The silence that still belongs to the melody

Why a sequence of now-points is not yet succession

Place four photographs on a table: a pendulum at the left, centre, right and centre again. The photographs display different positions. Nothing in any single photograph displays movement. An observer can infer motion by comparing the ordered set, but the inference is an act performed over already available items. Hearing a melody seems different. The succession itself is heard. The second note is heard as following the first, and the first is heard as having led into the second.

The snapshot model therefore faces a dilemma. If only the current tone is genuinely present, succession is not directly experienced. If previous tones are replayed as present alongside it, the listener should hear a chord. Neither result matches the case. Husserl’s retention occupies the difficult middle: the elapsed tone remains intentionally available, but in a mode marked by increasing pastness. It is not still sounding and not yet recollected as a separate episode.

Retention is not recollection

Suppose the quartet stops, and ten minutes later you deliberately remember its opening phrase. That act has a new object: the earlier performance as remembered now. You can attend to it, misremember it or compare it with a score. Husserl calls this secondary memory or recollection. Retention is different. While the phrase is still unfolding, you need not turn backwards, retrieve a record or represent the first note as a separate past event. Its having-just-been is built into how the later note appears.

This distinction can be tested in ordinary experience. During fluent speech, you understand the end of a sentence through words that are no longer acoustically present. Yet you do not normally experience a series of memory retrievals between syllables. When attention breaks and you ask, “What did you just say?”, the temporal object has already changed. You now reconstruct an episode that had formerly carried itself forward.

Protention is more basic than a prediction sentence

It is tempting to translate protention as “the brain predicts the next note”. Sometimes the analogy is useful. A familiar cadence can support a precise expectation. But Husserl’s claim is broader. Even an unfamiliar sound is encountered as continuing, ending, changing or possibly being interrupted. Protention can be an indeterminate openness rather than a detailed forecast. It supplies the future-facing edge through which an event can fulfil, disappoint or redirect what was underway.³

Consider surprise. A raw mismatch between an expected and observed symbol is not yet experienced surprise. To count as surprise within the phenomenon, the current event must overturn a live anticipation belonging to this unfolding object. The violation changes not only the next prediction but the interpretation of what preceded it. The joke’s final word makes the earlier sentence newly intelligible. The deceptive cadence retroactively reveals how strongly closure had been expected.

flowchart LR
  A["The silence that still belongs<br/>to the melody"] --> B["A present with depth and direction"]
  B --> C["What a measurement can and<br/>cannot inherit"]
  C --> D["The horizon-dependence laboratory"]
  D -. "re-test" .-> A
  classDef first fill:#0a1424,stroke:#7B4CD8,stroke-width:2px,color:#e6f3ff
  classDef step fill:#0a1424,stroke:#3f7da0,stroke-width:1px,color:#e6f3ff
  class A first
  class B,C,D step

A present with depth and direction

The living present is not a container

The phrase “living present” can mislead. It sounds like a box wide enough to contain several moments. Husserl’s analysis is more dynamic. The present is “living” because it temporalises. A fresh phase comes to prominence, prior phases recede in an ordered way, and the future horizon is revised. The field is continually self-differentiating. Its unity is not the simultaneity of many contents but the organised passage among modes of givenness.

This gives the present an asymmetry. Retention concerns what has occurred and carries determinate traces of it, although increasingly modified. Protention concerns what has not occurred and is often underdetermined. The two cannot simply be mirror images. The next note may take several possible forms; the previous note cannot become one of several alternatives without the experience changing into doubt or reconstruction. The temporal field has an arrow because fulfilment can settle an anticipation, whereas retention already bears the mark of occurrence.

The object’s time and the flow that discloses it

Two temporal orders are easy to collapse. The melody has an order: C occurred before D, the first phrase lasted four bars, and the cadence ended at a measurable time. This is the time of the object. The hearing of the melody also has a changing structure: C first appears, then recedes while D becomes prominent, and the whole phrase is carried towards completion. This is not a second melody hidden inside the listener. It is the temporal articulation through which the first melody can be experienced as one.

If consciousness of succession required a second observer to inspect a sequence of internal states, that observer would need its own consciousness of succession. The explanation would regress. Husserl therefore does not treat time-constituting consciousness as one more temporal object presented to an inner spectator. The flow is ordinarily lived through rather than watched. Reflection can later make a phase of it thematic, but reflection itself arrives with retention and protention and cannot stand wholly outside the flow it examines.¹

This distinction also disciplines system design. A log may record the object-time of inputs with perfect timestamps. A recurrent process may carry state across those timestamps. A user may experience a coherent event. These three facts can correlate while remaining non-identical. The log gives ordered evidence. The recurrent process supplies one possible causal bridge. The experienced flow is the phenomenon to be explained. Timestamped order is necessary for many temporal tasks, but order written in a ledger is not yet temporal presence.

A smallest useful calculation

An equation cannot prove Husserl’s account, but a toy model can expose the causal distinction. Let each note have a pitch value x. Let the recent intervals be Δx. Define a retained trend r as a weighted combination of the two most recent intervals. Use that trend to centre a distribution over the next note. This is not phenomenological retention. It is a deliberately thin functional proxy that lets us intervene.

The first line says that the latest interval contributes 60 per cent of the retained directional trend and the previous interval contributes 40 per cent. The second says that the next-note expectation is centred one retained trend away from the current note. The assumption is severe: pitch motion is treated as locally linear and context is compressed to two intervals. Remove that assumption and a richer model may encode harmony, metre, learned style or multiple timescales.

Fulfilment revises the past-facing field

When an expected note arrives, it does more than score well against a forecast. It can complete a phrase. Completion changes the status of earlier phases: what had been preparation becomes preparation-for-this-arrival. When a different note arrives, the phrase can be heard as detour, joke, mistake or beginning of a new pattern. The new phase therefore updates both sides of the field. Protention is fulfilled or disappointed; retention is reconfigured from the newly achieved present.

This two-sided revision explains why a temporal object can remain one object despite change. Identity is not sameness of content. A melody is precisely the ordered change of its phases. Its unity survives because each phase inherits and transforms a horizon. If every note began from zero, there would be no unfolding identity. If every note merely appended to an inert list, the list could preserve order without any phase being experienced as the current continuation of the whole.

One phrase, several scales

A syllable unfolds over tens or hundreds of milliseconds. A sentence may require seconds. A narrative establishes expectations over minutes. There is no reason to assume one universal window. Husserl’s formal structure describes a relation among retained, impressional and protentional phases. It does not set a neural duration. An empirical mechanism may implement several nested timescales, and the relevant scale may depend on the object being constituted.

This matters because “the present” is easily turned into a number. Ask how long now lasts, and an experimenter may report an integration window. Yet the duration of a mechanism, the duration of a stimulus structure and the experienced unity of an event are different variables. Their correlation is a research question, not a definition.

What a measurement can and cannot inherit

Three serious rival models

Husserl’s account belongs to a wider dispute about temporal consciousness. A retentional model says momentary consciousness represents a temporally extended event through retentional structure. An extensional model says the conscious episode itself extends through objective time, so successive phases of experience jointly present succession. Discrete or postdictive models propose that unconscious processing accumulates evidence before a conscious percept is assigned, sometimes after later input arrives.³¹³

These models can agree on many behaviours. All may explain why later context changes a percept. They disagree about what is extended, what is represented and when conscious content becomes determinate. No single illusion settles the dispute because the same response can be generated by different architectures. The discriminating work comes from interventions that preserve output while changing internal temporal organisation.

Neural windows constrain mechanisms, not phenomenological identity

Neuroscience offers strong evidence that different cortical regions integrate information over different timescales. Temporally scrambling a film disrupts higher-order regions more than early sensory areas when the scramble destroys long-range context. Narrated stories show a related hierarchy, from short acoustic windows to longer sentence and narrative windows.⁸⁹ Intracranial recordings also find slower dynamics and longer accumulation in higher-order regions.¹⁰

Music provides a sharper link to our case. During naturalistic listening, cortical responses track computational estimates of pitch and note-onset expectation. The finding supports the claim that anticipated musical structure influences neural processing. It does not show that a prediction-error signal is protention, or that the measured response is the experience of temporal flow.¹²

Repeated exposure to a film can shift event-specific neural patterns earlier, with different regions anticipating at different timescales.¹¹ Again, this is evidence for learned anticipatory processing. It becomes evidence about protention only through a bridge argument specifying which relation is preserved. “Both concern the future” is resemblance, not mechanism.

The tempo intervention

A useful mechanism question is whether temporal integration is yoked to absolute time or to the duration of meaningful structures. Stretch speech so every phoneme lasts twice as long. A structure-yoked mechanism should approximately double its integration window. A time-yoked mechanism should preserve a similar duration in seconds and therefore span fewer linguistic units.

Recent intracranial work used compressed and stretched speech to distinguish these possibilities. Integration windows in auditory cortex changed only slightly relative to the much larger change in speech-structure duration, favouring predominantly time-yoked integration in the measured regions.¹⁴ This result constrains neural models. It does not refute the claim that experienced objects have structure-sensitive horizons, because the experiment measured one mechanism in one modality at particular cortical sites.

Computational phenomenology: useful only with a break condition

Active-inference and predictive-processing accounts attempt to formalise temporally extended inference. Bogotá and Djebbara propose an “integrated continuity” in which past and future dependencies contribute to a modelled present.⁶ Grush’s trajectory-estimation approach similarly asks how an internal process can estimate a temporally extended trajectory rather than sample isolated states.⁵ These proposals are valuable because they generate architectures and interventions.

The mapping becomes overconfident when a computational dependency is called retention merely because it uses past state, or protention merely because it predicts. A cache can preserve tokens without representing their having-just-elapsed. A Bayesian smoother can use future observations to infer a past trajectory without any future being experienced as anticipated. A recurrent network can carry state without there being a subject for whom the present is thick.

What changes under rival metaphysics

Suppose a machine exhibits the complete HDP pattern. The metaphysical question has not disappeared. A reductive physicalist can treat the profile as a clue to the physical organisation that realises temporal experience, but still owes a bridge from causal dynamics to experience. A biological naturalist can accept every functional result while arguing that living neurobiology supplies constitutive powers the machine lacks. The profile then supports competence, not mentality.

A functionalist has the strongest reason to care about preserved horizon relations. If the right causal roles constitute mentality, substrate replacement should be harmless when those roles are genuinely preserved. Yet the HDP covers only four roles. It does not establish the full functional organisation of a subject, nor show that an output-equivalent mimic preserves internal relations. An emergentist may instead expect experience only when temporal dependencies participate in a larger self-maintaining organisation. The decisive level and threshold must then be specified rather than inferred from complexity.

An idealist can regard the temporal field as a structure within awareness and treat computational models as descriptions of regularities among appearances, not generators of awareness. Neutral monism and dual-aspect views can hold that the same underlying process admits physical and experiential descriptions, but they still need a principled correspondence between the two. Panpsychism may place rudimentary experience widely, yet a melody raises the combination and ordering problem: why do many micro-events form one temporally unified field? Process philosophy is naturally hospitable to becoming, transition and relational identity, but conceptual fit is not an implementation proof.

The HDP redistributes the explanatory burden without deciding the ontology. It tells every position which temporal relations a candidate preserves and where an output-only description is too weak. It cannot tell us whether awareness is fundamental, biologically constituted, functionally realised, emergent or one aspect of a neutral process. That limitation is productive because it prevents one successful model from being mistaken for a metaphysical verdict.

stateDiagram-v2
  [*] --> Claim
  state "Why a sequence of now-points<br/>is not yet succession" as Claim
  state "Retention is not recollection" as Test
  state "Protention is more basic than<br/>a prediction sentence" as Revision
  Claim --> Test: expose mechanism
  Test --> Decision: evidence survives
  Test --> Revision: rival survives
  Revision --> Claim: narrow or rebuild
  Decision --> [*]

The horizon-dependence laboratory

A new causal profile, not a consciousness score

The practical contribution of this paper is the Horizon Dependence Profile, or HDP. It asks whether a system’s treatment of the current phase changes when we intervene on the just-past, the expected continuation, the temporal scale and the continuity of internal state. The four results remain separate. Collapsing them into one score would hide which temporal property is present.

The profile can be applied to a human psychophysics task, a recurrent model, a streaming agent or a signal-processing pipeline. The observable may be a classification, confidence distribution, action proposal, segmentation boundary or neural readout. What matters is matched intervention: keep the current input fixed while changing one temporal property.

How to interpret the profile

A positive H result permits the statement that the system’s current output depends on prior trajectory under the tested interface. A positive V result permits the statement that future alternatives were differentially prepared. A Y value near zero indicates a fixed-time mechanism; a value near one indicates proportional structure yoking. A positive R result shows that hidden trajectory state contributes beyond the copied public summary.

A negative result is informative only within the assay’s sensitivity. H may be absent because the architecture is genuinely snapshot-based, because the supplied histories are equivalent for the task, or because the readout ignores a state that exists elsewhere. V may be absent because the system carries broad openness rather than a sharp forecast. R may be absent because the summary is causally sufficient, not because there was never any temporal integration.

Most importantly, the profile is not ordered from “less conscious” to “more conscious”. A thermostat with hysteresis may show H. A predictive controller may show H and V. A recurrent speech model may show all four. These properties make a system more relevant to temporal-organisation research, but the consciousness inference still requires a theory, a candidate boundary and rival generators.

Good measurement also requires invariance checks. Repeat the assay across unfamiliar sequences so memorised templates cannot carry the result. Vary output interfaces so a verbal explanation is not confused with the underlying dependency. Compare interventions at matched difficulty, and estimate uncertainty for every profile component. If H appears only when one history is noisier, the manipulation has changed evidence quality rather than temporal organisation. If R appears only after a long pause, fatigue or decay may rival state discontinuity. The profile earns causal force by excluding these simpler generators, not by displaying four impressive numbers.

function horizonProfile(candidate) {
  const sameNow = 65;
  const rise = [60, 62, 64, sameNow];
  const fall = [69, 67, 66, sameNow];

  const a = candidate.observe(rise, { tempo: 1, reset: false });
  const b = candidate.observe(fall, { tempo: 1, reset: false });
  const H = distance(a.current, b.current);

  const expected = candidate.continue(rise, 67);
  const violated = candidate.continue(rise, 61);
  const V = Math.abs(violated.surprise - expected.surprise);

  const fast = candidate.window({ tempo: 2 });
  const slow = candidate.window({ tempo: 0.5 });
  const Y = Math.log(slow / fast) / Math.log(4); // 0 time-yoked, 1 structure-yoked

  const uninterrupted = candidate.observe(rise, { reset: false });
  const reset = candidate.observe(rise, { reset: true, preserveSummary: true });
  const R = distance(uninterrupted.current, reset.current);

  return { H, V, Y, R }; // Keep a profile. Do not collapse to one score.
}

The code encodes four invariants. The current sample is identical in the history test. The past and current sample are identical in the continuation test. Tempo scaling preserves event order. Reset preserves the declared public summary. In an actual study, each invariant needs an auditable stimulus manifest and repeated trials.

Decision instrument. The matrix predicts patterns before data collection. “Human listener” remains an empirical target, and the reset assay requires an ethically and conceptually appropriate analogue such as interruption or state disruption.