Skip to content

Keep the learner in control of the information flow with on-demand tools and supports

Principle · All principles
Evidence · 22 claims (9 for, 11 mixed, 2 against) · 22 studies (10 causal, 3 quant-synthesis, 2 review, 2 associational, 2 design, 2 theoretical, 1 qualitative), q1–q3 · 3 of 22 report an effect size · 20 claims rest on one study

Conditional relationship

This page owns one decision: whether the helps a learner may need while working through rich material (a glossary or dictionary for the words in a scene, an explanation in the first language, replaying a scene, a model answer or feedback after an attempt, a hint) are put under the learner's control, available at once on request, rather than withheld, pushed by the system on a fixed schedule, or handed out by an instructor who drives the session. For a learner meeting material a little above their level (a second-language scene or text with unknown words, a complex-skill performance), keeping the material whole and the helps one step away is expected to support comprehension of the whole and verbal-association learning of the words looked up (immediate-performance, delayed-retention), better than the same material with no helps, and at least as well as the same helps given on someone else's schedule. The earlier page's source, a 1985 account of interactive videodisc work for language learning, put it as exercises that "provide a holistic experience, while at the same time providing specific tools when needed".

No claim in the wiki compares on-demand, learner-controlled helps with the same helps pushed by the system or by an instructor, in language learning or in video. The claims test neighbouring parts:

  • That the helps themselves can help. Glosses beside a second-language text improved immediate vocabulary recall over no glosses in one randomised study with 135 Taiwanese undergraduates, but did not improve reading comprehension; the general glossing claim gathers 2 distinct studies. Captions on second-language video show large effects on comprehension and vocabulary in one synthesis. These glosses and captions were present, not requested.
  • That control over the help can matter, under a condition. In one laboratory motor-learning experiment, learners who could decide after a trial whether to see their result were more accurate at a 24-hour retention test than learners given the same feedback schedule without choosing; deciding before the trial gave no benefit. Who decides is one variable; when they decide is another.
  • That control in general often does nothing. Across 18 studies of learner control in educational technology the average effect was near zero (g = 0.05); the general learner-control claim gathers 11 distinct studies, and its narrower findings repeatedly report controls that learners rarely use (a stop/play button on an animation; a hint budget spent early or hoarded). Students who never asked for a hint on at least one problem were more likely to wheel-spin.

So the relationship is conditional on whether the learner uses the help, when the decision to use it is taken (after an attempt, at the point of difficulty, appears to matter more than before), what the help gives away (a hint or a gloss, against a full answer), the learner's level relative to the material (in the gloss study the benefit appeared only at the two middle proficiency levels), and the outcome measured (words looked up, against comprehension of the whole).

How it sits beside its neighbours: Segmenting owns whether a continuous stream is cut into pieces and who sets the pauses, and concludes that designed segments, not a pause button, carry that effect; this page takes segmenting as given and is about the content helps offered inside and between segments. Multimedia Learning holds the words-with-pictures model and pacing as one of its moderators. Scaffolding and Fading owns help that a tutor or system raises and lowers contingently; this page is the case where the learner, not the tutor, decides when help comes, and the two combine when the on-demand help is graduated. Self-Regulated Learning owns teaching learners to monitor and to seek help; the prompts and the help-use plan below borrow from it. Autonomy owns choice as a motivational condition; the choice here is a cognitive one, about information, and the evidence for it is different.

Default design, while the relationship is untested

The earlier page's guidance, kept as a concrete default a designer can act on and revise. Each step is a proposal; its evidence status is stated beside it. Durations are untested proposals for a 40–50 minute session built around a 2–4 minute video scene or a 300–600 word text.

  1. Meet the whole first (about 5 minutes). Play the scene through, or have the text read through once, without stopping and with the helps closed, so the learner has the context and its clues before looking anything up. Untested proposal; from the earlier page's "holistic experience". If the scene runs longer than about four minutes or explains interacting steps, cut it into segments at natural breaks first, as Segmenting describes.
  2. Open the helps for a second pass (15–20 minutes). The learner replays, backs up and looks up freely: a glossary of the words that occur in this scene with their meanings, captions in the language of the scene, a short explanation of cultural or grammatical points in the learner's first language, and a way to repeat any line. Each help is one keystroke or tap away and does not take the learner out of the scene. Evidence that glosses and captions help vocabulary when present: Earlier meta-analyses, as reported second-hand, find that glosses improve L2 vocabulary learning and that text-and-picture glosses outperform single-mode glosses [+M], All three L1 gloss conditions significantly improve immediate vocabulary recall over a no-gloss control in EFL reading [+M], Captioning second-language videos produces large benefits for comprehension and vocabulary acquisition [+M]. That putting them on request beats always-on or absent helps is untested here.
  3. Let the learner attempt, then decide whether to see the model (10 minutes). At each point where the learner must respond (answer a character, choose a branch, say the line), the learner responds first and then chooses whether to see or hear the model response or feedback. Carried from motor learning, untested for language: Learners who could decide after a trial whether to receive knowledge of results (Self-After, Self-Both) were more accurate on a 24-h motor retention test than learners who decided before the trial (Self-Before) and than their yoked counterparts. [+M]; a choice made before the attempt gave no benefit (Having control over the knowledge-of-results decision only before a motor trial (Self-Before) produced no statistically significant retention or transfer benefit over a yoked group receiving the same KR schedule. [~M]).
  4. Make a help a hint before it is an answer. Where a help can solve the task for the learner (a translation of the whole line, a full model answer, an AI assistant), offer a smaller step first (the key word, the first half of the reply) and the full answer only on a second request. Evidence about unrestricted answer-giving help, second-hand: Unrestricted GPT-4 access improved practice performance but harmed later exam performance, and a guardrailed hint-based version mitigated the harm, as reported by the whitepaper citing Bastani et al. [~W]; learners asked for this themselves in one design study (K-12 learners preferred incremental hints and multiple attempts over immediately revealed solutions [+W]).
  5. Watch for learners who use no help at all, and prompt them. If a learner has opened no help in a scene and then answers wrongly, prompt once ("the glossary has this word; replay line 4"). Do not force the help. Basis: Not requesting any hints in at least one problem is related to greater wheel-spinning, consistent with help avoidance [~M] (correlational, maths practice software), and controls are often left unused (Learner flow control over animations is rarely used, and the authors report no correlation between its use and knowledge-test results although the printed Spearman r is 0.464 [-W]).
  6. Give the learner a plan for help, not a quota. Tell learners when help is most worth taking (after a first try, on the line they could not follow, on a word that recurs), rather than rationing it. Basis: The Coach's hint budget was managed poorly: students split into early-spenders and savers, and within the Coach condition only those with a deliberate hint strategy scored higher [~W]; the plan is an untested proposal.
  7. Do not let the instructor drive. In class, the teacher does not stop the scene to explain every word or decide when learners may look things up; the teacher circulates, answers what the helps cannot, and runs the debrief. Untested; from the earlier page's warning that instructors who drive without tools "misuse the medium and reduce lessons to drill".
  8. End with an unaided pass (5–10 minutes). Replay the scene, or a new scene with the same words, with helps closed; the learner answers comprehension questions in the brief's form or takes their part in the exchange. This is the observation that says what the learner can do without the help; assisted success during step 2 does not. Untested proposal.
  9. Across sessions, move the helps from first language to the target language, and from first pass to second pass. A default arc for a unit of six to ten sessions: in the first two or three, glossary and explanations in the first language, captions on during the second pass; in the middle sessions, glossary definitions in the target language, captions on request only; in the last sessions, the first pass and the response points with all helps closed and the helps open only on the second pass. Move a learner to the next scene when they follow it on the unaided pass and respond correctly at most response points (set the number locally, for example four of five) with no more than a handful of look-ups on the second pass; return to the previous step of the arc if they cannot follow the unaided pass. The arc and its thresholds are untested proposals.

When a condition is not met: if learners have no device to control (a single projected screen), give each learner a printed glossary of the scene and a turn at calling for a replay; if the material is well within the learners' level, drop the helps and spend the time on the response points, since the gloss study found no benefit at its highest proficiency level (Proficiency moderates gloss effects: vocabulary benefits appear at Levels 2 and 3 but not Levels 1 and 4 [~M]).

Fitting the design to a situation

The three facts that most change the decision:

  • How far above the learner the material sits. Helps carry most when the learner can follow the gist but not every word; a learner who follows nothing needs easier material, and one who follows everything needs no helps. If unstated, ask: on one play-through without help, how much of the scene can the learner retell, and how many words in a sample paragraph are unknown to them?
  • Whether learners will use the help, and when. On-demand help works only if it is taken, at the point of difficulty, and not as a way round the task. If unstated, ask: in a trial scene, which helps did learners open, how often, and before or after trying?
  • What counts as success: the words, the whole, or the exchange. Glosses help the words looked up more reliably than comprehension of the whole text. If unstated, ask: will the learner be checked on vocabulary, on following the scene, or on taking part in a conversation like it, and with or without helps?
If the brief says… Then change… Basis
Learners are novice in the language, following little of the scene even with helps Use shorter scenes (under two minutes) or easier material; open the first-language glossary and captions from the first pass; keep step 8 to the gist (who, where, what happened) Proficiency moderates gloss effects: vocabulary benefits appear at Levels 2 and 3 but not Levels 1 and 4 [~M] (EFL undergraduates; no benefit at the lowest level)
Learners are advanced, following the scene without help Drop the glossary pass; keep replay and the response points; give helps only for idioms and cultural references Proficiency moderates gloss effects: vocabulary benefits appear at Levels 2 and 3 but not Levels 1 and 4 [~M] (no benefit at the highest level)
Learners are child (primary age) Fewer helps (a picture glossary and replay), large on-screen buttons, and an adult or peer nearby to prompt help use; shorter sessions (20–30 minutes) untested proposal; the pacing experiment with primary pupils (Letting learners pace instruction helped on complex material in one experiment, while learner control in general showed almost no effect across 18 studies [~S]) tested a pause control, not content helps
The goal is vocabulary (verbal-association) Make the glossary the centre of step 2: text and picture for each word, and recycle the looked-up words in the next scene Earlier meta-analyses, as reported second-hand, find that glosses improve L2 vocabulary learning and that text-and-picture glosses outperform single-mode glosses [+M] (second-hand report of syntheses); Marginal glosses, dictionary use, and repeated occurrence of unknown words each positively affect incidental vocabulary learning [+M] (advanced university learners of French)
The goal is comprehension of the whole, or taking part in an exchange (complex-skill) Do not expect the glossary to carry it; spend the time on the whole-first pass, captions, and the response points with model answers after each attempt L1 glosses do not significantly improve EFL reading comprehension compared with no glosses [~M]; Captioning second-language videos produces large benefits for comprehension and vocabulary acquisition [+M] (captions always on, not requested)
online-self-paced, no teacher present Build steps 5 and 6 into the software: a prompt after a wrong answer with no help opened, and a one-screen guide to when help is worth taking; log help use for the learner to see Not requesting any hints in at least one problem is related to greater wheel-spinning, consistent with help avoidance [~M] (maths practice software, correlational; carried to language video)
classroom with one teacher and many learners, one device each Learners work in headphones on their own device for step 2; the teacher circulates to those who have opened no help, then runs a five-minute whole-class check on the scene's two or three hardest lines untested proposal
Mixed modes: some learners remote, some in the room, or phone-only Put every help inside the player so it works on a phone screen (tap a word for its gloss, a replay-line button); run the response points and the unaided pass the same way for both groups; do not rely on a teacher stopping a projected screen untested proposal
One projector, no learner devices Give a printed glossary of the scene; let learners in turn call for a replay of a line; the teacher stops only when asked, not to explain untested; from the earlier page's warning against instructor-driven use
A single session only Run steps 1, 2, 3 and 8 once on one scene (5 + 20 + 10 + 5 minutes); do not claim retention untested proposal
A fixed short time budget (for example 20 minutes) One short scene: whole pass 3 minutes, helps open 10, two response points with model answers 4, unaided replay 3 untested proposal
Adults in work settings (a business trip, a customer desk, a clinical handover) Use scenes from the learner's own work; glossary entries for the phrases they will need to say, not only to understand; response points that are the exchanges they will have untested proposal; the earlier page's product example (a businessman arriving in the United States) is this setting
Learners with low confidence or low literacy Spoken glosses and replay before written ones; helps that cannot be seen by others; praise taking help as a sensible move untested proposal; Students and teachers respond favorably to embedded aids: approximately 90% of students reported no difficulty understanding or using the aids, and over 75% wanted similar aids throughout the book. [+W] reports acceptability only
The help is an AI assistant or any tool that can give the whole answer Set it to give hints first and the answer only on a second request; keep the unaided pass Unrestricted GPT-4 access improved practice performance but harmed later exam performance, and a guardrailed hint-based version mitigated the harm, as reported by the whitepaper citing Bastani et al. [~W] (second-hand, mathematics, not language)
Learners rewind and re-look-up repeatedly without progressing Treat it as a signal of difficulty, not of good use: offer an easier segment or a short explanation, rather than more replay Video-watching sequences involving pausing and rewinding are negatively correlated with math task performance in a flipped precalculus lesson [~W] (correlational; the authors do not suggest discouraging rewinding)
Stakes are high, or the test is taken without helps Judge readiness only on the unaided pass; keep the helps for practice Unrestricted GPT-4 access improved practice performance but harmed later exam performance, and a guardrailed hint-based version mitigated the harm, as reported by the whitepaper citing Bastani et al. [~W] (performance with help did not carry to the exam without it)
Scarce facilitator time (volunteers, one tutor checking in briefly) Put the prompts of step 5 and the help guide of step 6 in the material; the tutor's few minutes go to the unaided pass and the response points untested proposal
The brief names its own routine, glossary, rubric or assessment Keep it; change only what the evidence speaks to (helps on request after an attempt, hints before answers, one unaided check). If its rubric rewards tool use, keep that criterion and add an unaided item beside it untested proposal
The brief uses its own categories for the vocabulary or tasks (a job's phrase list, a course's topic list) Build the glossary and response points in those categories, not in categories of your own untested proposal
Help requests or the material are private or sensitive Keep help use private to the learner; do not display who looked up what; let the learner choose whether the log is shared untested proposal
The brief wants a whole performance kept whole (a full conversation, a full scene) Keep the scene and the exchange whole; put helps beside it rather than splitting it into item drills untested; from the earlier page's "holistic experience"
The brief reports an error learners actually made (a misheard word, a wrong reply) Use it as a response point: the learner hears the line, chooses a reply, then compares with the error and the model untested proposal

Ten of the twenty-two rows link a claim. None of them tests on-demand helps against the same helps given otherwise; each carries a neighbouring finding (glosses present against absent, captions always on, help avoidance in maths software, an AI assistant in mathematics) to this setting and says so. The rest are untested proposals.

Observation, state and explanation

What can be observed: which helps were opened, when (before or after an attempt), how often, on which lines or words; the response at each response point; and the result of the unaided pass. "Understands the scene", "is avoiding help" and "is leaning on the help" are inferred states. Success with the helps open shows what the learner can do with them, not without them.

Response Plausible explanations Observation that could change the interpretation
Opens no help and answers wrongly Does not know the help is there; avoids help (pride, being seen); misjudges own understanding Ask the learner to point to the help; offer it privately; ask for a confidence rating before the answer is shown.
Opens every gloss, answers well with helps, fails the unaided pass The help carried the task; words looked up were not retained; the unaided pass is harder than intended Re-test the looked-up words alone; give an unaided pass on the same scene the next day.
Rewinds the same line many times without progress The line is above the learner's level; a sound or word they cannot segment; distraction Show the line's caption once; ask what they hear; offer an easier version of the line.
Looks up words, follows the words, still misses the point of the scene Vocabulary was not the barrier; the cultural or pragmatic point was; comprehension needs the whole Ask what the character wanted; offer the first-language explanation of the situation rather than more glosses.
Spends the hint or help allowance on the first items No plan for help use; first items genuinely hardest Compare help use across items against item difficulty; teach the help plan and observe the next scene.

These rows are proposals, untested with learners as diagnostics. Asking about help use changes help use, so record each probe.

Evidence and qualifications

No claim in the wiki compares learner-controlled, on-demand helps with the same helps given by the system or an instructor. The claims below test parts of the model.

Learners, materials and outcomes differ across these claims (undergraduates reading EFL texts, primary pupils watching an animation, volunteers moving a slider, secondary and college students in maths and programming software), so do not rank them by effect label. None measures delayed comprehension of second-language video under on-demand against pushed helps.

The evidence by question

Further evidence, not yet read against this model

Every claim the earlier page cited is used in the model above (the annotated videodisc study, learner flow control, and learner pacing). These further claims were read for this rewrite and bear on the page only at its edge:

Objective and learner-valued goal

The designer's objective is usually comprehension of material like the scene, the words it carries, or taking part in an exchange like it, unaided, at a stated horizon. Elicit separately what the learner wants: to get through an immigration desk and a hotel check-in on a business trip, to follow a television series, to pass a course test, to stop feeling lost in meetings. Agreement and divergence are both observations. A learner who values finishing the scene quickly may open every help and retain little; one who values not looking slow may refuse help they need; one who wants to talk may skip the glossary and go straight to the response points. Help use reflects the goal as well as the difficulty.

For example, a business traveller (the earlier page's product example) may want to handle the questions at a customs desk, while the designer's objective is the scene's vocabulary. Glossary work serves the objective; an unaided run of the customs exchange tests the learner's goal. Record both, so success on a vocabulary check is not read as the capability the learner came for.

What would revise this model?

The expectation should weaken if comparisons of the same helps offered on request against the same helps pushed by the system or the teacher, with comparable learners and an unaided outcome, show no advantage for on request, or show that learners who choose their help retain less at a delay. If such comparisons repeat the motor-learning result that only a choice made after an attempt helps, step 3 becomes the core of the model. If learners at the target level rarely open the helps, availability alone is not the design and the prompts in steps 5 and 6 carry it. If glosses help vocabulary but not comprehension in video as they did in reading, the model should claim the words, not the whole. Do not protect the model by calling every failure "the learner did not use the tools properly" after the fact.

Use of a help, assisted performance, unaided immediate performance, delayed retention and use in the setting the learner values are separate claims. The present evidence establishes no number of helps, no order of helps, and no threshold for moving on.

Claims

  • Segmenting — cutting a continuous scene into pieces and deciding who sets the pauses; this page assumes it and adds the content helps.
  • Multimedia Learning — the words-with-pictures model, with pacing as a moderator.
  • Scaffolding and Fading — help raised and lowered by a tutor or system; graduated on-demand helps borrow its ladder.
  • Self-Regulated Learning — monitoring and help-seeking, which on-demand helps depend on.
  • Autonomy — choice as a motivational condition, a different reason for learner control.

Examples

Key Sources

  • Gale, Larrie E.; Brown, Bruce L. (1985). A Theory of Learning and Skill-Acquisition Applied to Interactive Video: Activities at the David O. McKay Institute, Brigham Young University. In Gillespie, Junetta B., Ed. Video and Second Language Learning. Urbana, Language Learning Laboratory, University of Illinois at Urbana-Champaign. https://eric.ed.gov/?id=ED278252
Page metadata
Field Value
type principle
id learner-control-information-flow-with-on-demand-tools
title Keep the learner in control of the information flow with on-demand tools and supports
description For a learner working through second-language video or text a little above their level, keeping the material whole and putting helps (glossary, replay, first-language explanation, model answer after an attempt, hints before answers) one step away under the learner's control is expected to support comprehension and vocabulary at least as well as withholding or pushing the same helps; no claim here tests on-demand against pushed helps, and learner control in general averages near zero.
status review
canonical True
generated by claude/unspecified on 2026-10-15
trust unverified — no one has confirmed this page's content
sources 1 — gale-1985

Frontmatter is the page's machine-readable record — see the Schema & Guide.