04 — Structure, attention and what to leave out¶

Mayer's multimedia principles were developed for narrated animations, i.e. for exactly the "narration + limited or no visual channel" situation we are in. The subset below survives the translation to audio-only.

4.1 Segmenting — [A], median effect ≈ 0.98¶

Break a lesson into learner-paced segments instead of one continuous stream. This is the same finding as the transient information work in 02, arrived at from the other direction, and it is one of the largest effects in the multimedia literature.

Implication. Segment boundaries are structural objects in our script, with an audible marker and a pause, sized by estimated seconds of narration rather than by source paragraphs. In a future player (Part 2) they become the natural stop/resume points. → SEG-*.

4.2 Pre-training — [A]¶

Learners do better when they already know the names and characteristics of the key concepts before the explanation starts. Pre-training offloads part of the essential processing so working memory can spend itself on the mechanism.

Implication. Every document and every major section opens with a short term pre-load: the handful of terms and symbols you must own before the section makes sense, each with a one-line definition, delivered before the exposition rather than in-line. This is also where acronyms and symbol pronunciations get established once. → PRE-*.

4.3 Signaling — [A], effect ≈ 0.52¶

Cues that expose the organization of the material improve learning. In audio there is no bold text, no layout, no figure to point at — signaling must be entirely verbal.

Implication. Explicit verbal signposting is mandatory, not stylistic: enumerate ("there are three reasons; here is the first"), announce structure before content, mark transitions, and mark importance ("this next sentence is the core claim of the paper"). Verbal signposts also serve the re-entry function from §2.3. → SIG-*.

4.4 Coherence — [A], effect ≈ 0.97¶

Removing interesting-but-extraneous material improves learning. This is the largest and most counter-intuitive lever available to us: the single most effective thing mimem can do is delete.

Combined with the seductive-details literature (§3.4), the message is blunt: tangential-but-fun content actively displaces main ideas in memory.

Implication. Aggressive triage of source material. Author affiliations, funding statements, acknowledgements, running heads, reference lists, DOIs, "the remainder of this paper is organized as follows", boilerplate ethics statements, and most hedging clauses are dropped, not narrated. Everything retained must earn its place against the coherence principle. → COH-*.

4.5 Prequestions / pretesting — [A] for the specific effect, with a sharp limit¶

Asking questions before the content is presented improves learning — but the meta-analytic picture is precise about what it improves: g ≈ 0.54–0.66 for the specifically prequestioned content, and essentially zero (g ≈ 0.04) for everything else in the same lesson. Eye-tracking shows why: prequestions redirect attention onto the prequestioned sentences.

Implication. This is a targeting instruction, not a decoration. Prequestions must be generated from the same item pool as the final retrieval items, so that attention is steered onto exactly what we intend the listener to retain. A generic hook ("ever wondered how batteries age?") buys nothing measurable. And because attention is redirected away from non-prequestioned content, the number of prequestions per section must be small and deliberate (2–4). → PRQ-*.

4.6 Redundancy — [B], and why it does not forbid our repetition¶

Mayer's redundancy principle says: don't present identical on-screen text and narration simultaneously. It is about simultaneous duplication across channels, not about repetition over time. Spaced, paraphrased repetition (§1.2, §1.5) is a different mechanism and remains desirable.

Implication. Worth stating explicitly so that neither we nor a future contributor "optimizes away" the repetition machinery by misapplying this principle. The written study.md and the spoken audio track are deliberately not identical; they are two renderings for two uses.