McGurk effect¶

Show that speech perception is intrinsically multimodal: a face articulating /ga/ over an audio /ba/ is heard as /da/, because the auditory and visual streams are fused into a single percept — by reliability-weighted, coherence-gated combination — before the result reaches awareness.

Core Idea¶

The McGurk effect, reported by McGurk and MacDonald in Nature (1976), is a multisensory speech-perception illusion in which conflicting visual and auditory phoneme information is not resolved in favour of either channel but is instead fused into a third percept that corresponds to neither input alone. In the standard demonstration, a viewer sees a face on video articulating the syllable /ga/ while the audio track carries the syllable /ba/; most viewers consciously hear /da/ or /tha/ — a fusion syllable generated by the integration — even though neither input alone carries that syllable. Closing the eyes, which removes the visual channel, restores accurate perception of the audio /ba/, and opening them again makes the illusory fusion percept return immediately.

The structural claim established by the illusion is that speech perception is intrinsically multimodal: the auditory and visual articulatory streams are combined into a single perceptual decision before that decision enters awareness, not by a post-perceptual inference that a listener can override. This pre-attentive integration is demonstrated by cognitive impenetrability — the fused percept is robust to knowing about the illusion, to repeated exposure, and to deliberate instruction to "hear the audio" — and it obeys a reliability-weighting rule: the visual channel dominates the integration more strongly when the auditory channel is degraded by noise and less strongly when audio quality is high, consistent with a weighted-combination process. Integration also requires spatio-temporal coherence: when the audio source is perceptually separated from the face, or when the auditory and visual streams are misaligned by more than roughly 100 milliseconds, integration breaks down and the auditory syllable is heard as delivered. Susceptibility varies across language communities — it is reliably lower in Japanese-speaking populations than in Spanish- or English-speaking populations — suggesting that the weights assigned to articulatory visual cues during integration are calibrated by linguistic experience.

Structural Signature¶

Sig role-phrases:

the multi-channel perceiver — a listener with two simultaneously active sensory channels (auditory + visual articulation)
the conflicting inputs — channel-specific signals carrying incompatible information about one shared speech event (face says /ga/, audio says /ba/)
the pre-attentive integrator — a mechanism that combines the streams into one percept before the result enters awareness
the fused percept — a third syllable (/da/) corresponding to neither input alone
the cognitive impenetrability — the fusion survives knowing the trick, repeated exposure, and the instruction to "hear the audio," so it is integration not overridable inference
the reliability-weighting rule — the visual channel dominates more as the audio is degraded, less as it is clean (a measurable weighting function)
the coherence gate — fusion happens only within a shared-source window; split location or asynchrony beyond ~100 ms breaks it and the audio is heard as delivered
the experience-calibrated weights — susceptibility tracks language community, so the integration weights are tuned by linguistic exposure rather than fixed

What It Is Not¶

Not a cognitive misinterpretation you can correct by knowing the trick. The fusion is cognitively impenetrable: it survives full knowledge of the illusion, repeated exposure, and the explicit instruction to "hear the audio." That impenetrability is what places it on the pre-attentive side of the fork — it is integration completed before awareness, not a judgment a listener could reason their way out of.
Not the eyes merely assisting a straining listener. Vision is not an optional aid layered on top of a finished auditory percept; it is a constituent of the heard percept, combined with the audio before the result reaches consciousness. The eyes-closed/eyes-open toggle, which flips the percept in one second, shows the visual stream is part of what is heard, not a helper to it.
Not one channel overriding the other. The conflict is not resolved by privileging vision or audition; the streams fuse into a third syllable (/da/) that corresponds to neither input alone. Reading it as visual capture of the audio misses that the percept belongs to neither channel — it is a genuine combination.
Not unconditional integration of any two streams. Fusion requires spatio-temporal coherence: it happens only within a window where the inputs plausibly share one source. Split the apparent source, or misalign the streams past roughly 100 milliseconds, and integration breaks down — the auditory syllable is heard as delivered. The effect marks a precise boundary, not a blanket merging.
Not fixed universal wiring. Susceptibility varies across language communities — reliably lower in Japanese-speaking than in Spanish- or English-speaking populations — so the weights assigned to visual articulatory cues are calibrated by linguistic experience, not hard-wired. Treating the effect as a perceptual constant misses that exposure tunes the weighting.

Scope of Application¶

The McGurk effect lives across the subfields that study audiovisual speech — perception science, neuroscience, audiology, and linguistics — wherever a perceiver fuses conflicting auditory and visual articulation pre-attentively; its reach is within that one integration substrate viewed through different instruments, plus a literal applied implication wherever streams plausibly share a source (the sibling cross-modal illusions are co-instances of the multisensory-integration parent, not McGurk).

Perception and cognitive science — the canonical demonstration that perceptual modalities are not encapsulated, and a foundational case for Bayesian-causal-inference accounts of multisensory integration.
Neuroscience — the fused percept is used to localize integration regions (superior temporal sulcus, premotor cortex) via fMRI and lesion studies and to drive audiovisual mismatch-negativity paradigms.
Audiology and clinical communication — bears on cochlear-implant and hearing-aid users and lip-reading training, with cross-linguistic and developmental variation in susceptibility studied as a clinical marker.
Linguistics — cited as evidence for the motor theory of speech perception and against strictly auditory-only models of phoneme recognition.
Speech technology and HCI — because audiovisual misalignment degrades intelligibility (not merely aesthetics), lip-sync tolerances in dubbing, video conferencing, animated characters, and AR/VR are tightened to keep streams within the coherence window.
Cross-cultural psychology — susceptibility tracks language community (lower in Japanese, higher in Spanish/English speakers), making the effect a probe of how linguistic experience calibrates the integration weights.

Clarity¶

Naming the McGurk effect overturns a tacit picture of speech perception in which hearing is auditory and the eyes, at most, assist a listener who is straining to make out words. The illusion makes the opposite legible and undeniable: that the visual articulatory stream is not an optional aid layered on top of a finished auditory percept but a constituent of it, combined with the audio before the result reaches awareness. The decisive distinction the effect sharpens is between pre-attentive integration and post-perceptual inference — between a percept the brain has already fused and a conclusion a listener could reason their way out of. Cognitive impenetrability is what settles it: because the fusion survives knowing the trick, repeated exposure, and the explicit instruction to "hear the audio," the integration cannot be a judgement laid over the sound; it is the sound, as delivered to consciousness. The eyes-closed/eyes-open toggle turns this from argument into a one-second demonstration.

That reframing lets a perception researcher ask sharper, quantitative questions in place of the blunt "do vision and audition interact?" Because integration follows a reliability-weighting rule — the visual channel dominates more when the audio is noisy and less when it is clean — the question becomes how much weight each channel earns and under what conditions, which is a measurable function rather than a yes/no. And because fusion demands spatio-temporal coherence, breaking down when source location is split or the streams drift past roughly 100 milliseconds, the effect marks the precise boundary at which the system stops treating two inputs as one event. The cross-linguistic variation sharpens the further question of where those weights come from: if susceptibility tracks language community, the integration weights are calibrated by experience, not fixed — turning "is multimodal speech perception universal?" into an empirical study of how exposure tunes the weighting.

Manages Complexity¶

The behaviour of audiovisual speech perception, taken case by case, is a thicket of seemingly disconnected facts: that a /ga/ face over /ba/ audio yields heard /da/; that closing the eyes restores the audio and opening them reinstates the illusion; that knowing the trick, repeated exposure, and the instruction to "hear the audio" all fail to dispel it; that the visual channel matters more in noise and less in quiet; that splitting the apparent source or misaligning the streams past roughly 100 milliseconds makes fusion collapse; that Japanese speakers are less susceptible than Spanish or English speakers. The McGurk effect collapses this list to one structure that an analyst can carry instead of memorising the cases: speech perception is multimodal integration into a single percept before awareness, by reliability-weighted combination conditioned on the inputs plausibly sharing one source. Once that is the object, the qualitative outcome of any manipulation follows by tracking a few parameters the structure names rather than re-deriving each result. The reliability of each channel sets the weights and thus the direction of capture — degrade the audio and the visual articulation dominates more, clean it up and it dominates less — turning the blunt "do vision and audition interact?" into a measurable weighting function. Spatio-temporal coherence sets whether integration happens at all — within the source-plausibility window the streams fuse, outside it (split location, asynchrony beyond ~100 ms) the system treats them as two events and the audio is heard as delivered — marking the precise boundary of the "one event" inference. Cognitive penetrability sorts the percept onto the correct side of a load-bearing fork: because the fusion survives knowledge, exposure, and instruction, it is pre-attentive integration, not a post-perceptual judgment a listener could reason out of — which is exactly what the eyes-closed/eyes-open toggle demonstrates in one second. And linguistic experience sets where the weights come from — susceptibility tracking language community means the integration weights are calibrated by exposure rather than fixed, recasting "is multimodal speech perception universal?" as an empirical study of tuning. A high-dimensional "catalogue of audiovisual speech phenomena" reduces to a weighted-combination process with a coherence gate and an experience-set weighting, off which the analyst reads capture direction, breakdown threshold, and the pre- versus post-perceptual status of the percept.

Abstract Reasoning¶

The McGurk effect licenses a locus diagnostic that reasons from the robustness of an illusion to the stage at which two streams are combined. The signature inference distinguishes pre-attentive integration from post-perceptual inference by a cognitive-impenetrability test: because the fused percept survives knowing the trick, repeated exposure, and the explicit instruction to "hear the audio," the analyst reasons that the integration cannot be a judgment laid over the sound that a listener could reason their way out of — it is the sound as delivered to consciousness, combined before the result reaches awareness. The eyes-closed/eyes-open toggle supplies the causal probe that turns this from argument into demonstration: removing the visual channel restores the audio percept and reinstating it returns the fusion immediately, so the analyst reasons that the visual articulatory stream is a constituent of the heard percept, not an optional aid — the toggle manipulating one input and reading the percept flip as evidence of constitutive integration. This overturns the tacit picture in which hearing is auditory and the eyes merely assist, replacing "do vision and audition interact?" with the determinate claim that the visual stream is fused into the percept pre-attentively.

The predictive move reads the direction of capture off a reliability-weighting rule: the analyst predicts the visual channel dominates the integration more when the auditory channel is degraded by noise and less when audio is clean, so degrading the audio is forecast to strengthen visual capture — turning the blunt interaction question into a measurable weighting function of channel reliability. The boundary-drawing move uses the spatio-temporal coherence requirement to fix where integration happens at all: the analyst predicts fusion within the window where the streams plausibly share one source, and predicts breakdown — the auditory syllable heard as delivered — when the apparent source is split or the streams drift past roughly 100 milliseconds, marking the precise threshold at which the system stops treating two inputs as one event. So the analyst reasons from "the audio is suddenly heard correctly despite a conflicting face" to "the coherence gate was crossed — location split or asynchrony beyond the window — so the inputs were registered as two events." The experience-calibration move reasons from cross-linguistic variation about where the weights come from: because susceptibility tracks language community (lower in Japanese-speaking than Spanish- or English-speaking populations), the analyst infers the integration weights are calibrated by linguistic experience rather than fixed, and predicts that exposure history tunes the visual weighting — recasting "is multimodal speech perception universal?" as an empirical study of how experience sets the weights. The applied/interventionist corollary follows from the same constitutive claim: because audiovisual misalignment degrades intelligibility and not merely aesthetics, the analyst predicts that lip-sync error in dubbing, conferencing, or animated characters beyond the coherence window will impair content recognition, and reasons that the design lever is keeping the streams within the source-plausibility window. The boundary-drawing scope is a perceiver with two simultaneously active channels carrying conflicting information about one event and a pre-attentive integrator: the analyst predicts a fused, impenetrable percept where those hold, and reasons that the same weighted-combination-plus-coherence-gate logic governs sibling cross-modal illusions while the speech-specific content (phoneme fusion, articulatory cues, language-tuned weights) is what this effect adds.

Knowledge Transfer¶

Within speech and multimodal-perception research the McGurk effect transfers as mechanism across the subfields that study it, because each is looking at the same pre-attentive, reliability-weighted, coherence-gated integration through a different instrument. In cognitive science it is the canonical demonstration that perceptual modalities are not encapsulated; in neuroscience the same fused percept localizes integration regions (superior temporal sulcus, premotor cortex) and drives audiovisual mismatch paradigms; in audiology it bears directly on cochlear-implant and hearing-aid users and on lip-reading training, where cross-linguistic and developmental variation in susceptibility is read as a clinical marker; in linguistics it is evidence in the motor-theory-of-speech debate and against auditory-only phoneme models. Across all of these the same handles carry intact: the cognitive-impenetrability test that places the percept on the pre-attentive side of the fork, the reliability-weighting rule that predicts visual capture rises as audio degrades, the spatio-temporal coherence window (~100 ms, shared source) that fixes where fusion happens at all, and the experience-calibration reading of cross-linguistic variation. The vocabulary travels with it because the substrate — a perceiver with conflicting simultaneous channels and a pre-attentive integrator — is constant beneath the differing measurement.

Beyond speech, the transfer splits cleanly into three kinds. First, the applied design implication transfers literally wherever its precondition holds, and this is closer to an instrument-reach than a metaphor: because audiovisual misalignment degrades intelligibility and not merely aesthetics, lip-sync error beyond the coherence window impairs content recognition in dubbing, video conferencing, animated characters, and AR/VR — the same weighted-combination-plus-coherence-gate logic, the same source-plausibility window, just a different stream pair. The boundary to watch there is over-reading: it predicts intelligibility loss when streams plausibly share a source and drift past the window, and says nothing about misalignments outside that regime.

Second, within perception the integration mechanism itself genuinely recurs across a sibling family of cross-modal illusions — the ventriloquism effect (visual–auditory integration for spatial location), the rubber-hand and body-transfer illusions (visual–tactile integration for body ownership), the sound-induced flash illusion (audio biasing visual count). These are not analogies to McGurk; they are co-instances of the same pre-attentive, reliability-weighted, coherence-gated combination operating on different objects. So the honest report is that the portable content is the parent — pre-attentive multisensory integration with Bayesian-style reliability weighting and source-attribution gating (multisensory_integration, crossmodal_binding, Bayesian causal inference) — and the cross-illusion lesson should be carried by that parent, not by "McGurk." What stays home is the speech-specific cargo: phoneme fusion, articulatory visual cues, the motor-theory ties, and the language-tuned weights. Those are what make this the McGurk effect rather than its siblings, and they do not transfer even to the ventriloquism effect next door.

Third, looser invocations — "a McGurk effect" for any case where context overrides a signal, or where one input colors the reading of another in non-perceptual judgment — are analogy: they borrow the shape (the percept differs from any single input) while dropping the pre-attentive integrator, the reliability weighting, and the coherence gate that give the original its predictive structure, and should be marked as such. The clean summary: within speech perception it transfers as mechanism; the lip-sync implication transfers literally as an instrument wherever the precondition holds; the integration machinery recurs across sibling illusions and that lesson belongs to the multisensory-integration parent; the speech-specific content stays home. See Structural Core vs. Domain Accent.

Examples¶

Canonical¶

Harry McGurk and John MacDonald reported the effect in Nature in 1976 ("Hearing lips and seeing voices"), having stumbled on it while a dubbing mismatch made a played-back syllable sound wrong. In the canonical demonstration, a viewer watches a video of a face repeatedly articulating /ga/ while the soundtrack, dubbed onto that face, carries /ba/. Most viewers report hearing neither: they consciously perceive /da/ (or /tha/), a fusion syllable present in neither the audio nor the lips. The clincher is a one-second manipulation — close the eyes and the true audio /ba/ is heard cleanly; open them and /da/ instantly returns. The percept is unmoved by knowing the trick, by repetition, or by being told to "just listen to the sound," establishing that the two streams are combined before the result reaches awareness.

Mapped back: The viewer with simultaneous ears and eyes is the multi-channel perceiver; the /ga/ lips over /ba/ audio are the conflicting inputs about one speech event. The brain's pre-attentive integrator merges them into /da/ — the fused percept belonging to neither channel. That it survives full knowledge and the instruction to hear the audio is the cognitive impenetrability placing it before awareness, and the eyes-closed/eyes-open toggle proves vision is a constituent, not an aid.

Applied / In Practice¶

Broadcast and telecommunications engineering builds directly on the coherence requirement. Because audiovisual asynchrony degrades not just aesthetics but speech intelligibility, standards bodies set tight lip-sync tolerances: recommendations such as ITU-R BT.1359 hold the audio-video offset to roughly a hundred milliseconds or less (audio may lead only slightly and lag only modestly) before misalignment becomes objectionable and comprehension suffers. Television production, video-conferencing codecs, film dubbing, and animated-character animation all budget for keeping the voice and the visible articulation within this window, because once the streams drift past it the viewer stops fusing them into a single speech event and intelligibility falls.

Mapped back: The viewer of dubbed or streamed video is the multi-channel perceiver; the soundtrack and the on-screen mouth are the two streams whose fusion the engineering must preserve. The lip-sync tolerance is a direct engineering encoding of the coherence gate — fusion (and the intelligibility gain it brings) holds only within a shared-source temporal window, and asynchrony beyond roughly 100 ms breaks it. That visual articulation contributes most when audio is degraded (noisy channels, hearing loss) is the reliability-weighting rule the same designs exploit by prioritizing clean lip visibility.

Structural Tensions¶

T1: Genuine fusion versus channel override (a third percept or vision capturing audio). The effect's headline claim is that conflicting streams fuse into a third syllable (/da/) belonging to neither input — not that one channel wins. This is what distinguishes McGurk from mere visual dominance and makes it evidence for constitutive integration rather than weighting-to-a-winner. But the phenomenon is not uniform: some stimulus pairs yield a true fusion, others yield a "combination" (both heard in sequence), and still others yield outright visual capture where the audio is simply overwritten. The tension is that the concept's most striking commitment — a percept present in neither input — holds cleanly only for a subset of cases, while the broader family shades continuously from fusion through combination to capture. Insist on fusion and you narrow the effect to its purest examples; admit the whole range and the neat "third percept" story becomes one outcome of a weighted process among several. Diagnostic: Is the reported percept genuinely present in neither input (fusion), or is it one channel overwriting the other (capture) dressed up as integration?

T2: Cognitive impenetrability versus experience-tuned weights (fixed to cognition, plastic to history). The effect is placed on the pre-attentive side of the fork precisely because it is impenetrable — surviving knowledge of the trick, repetition, and the instruction to hear the audio. Yet susceptibility varies reliably by language community and by developmental history, which means the integration weights are calibrated by long-run linguistic exposure. These two facts pull in opposite directions on "how fixed is it": the percept cannot be moved by any online act of will or belief, yet it has been shaped, over years, by experience. The resolution is that impenetrability is a claim about the timescale of cognition (no reasoning-out-of-it in the moment) while plasticity is a claim about the timescale of learning (weights set by exposure) — but the concept's rhetoric of an automatic, hard illusion sits uneasily with its own evidence that the weights are learned. Diagnostic: Is the invariance being claimed impenetrability to in-the-moment cognition (true) or fixedness across experience and development (false — the weights are tuned)?

T3: The clean weighting rule versus the phenomenon's variability (a tidy function over a noisy effect). The reliability-weighting rule gives the effect its predictive elegance: visual capture rises as audio degrades, a measurable function of channel reliability. But the McGurk effect is notoriously variable — susceptibility ranges widely across individuals, stimulus tokens, talkers, and labs, with the same paradigm yielding very different fusion rates. The tension is that the concept sells a smooth, lawful weighting process while the raw data are heterogeneous enough that "does this person show the effect at all?" is itself uncertain. The weighting rule is real as a within-subject trend but coexists with between-subject and between-stimulus scatter large enough to undercut any single population estimate. Present it as a clean function and you overstate its determinacy; foreground the variability and the lawful weighting looks like a signal buried in substantial noise. Diagnostic: Is the weighting claim a within-subject trend (audio degradation increases this perceiver's visual reliance) or a population constant — and does the variability across people and tokens swamp the rule?

T4: Vision as constituent versus audio faithfully encoded (what the eyes-closed toggle really shows). The eyes-closed/eyes-open toggle is the effect's one-second proof that vision is a constituent of the heard percept, not an aid. But the very same toggle shows that when the eyes close, the true audio /ba/ is heard cleanly and immediately — which means the auditory input was faithfully registered all along and only combined with vision at a later stage. So the toggle simultaneously supports "vision is part of what you hear" and "the audio was intact independently of vision." The tension is between reading the fused /da/ as evidence that there is no vision-free auditory percept and the toggle's demonstration that an accurate auditory percept is one blink away. Constitutive integration and preserved unimodal encoding both hold, and the concept's stronger phrasing (vision is the sound as delivered) glosses over the fact that the audio is recoverable the instant vision is removed. Diagnostic: Does the setting require vision to hear anything coherent (strong constituency), or is an accurate auditory percept available the moment the visual stream is removed (integration over an intact unimodal encoding)?

T5: The coherence gate as smart heuristic versus its exploitability (source-inference that dubbing fools). The coherence gate is an elegant adaptation: fuse two streams only when they plausibly share one source — same location, aligned within ~100 ms — and treat them as separate events otherwise. This protects the perceiver from spuriously merging unrelated sights and sounds. But the gate is a heuristic on proxies for common source (spatial and temporal coincidence), not on source identity itself, so it is straightforwardly fooled: a dubbed face and a mismatched soundtrack share neither talker nor true source, yet pass the spatio-temporal test and get fused, producing the illusion. The tension is that the same gating that makes integration ecologically smart is exactly what makes it exploitable by any stimulus engineered to satisfy the proxies without sharing a source. The feature is protective in the natural environment and misfiring in the edited one. Diagnostic: Do the streams genuinely share a source, or do they merely satisfy the spatio-temporal proxies (co-located, aligned within the window) that the gate mistakes for shared source?

T6: Autonomy versus reduction (the named speech illusion or the multisensory-integration parent). The McGurk effect is a specific, canonical demonstration with irreducible speech-specific cargo — phoneme fusion, articulatory visual cues, the motor-theory ties, the language-tuned weights — and in situ, as a probe of speech perception or a lip-sync design constraint, that specificity is exactly the point. But the integration machinery it exhibits is not proprietary: pre-attentive, reliability-weighted, coherence-gated combination is the parent (multisensory_integration, crossmodal_binding, Bayesian causal inference), and the ventriloquism, rubber-hand, and sound-induced-flash illusions are co-instances of that parent, not analogies to McGurk. The speech-specific content does not transfer even to the ventriloquism effect next door. The tension is between a named illusion that earns its own study and the recognition that its portable structure belongs to the multisensory-integration parent, while looser "a McGurk effect" for any context-overrides-signal case is mere analogy dropping the integrator, the weighting, and the gate. Diagnostic: Resolve toward multisensory_integration (with reliability weighting and source-attribution gating) when carrying the lesson to sibling cross-modal illusions; toward the named McGurk effect when phoneme fusion, articulatory cues, and language-tuned weights are specifically in play.

Structural–Framed Character¶

The McGurk effect sits toward the structural end of the spectrum but stops short of the pole — best read as mixed-structural, and it is a more structural case than the "effect" label would suggest, because what the illusion exhibits is a genuine, evaluatively-neutral perceptual mechanism rather than a normative judgment or a lab-constituted artifact. On three of the five criteria its structural credentials are strong. Its evaluative weight is nil: naming the effect convicts and praises nothing — it describes how audiovisual speech is fused, a fact about the perceptual system that is neither good nor bad, unlike a bias or a fallacy whose naming renders a verdict. It is largely not human-practice-bound: the pre-attentive, reliability-weighted, coherence-gated integration runs in any perceiver's brain automatically — cognitively impenetrable, unmoved by knowing the trick — so the mechanism operates without an observing analyst and does not dissolve when the study of it is withdrawn. Its institutional origin is essentially none: this is a discovered phenomenon (McGurk and MacDonald stumbled on it via a dubbing mismatch), a fact of how the brain combines articulatory streams, not an artifact of a tradition, agency, or evaluative practice — the name marks the discoverers, not an institution's construction. The one genuine wrinkle on the framed side of these three is that the illusion specifically requires a contrived, incongruent stimulus that does not arise in natural (congruent) speech, so "the McGurk effect" as a named demonstration is elicited by an engineered probe even though the integration mechanism it reveals is fully natural.

What keeps it off the structural pole is the remaining pair. On vocab_travels it scores low: its operative vocabulary — phoneme fusion, articulatory visual cues, the coherence gate, language-tuned weights, the motor-theory ties — is pinned to the speech substrate and does not float free, and on import_vs_recognize the transfer is bimodal in the way the entry stresses — within audiovisual-speech research the mechanism is recognized through different instruments, but the sibling cross-modal illusions (ventriloquism, rubber-hand, sound-induced flash) are co-instances of the parent, not analogies to McGurk, and looser "a McGurk effect" for any context-overrides-signal case is import-by-analogy that drops the integrator, the weighting, and the gate. The portable structural skeleton is pre-attentive multisensory integration — reliability-weighted, coherence-gated combination of conflicting channels into a single percept before awareness — and that skeleton is exactly what the McGurk effect instantiates from its parent primes (multisensory_integration, crossmodal_binding, Bayesian causal inference), not what makes "McGurk" itself travel: the cross-illusion reach belongs to that parent, while the speech-specific cargo — phoneme fusion, articulatory cues, the language-calibrated weights — stays home and does not transfer even to the ventriloquism effect next door. Its character: a real, evaluatively neutral, cognitively-impenetrable perceptual integration mechanism recognized across the perception sciences, elicited as a named illusion by a contrived speech stimulus and stated in speech-specific vocabulary that pins it to its home domain — structural in the integration skeleton it borrows from its multisensory-integration parent, hence mixed-structural rather than a free-floating prime.

Structural Core vs. Domain Accent¶

This section decides why the McGurk effect is a domain-specific abstraction and not a prime, and carries the case for its domain-specificity along with it.

What is skeletal (could lift toward a cross-domain prime). Strip the speech and a thin relational structure survives: two conflicting sensory channels carrying information about one event are fused, before awareness, into a single percept by reliability-weighted combination gated on the inputs plausibly sharing a source. The portable pieces are abstract: a multi-channel perceiver, channel-specific reliabilities that set the weights, a pre-attentive integrator, a coherence gate that decides whether to fuse at all, and weights calibrated by experience. That skeleton is genuinely substrate-portable — it recurs across the ventriloquism effect, the rubber-hand illusion, and the sound-induced flash illusion — and its recurrence is mechanism, not metaphor, which is exactly why those siblings are co-instances of the parents multisensory_integration and crossmodal_binding (with Bayesian causal inference). It is the integration core the McGurk effect shares, not what makes it the McGurk effect.

What is domain-bound. Everything that makes the concept the McGurk effect in particular is speech-perception furniture that does not survive extraction: phoneme fusion (the specific /ba/ + /ga/ → /da/ third syllable), articulatory visual cues, the language-tuned weights (susceptibility lower in Japanese than in Spanish- or English-speaking populations), the motor-theory-of-speech ties, and the ~100 ms audiovisual coherence window measured on articulating faces. The decisive test: this speech-specific content does not transfer even to the ventriloquism effect next door — a spatial visual–auditory integration with no phonemes, no articulation, no language-calibrated weighting. Remove the articulatory speech substrate and "phoneme fusion" and "articulatory cue" have no referents, leaving only the bare weighted-combination-plus-coherence-gate that belongs to the parent, not to McGurk.

Why this does not clear the prime bar. A prime's vocabulary travels and its transfer is recognition of the same mechanism, not analogy. The McGurk effect's transfer is threefold, and only its narrowest band is proprietary. Within audiovisual-speech research the mechanism is recognized intact through different instruments (fMRI localization, audiology, the linguistics debate). Its lip-sync design implication transfers literally as an instrument wherever streams plausibly share a source and drift past the window — but that is the same coherence-gate logic on a different stream pair, not the McGurk effect proper. And the integration machinery recurs across the sibling illusions — but that lesson belongs to multisensory_integration/crossmodal_binding, not to McGurk, while looser "a McGurk effect" for any context-overrides-signal case is analogy that drops the integrator, the weighting, and the gate. When the portable structure is genuinely needed, the parent already carries it in more general form; the phoneme fusion, articulatory cues, and language-tuned weights are home-bound cargo that stays in speech perception.

Relationships to Other Abstractions¶

Current abstraction McGurk effect Domain-specific

Parents (1) — more general patterns this builds on

McGurk effect is a decomposition of Bayesian Cue Integration Prime

Removing audiovisual-speech furniture leaves Bayesian Cue Integration's reliability-weighted fusion of simultaneous noisy cues to one latent speech event, conditional on common-source coherence.

Hierarchy path (1) — routes to 1 parentless root

McGurk effect → Bayesian Cue Integration → Precision Weighting → Aggregation → Micro Macro Linkage

Not to Be Confused With¶

Ventriloquism effect. The mislocalization of a sound toward a synchronized visual source — a puppet's moving mouth "pulls" the heard voice to it. It is the closest neighbor, but it integrates the visual and auditory streams for spatial location, with no phonemes, articulation, or language-tuned weights. It is a co-instance of the same multisensory-integration parent, not a form of McGurk. Tell: is the fused dimension which syllable is heard from conflicting articulation (McGurk), or where a sound is located from a visual cue (ventriloquism)?
Rubber-hand / body-transfer illusion. Visual–tactile integration for body ownership: synchronized stroking of a fake hand and the hidden real one makes the fake feel like one's own. Another sibling co-instance of the integration parent, operating on ownership rather than speech — no articulatory fusion. Tell: is the integrated percept a speech syllable (McGurk) or a sense of which body part is one's own (rubber-hand)?
Sound-induced flash illusion. Audition biasing vision: a single flash paired with two beeps is seen as two flashes. It is the reverse-direction sibling (audio shaping visual count), again a co-instance of reliability-weighted crossmodal integration, not a speech effect. Tell: does audio alter a visual count (sound-induced flash) or does vision alter a heard syllable (McGurk)?
Visual capture / lip-reading assistance. Outcomes where one channel overrides the other — vision simply overwriting the audio, or vision merely aiding a straining listener without changing the percept. The McGurk fusion is a third syllable present in neither input, not a winner-take-all override and not an optional aid. Tell: is the reported percept present in neither input (fusion — McGurk), or is one channel overwriting or assisting the other (capture / lip-reading)?
Top-down / post-perceptual inference. A cognitive judgment laid over a finished percept that a person can reason their way out of once they know better (as in many context effects). McGurk fusion is cognitively impenetrable — completed pre-attentively, surviving knowledge of the trick and the instruction to hear the audio. Tell: can the reading be corrected by knowing the truth (post-perceptual inference), or does it persist unchanged despite full knowledge (pre-attentive integration — McGurk)?
multisensory_integration / crossmodal_binding (the parent mechanism). The substrate-neutral pattern of pre-attentive, reliability-weighted, coherence-gated combination of channels — which McGurk instantiates on speech and which its sibling illusions co-instantiate. The cross-illusion lesson rides this parent, not McGurk. Tell: is the claim about weighted crossmodal fusion in general (the parent), or specifically phoneme fusion with articulatory cues and language-calibrated weights (McGurk)? (Treated more fully in an earlier section.)

Neighborhood in Abstraction Space¶

McGurk effect sits in a moderately populated region (52^nd percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Communication Channels & Modality (11 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12