance levels have different requirements. The How X Works hub links this problem to wider explanations of learning systems.
How a learner can find the missing explanation
If a video says “as you can see,” “this one,” “watch what happens” or “the arrow here,” pause and ask whether the spoken words actually identify the object or relationship. If they do not, the captions may be faithfully reporting an incomplete explanation. That distinction matters. There may be nothing wrong with the transcription service; it cannot accurately transcribe facts that nobody spoke.
Record the timestamp and formulate a precise question. Instead of “I cannot understand the video,” write, “At 2:14, which quantity does the arrow represent, and which way is it pointing?” A teacher can answer that question quickly, and the response can improve the resource for the next student. If an accessible transcript, labelled diagram or companion notes are already available, use them to reconstruct the relationship.
When a diagram is central to a calculation, ask for the variables, units and relative positions. When it shows a process, ask for the starting condition, the ordered changes and the endpoint. When a teacher edits a sentence on screen, ask which words changed and why. Do not fill gaps by inventing what an unseen chart probably shows. In Science and Mathematics, plausible guesses can produce confident but false conclusions.
After receiving the description, test your understanding on a different example. If the video shows a left-to-right energy-transfer arrow, try explaining the direction of transfer in a new scenario with different objects. If it shows a map changing over time, describe a different comparable pair of maps. The objective is independent explanation of the relationship, not memorising one narration.
How parents can support access without assuming a learning deficit
A parent may notice that a child reads the subtitles repeatedly and still cannot answer the exercise. The natural conclusion is that the child needs to concentrate harder. But if the exercise depends on information displayed silently, reading those subtitles ten times will never supply the missing diagram. Before increasing practice time, inspect whether the necessary information is in the accessible material at all.
Play the clip without looking at the screen and listen for every essential fact. Then read the captions alone. If you cannot name the arrow, its direction or the meaning of the graph, the child may have an information-access problem. This simple household test is not a full accessibility audit, but it can reveal a useful question to take to the teacher.
Be careful not to assume that all children experience the same barrier. Some learners need accurate captions of speech; some need visual description; others benefit from a labelled still image, slower playback, larger text or an accessible data table. A child’s specific needs, the task and the format all matter. Offer access to an appropriate explanation rather than treating one method as the universal answer.
Ask the school for the missing content in a respectful, concrete way. “Could you describe the graph at 3:40, including its axes and direction of change?” is more actionable than “Please make the video better.” A corrected narration, one-paragraph note or descriptive transcript may be enough. Parents need not demand expensive technology when a clear and accurate explanation will solve the educational problem.
How teachers can script accessible explanation from the beginning
Before recording, mark the moments when pupils must see something to understand the next statement. These might include newly introduced labels, arrows, graph axes, equation transformations, maps, demonstration results or changes made to an essay. For each point, draft a sentence that carries the essential fact in speech. The sentence should support learning even when the screen is not visible.
Replace “as you can see, the second line is higher” with a precise statement such as “At the final measurement, the second line represents 18 degrees Celsius while the first represents 12 degrees Celsius.” The exact values must come from the actual chart. Replace “this arrow goes there” with “the arrow points from the source toward the lamp through the closed path,” where that is the intended diagram. Precision benefits everyone, not only pupils using assistive technology.
Keep the narration matched to what is on screen. If a teacher changes the example during editing, check whether the spoken variables and visual labels still agree. A descriptive sentence that refers to an earlier graph can be as misleading as a missing sentence. When numbers are central, verify them against the source dataset or written solution, not memory.
Record a sample scene, then perform an audio-only test. Can a listener identify the major relationship? Ask a colleague who did not prepare the video. Their questions may reveal ambiguous references that the teacher no longer notices because the diagram is so familiar. Only after the content is clear should the captions be checked for technical transcription accuracy.
The two-pass media audit: sound alone and text alone
The first pass is audio-only. Close your eyes or turn the screen away while listening to the lesson. Identify every point where the teacher refers to something unspoken: “this,” “that colour,” “the top row,” “the moving arrow.” Check whether an essential relationship remains inferable. The test is imperfect because an experienced teacher may infer missing information using prior knowledge; a novice may not. Ask what a pupil at the target level would need.
The second pass is text-only. Read the captions or basic transcript without watching the video. This detects omissions in how speech and sound are represented, but it also reveals how much the explanation relies on visuals. If the student cannot know the plotted variables from the text, a descriptive alternative is needed. If the captions themselves mishear a term, that is an additional problem with its own repair route.
Then review the actual video with all modalities present. Do the verbal statements accurately describe the diagrams? Are the descriptions timed with the right event? Are important safety warnings conveyed? Does the accessible transcript make the same scientifically justified claims? A technically accessible sentence can still be false; accessibility and content accuracy must be verified together.
The separate eduKateSG guide Why Automatic Captions Can Change a Science Explanation owns transcription mistakes. This page owns a different failure: information not present in the spoken track at all. A complete quality check needs both.
Case study: a bar chart with missing axis meanings
A video displays a bar chart and the teacher says, “The blue bar is twice the orange one.” The captions reproduce the sentence correctly. Yet the bars may show different experiments, years or student groups. Without the axis labels and units, the comparison cannot be interpreted responsibly. Twice what? Measured in which units? Are the bars actually directly comparable?
A useful description states the plotted variable, the compared categories, the units and the relevant displayed values. If the chart reports 40 minutes and 20 minutes, say so. If it uses a truncated axis or different denominators, flag the limitation. Avoid inventing numbers where only a qualitative height difference can be seen. Visual description is an act of faithful reporting, not licence to exaggerate a graph’s meaning.
The teacher can also provide a small accessible data table. For many quantitative graphs, that is a better companion than a long verbal tour through every bar. A learner can then evaluate whether the ratio claim follows from the data and whether other information is needed.
Case study: a geometry diagram uses colour as an unspoken clue
A teacher displays two blue angles and two red angles, explaining that “these match.” Captions accurately reproduce those words, but the blue and red relationships are not named. A reader relying on audio or text has no way to identify which angle pair the teacher means. A pupil watching a grayscale copy may experience a similar difficulty.
A better explanation identifies the vertices and relationships explicitly: “Angle ABC is equal to angle DEF because the corresponding lines are parallel,” if that is the justified relationship in the particular diagram. The visual can retain colour as a helpful secondary cue. The words should not depend on colour alone to establish the mathematics.
A description must also distinguish a given relationship from one inferred in the proof. If the teacher says two angles are equal, explain whether equality is marked as given or follows from a geometric theorem. This protects the logic of the lesson as well as access to the diagram.
Case study: a Science simulation changes one variable silently
An interactive demonstration increases temperature while other quantities remain constant. The narrator says, “Notice how the graph changes.” A learner who cannot see the control slider cannot know which variable was adjusted. Even a sighted learner might misread the graph if the on-screen control label is small or changes too quickly.
The accessible explanation should state the original setting, the changed setting, what remained fixed and the observed response. “We raise temperature from X to Y while holding the other controls fixed; the plotted rate rises over these displayed measurements,” with verified values, conveys the essential comparison. It should also say whether the model is a simulation rather than direct laboratory evidence.
A text alternative can record the same before-and-after values and the model’s assumptions. When a claim needs experimental evidence, do not treat the animation alone as empirical proof. A simulated result is generated by a model whose assumptions should be available for inspection.
Case study: a historical timeline appears without date narration
A History video scrolls through events on a timeline, while the speaker says, “That came much later.” Captions are accurate, but the student relying on text does not know which event, date or interval “that” describes. This is a reference problem and a chronology problem. The relevant information may be written on the screen and never spoken.
The remedy identifies the events by name and date and gives the relationship between them. If the timeline uses approximate dates, the spoken account should preserve that uncertainty. “Around the late nineteenth century” should not silently become “in 1890.” Historical sources differ in precision, and responsible description should not manufacture certainty to make the voiceover sound crisp.
Teachers can add a chronological table alongside the video. A learner can then reconstruct the order and ask what evidence supports the intervals. The lesson becomes less dependent on animated motion and more resilient across devices and learning modes.
Case study: a teacher’s handwriting contains the missing formula
A lesson video shows a teacher writing an equation step by step while saying, “This is the formula you need.” Captions convey the sentence, not the handwritten notation. Even when the equation is obvious to a trained viewer, it may be unreadable on a small phone. The student needs the actual expression and its variable definitions in an accessible form.
The simplest improvement may be to speak the essential equation and provide it in a properly formatted text or mathematical representation. If an equation is complex, a dedicated accessible notation format or companion description may be better than an awkward spoken stream. Teachers should avoid relying on a screenshot alone for a formula central to the task.
The student should then use the equation on a fresh problem, defining variables and units correctly. Access to the equation is necessary but not sufficient for understanding; the teacher must still explain when and why the equation applies.
Failure modes when adding description
One failure is using automated image recognition to describe a scientific diagram without expert checking. A general model might call a flow arrow a motion arrow or overlook a negative sign. Another is describing every decorative detail while omitting the axis labels. A third is providing one long descriptive paragraph far from the time at which the diagram matters. A fourth is changing the narration but neglecting to regenerate or verify the captions.
Other failures include contradictory verbal and visual versions, on-screen text too small to read, explanations that substitute subjective interpretation for observable fact, and descriptive transcripts that merely repeat vague spoken phrases. Repairs should be tested against the actual learning task. Can the viewer identify the variables, the relationship and the next action? Can the student use that information to answer without guessing what was visually omitted?
Do not treat accessibility as a compliance sticker applied after production. It is a design question about which information must be available and through what route. The W3C recommends planning visual description early because doing so can reduce the need for separate corrective tracks in suitable educational material.
Frequently asked questions
Are accurate captions enough for every learning video?
No. They convey the audio information, but visual-only facts such as graph labels, arrow directions and on-screen instructions may still be inaccessible. Those need appropriate description or an equivalent accessible alternative.
Is audio description the same as subtitles?
No. Subtitles and captions represent spoken or audible content in text. Audio description communicates important visual information through an audio route. A descriptive transcript can combine spoken and visual information in text.
Must a teacher describe every object visible on the screen?
No. Prioritise what is necessary to understand the lesson, the task, the evidence and any safety instructions. A flood of decorative detail can obstruct the central explanation. Accuracy, relevance and timing matter.
Can the existing teacher narration be enough?
Yes. If it already conveys all necessary visual information clearly, a separate description track may not be needed for that content. W3C calls an approach that integrates the visual facts into the main speech integrated description.
Is a basic transcript a complete replacement for the video?
Not if it excludes essential visual-only information. A descriptive transcript or other accessible representation must include those facts. Complex graphics may also benefit from accessible tables or structured mathematical notation.
What if the captions are wrong as well?
Correct the transcription issue and the visual-information gap separately. An accurate description will not repair a wrong technical word in the captions, and a corrected caption will not describe an unspoken chart. Both quality checks may be needed.
How should we check whether the improvement worked?
Ask a learner or colleague who does not see the diagram to explain the intended variables and relationship using the available audio or accessible alternative. Then test understanding with a new example. The method is an access and comprehension check, not a guarantee that every student will master the topic.
The lasting principle
A video can contain every scientific fact on the screen and still fail to communicate those facts to a learner who depends on other modes. Captions preserve the information in the sound; visual description preserves necessary information in the image. The strongest educational video plans both, checks them against the actual model or data and gives pupils a fair route from observation to explanation. That is how a silent diagram stops being a hidden prerequisite to understanding the lesson.