Abstract:
A story can remember the past only if somebody can remember the story. This article asks how oral narrative solved that problem at scale. Drawing on the Homeric and Indian epics, it argues that narrative memory is distributed across embodiment, being-with, dwelling, multimateriality and multisymbolization: voice and meter, teachers and audiences, occasions and places, instruments and objects, formulas and genealogies. Memory emerges as constrained reconstruction, and fidelity itself has dimensions and costs. Writing, print and digital media redistribute and recombine this mnemonic labour, shifting scarcity from retention to recursive relevance, which makes the distinction between preservation and living memory decisive.
I. Introduction: The Story Has to Survive Before It Can Remember Anything
Narrative can make absent events available. It can carry a humiliation across thirteen years, a curse across a sea, the deeds of the dead into the lives of the unborn. Yet narrative is itself an absent event whenever it is not being told. Between one performance and the next, the story exists nowhere as a performance; it has fallen silent, and if nothing brings it back it will not come back. Before narrative can preserve a consequential past, then, it has to solve the problem of its own persistence. This yields a paradox that governs everything that follows: a story can remember the past only if somebody can remember the story.
The paradox is especially acute for oral epic. The Odyssey and the Mahabharata preserve vast quantities of narrated history, and neither began as a written composition whose exact wording was continuously available in a stable archive. The Odyssey runs to some twelve thousand lines of hexameter; the Mahabharata is traditionally reckoned at a hundred thousand verses. Both carry intricate genealogies, long causal chains, dozens of named participants and events whose consequences surface hundreds or thousands of lines after their occurrence. The obvious question is practical before it is aesthetic: how does a tradition retain enough narrative structure to be reproduced repeatedly, across performers and generations, without the help of a text?
One answer should be set aside at once. The question is not how a single bard memorised twelve thousand or a hundred thousand lines. That formulation presupposes the very things whose absence defines the problem: a fixed text, stored whole, retrieved by an individual mind. It imports the model of the library into a world that had none, and then marvels at how prodigious the librarians must have been. The better question is what distributed arrangements made a large narrative reliably reconstructable, so that it could be performed again, recognisably the same, by people who had never seen it written.
The argument of this article is that narrative memory is a multimediated accomplishment. Oral narrative is stabilized through embodied rhythm and performance, through interrecursive relations among performers, teachers and audiences, through recurrent places and occasions, through material supports, and through symbolic patterning. Meter, formula, recurrent scenes, performance settings, audience expectation, teacher-pupil transmission, genealogy and ritual repetition are therefore part of the machinery that makes large-scale narrative possible at all, rather than embellishments added to a narrative that already exists independently of them. Writing, when it comes, changes narrative by redistributing mnemonic labour away from repeated embodied and interrecursive reconstruction toward durable material and symbolic stabilization, and that redistribution changes what narrative can become. Print and digital media redistribute the labour again, each in its own direction.
The article proceeds from the general to the historical. Sections II and III set out an account of memory as reactivation rather than storage and trace the five mediations through which narrative remains available. Sections IV and V examine meter and formula as devices for making language returnable. Sections VI to VIII show how the Homeric and Indian epics stage remembering as interrecursive, and how audiences function as part of the mnemonic infrastructure. Section IX treats narrative form as a metabolic economy. Sections X to XIII follow the redistribution of mnemonic labour through writing, print and digital media, and Section XIV distinguishes preservation from living memory. The claim that holds them together is simple to state: the history of narrative media can be understood as a history of where the work of remembering is done.
II. Memory Is Not Storage
The first step is to reject the container model of memory. Neither minds nor societies simply store narratives in the way a digital file sits unchanged on a drive, waiting to be opened. Remembering is reconstructive. Past events become available through present configurations: through what a body can do now, what others can supply now, what a place and an occasion make relevant now, what objects and symbols now present. Frederic Bartlett's experiments, in which a Native American story was passed from one English reader to the next and steadily reshaped toward the readers' own expectations, established nearly a century ago that recall is a construction guided by schemas rather than the retrieval of a copy (Bartlett 1932). The point generalizes. A scar does not literally remember the wound, and a bed does not literally remember the marriage. Likewise, a performer need not contain a fixed text and retrieve it whole.
A basic distinction should govern the rest of the argument: the distinction between retention and reactivation. Retention is the persistence of constraints, traces, patterns and structures through which a past can later be recovered. A trained voice, a remembered melody, a genealogy learned in childhood, a place associated with an event, a verbal formula, an inscribed tablet: each retains something, in the sense that it narrows what can later be reconstructed. Reactivation is the present recursive process through which those traces become available again. A performance, a recitation, a reading, a conversation that begins with "do you remember": each takes up what has been retained and makes it present. Narrative depends on both. Retention without reactivation is inert, and reactivation without any retained constraint would be invention rather than remembering. Between these limits, reconstruction can range widely and still count as memory, because something from the past constrains it.
An oral tradition therefore faces a double requirement. It needs enough retention that each performance remains recognisably the same story, and enough flexibility that the story can be reconstructed under changing conditions: a different singer, a different audience, a shorter night, a patron who wants his ancestors mentioned. The requirement leads to an important negative claim: fidelity need not mean verbatim identity. Albert Lord's South Slavic singers insisted that they sang a song exactly as they had heard it, word for word, and the recordings made of their performances showed extensive variation in wording from one telling to the next (Lord 1960). The contradiction dissolves once one asks what the singers meant by a "word": for them it could be a line, a phrase, a scene, a unit of the story, and what remained the same was the song in that sense.
The relevant question for oral tradition, then, concerns less whether every sequence of words remains fixed than which narrative relations, episodes, characters, causal chains, genealogies, formulas and recognitional structures remain stable enough for the narrative world to persist. Some traditions do achieve verbatim fidelity without writing. The Vedic hymns have been transmitted orally for thousands of years with extraordinary exactness, through techniques of recitation in which the text is chanted word by word, in overlapping pairs, and in intricate braided sequences, so that every syllable is fixed by several independent patterns at once. That achievement, however, required an immense investment of embodied training and specialized social organisation, and it was undertaken for texts whose exact sound was itself held to matter. Many oral epic traditions have tolerated substantially greater verbal variation, while protecting character, episode, sequence and relational structure with considerable tenacity. The contrast already suggests that fidelity is not a single quantity, a point Section IX develops.
III. The Five Mediations of Narrative Memory
Rather than five storage systems, the five mediations are better understood as five dimensions through which narrative can remain recursively available, each contributing its own kind of constraint to the reconstruction of a story, and each interlocking with the others.
Embodiment. Narrative memory lives in voice, breath, rhythm, meter, gesture, timing, auditory expectation, and muscular and vocal habit. A singer trained in a tradition can continue a patterned line because the body has learned its cadence, in much the way a skilled walker does not decide where each foot goes. The line comes to the voice partly because the voice already knows what shape a line has. The key claim here is that meter is symbolic structure made bodily reproducible. It is precisely the kind of cross-mediational device that a multimediated account predicts, one in which symbolic possibilities are constrained by embodied rhythm, so that a sequence becomes easier to reconstruct because only certain continuations fit the body's expectation of what comes next.
Being-with. Narrative is often remembered with others. Teacher and apprentice, performer and audience, co-performers, and previous generations whose formulations are inherited all take part. Audience members know when something has been omitted, altered or mistold. Correction, anticipation, response and recognition distribute the labour of remembering across everyone present, and across many who are absent but whose earlier performances shaped what counts as the story. Hence a strong proposition: tradition is interrecursive memory extended through generations. The proposition needs immediate qualification if it is to remain exact. Dead performers are not present interrecursive counterparts: they cannot respond to what the current singer does, and nothing the singer does can become a condition of their further activity. What persists of them are traces of their earlier recursive activity, sedimented in formulas, teaching lineages, expectations, remembered performances and, later, recordings. Tradition, more precisely, extends the consequences of past interrecursivity across generations by continually recruiting new interrecursive participants. Teacher and pupil are interrecursive now; performer and audience are interrecursive now; their predecessors are not, but what those predecessors did helps constitute the field within which the present interrecursivity occurs. No single performer needs to contain the whole tradition, because each performance is generated within a field that is partly sedimented by past relations and partly reconstructed, actively and in the present, through current ones. The tradition is, in this sense, a consequential past of its own, kept available by the living who continue to respond to it. Lord's account of how a South Slavic singer learned the craft shows the relational shape of this extension. A boy first listened, often for years, absorbing the stories, the rhythm and the phrases without singing. He then began to sing for himself, fitting phrases to the meter, usually alone or among intimates. Only later did he sing before an audience, whose response taught him what a full and acceptable song required (Lord 1960). Rather than being handed a text at any stage, he was placed, progressively, inside a network of performers and listeners whose recognition defined what he was learning. Jan Assmann's distinction between communicative memory, carried in everyday interaction across roughly three generations, and cultural memory, stabilized in institutions, rituals and texts that outlast living recollection, describes two regimes of this extension (Assmann 2011).
Dwelling. Narration happens somewhere and somewhen: at night, in a particular season, by a fire, in a forest clearing, on a route of pilgrimage, at the place where the story is said to have happened. The environment can scaffold memory without containing symbolic content in any narrow sense. Keith Basso's work among the Western Apache showed how place-names in the landscape each evoke a story, so that travelling through country is a way of being reminded, and a speaker can recall a whole moral history to a listener simply by naming a place (Basso 1996). A story told in the same place, season or ritual cycle is easier to reactivate because the setting restores part of the constellation within which the narrative ordinarily lives. The Mahabharata is itself recited, in its outermost frame, to sages gathered in the Naimisha forest, and the forest is part of the scene of remembering.
Multimateriality. Long before writing, material form is involved in narrative memory. Musical instruments, from the Homeric lyre to the one-stringed gusle of Lord's singers, shape and pace performance. Built settings such as the feasting hall, the court and the shrine gather audiences and cue the kinds of stories that belong there. Monuments, graves, weapons and genealogical markers anchor narrated events in objects that can be pointed to. Among the Luba of central Africa, specialists of the mbudye society read histories from the lukasa, a hand-held board studded with beads and shells, whose arrangement prompts the recitation of kings, migrations and sacred places, and whose readings vary with the reader and the occasion (Roberts and Roberts 1996). Later come tablets, manuscripts, books, print, recordings and digital archives. Material anchors do not merely store content; they reduce the amount of active reconstruction required. That becomes central once writing enters the analysis.
Multisymbolization. Names, genealogies, formulaic phrases, repeated episodes, parallel structures, conventional openings, type-scenes, stock epithets, meters, curses, prophecies, refrains and the conventions of narrative sequencing itself all reduce the number of plausible next moves. A prophecy tells a listener what must eventually happen; a genealogy tells who must be mentioned; a type-scene tells what steps a feast or an arming requires; a formula tells which words fit here. David Rubin's cognitive study of oral traditions showed how constraints of meaning, imagery and sound combine, each alone insufficient, to narrow recall until the next line becomes nearly inevitable (Rubin 1995). This gives a formulation that the rest of the article depends on: mnemonic form works by constraining possibility. The easier it is to know what could plausibly come next, the easier the narrative is to reconstruct.
The mediations work together, and their conjunction is what makes the tradition robust. A singer at a feast, before an audience who know the story, with a lyre in hand and a meter in the body, drawing on formulas and scenes learned from a teacher, reconstructs a narrative under five sets of constraints at once, any of which can compensate for a lapse in another. Oral narrative memory, on this account, is the recurrent alignment of embodied, relational, environmental, material and symbolic constraints, rather than symbolic content stored in a brain.
Constraint is the concept that unifies these observations, and it will run through the rest of the argument. Meter constrains; formula constrains; audience expectation, genre, place and material cues constrain. Writing, when it comes, constrains wording far more tightly; print constrains variation among copies; an archive constrains reconstruction by keeping traces retrievable. Memory, on this account, sits between the retrieval of a stored copy and unlimited present invention, the two alternatives the argument rejects, and it can be named precisely: narrative memory is constrained reconstruction. A past remains a past, rather than becoming a present fabrication, because present reconstruction encounters constraints left by what came before. The history of narrative media can then be read as a history of changes in which constraints are retained, where they are lodged, and what it costs to maintain them.
IV. Meter as a Technology of Returnability
Meter deserves separate treatment, because it is at once the most obvious and the least theorized feature of early long narrative. Very early long narratives are, with striking regularity, metrically organised. Homeric epic is composed in dactylic hexameter, a line of six feet with a flexible interior and a fixed close. The Mahabharata is composed overwhelmingly in the shloka, a stanza of four eight-syllable quarters with constrained patterns of long and short syllables toward the end of each quarter, interspersed with passages in longer meters. The two are very different devices, and nothing in the argument depends on treating them as the same. The claim is more general: meter makes language more returnable.
It does so in several ways at once. Meter limits acceptable forms, since only words and phrases of certain shapes fit certain positions. It creates expectations, so that a listener, and a performer, feels where a line should close. It creates temporal slots, so that the line has a known length and a known internal structure into which material can be fitted. It allows partial reconstruction from rhythm, since a half-remembered phrase can be completed by what the meter requires. And it joins symbolic structure to bodily timing, so that the memory of the line is carried in breath and pulse as well as in the recollection of words.
The interaction of meter and phrase can be seen at the level of a single line. The Homeric hexameter characteristically divides at a caesura in its third foot, and the stretch from that point to the line's end is a slot of known shape. A large set of formulas exists to fill exactly such slots. A singer who has brought the line to its caesura and needs to name the hero as subject has, ready to hand, a phrase of the right length and rhythm; the choice has been made by the tradition, and the singer's attention is free for the next line. The meter, in other words, both constrains what may be said and creates positions at which the tradition has already prepared what to say.
For this reason, meter is better understood as a technology of recursive returnability than mainly as aesthetics. Returnability here means the possibility of re-entering a sequence after interruption and resuming it. A line can be re-entered because its shape is known. A sequence can be resumed because the rhythm has continued in the body even where the words have faltered. A performer who momentarily loses the exact wording still inhabits a patterned space of possible continuations, and the meter narrows that space enough for a fitting continuation to be found. A broader thesis follows: form is one way narrative protects itself from forgetting.
Rhyme, where it is used, does comparable work by different means, binding line-ends to one another so that each close calls up the next. Alliteration in Germanic verse binds the halves of a line. Parallelism in much Near Eastern and Mesoamerican verse lets the second member of a pair be reconstructed from the first. Each is a way of making language anticipate itself. Formula and repetition, discussed below, reduce the search space further.
One distinction must be kept carefully. The strongest argument here is not that ancient performers memorised fixed texts more efficiently because they had meter. That would return to the storage model by another route. Milman Parry's analysis of Homeric diction, extended by Lord's fieldwork, showed that the relevant process in epic traditions is composition in performance: the singer rebuilds the song each time from a stock of formulas, scenes and story-patterns, within the meter, rather than reciting a text learned by heart (Parry 1971; Lord 1960). Meter matters for such a process because it makes reconstruction and transmission more robust. It does not store the song; it makes the song rebuildable. The Vedic case, where meter supports verbatim recitation, and the epic case, where it supports recomposition, show that the same kind of device can serve different mnemonic regimes, depending on what else the tradition has decided must be fixed.
V. Formula, Type-Scene, and the Reduction of Mnemonic Search Space
Beyond meter lies a larger apparatus of recurrence: recurring phrases, repeated scenes, conventional assemblies, feasts, arming scenes, journeys, recognitions, genealogical introductions, combat sequences and formulaic epithets. Parry defined the formula as a group of words regularly used under the same metrical conditions to express a given essential idea, and he showed that the Homeric system of noun-epithet formulas was both extensive and economical, providing, for many characters, one formula for each metrical position and grammatical case (Parry 1971). Odysseus can be "much-enduring godlike Odysseus" where the end of the line needs that shape, and "Odysseus of many wiles" where it needs another. Dawn is rosy-fingered; Athena is grey-eyed; the sea is wine-dark. Scholarship since Walter Arend has identified the typical scenes of Homeric narrative, sequences such as arming, sacrifice, hospitality and assembly that recur with a stable order of elements and variable detail (Arend 1933). When a guest arrives in the Odyssey, water is brought for his hands in a golden jug over a silver basin, a table is set, bread and meat are served, and only then are questions asked.
Lord extended the notion from phrase to scene, calling the recurrent groups of ideas from which songs are built "themes": the assembly, the summons, the letter, the arming of the hero, the journey, the return (Lord 1960). The Homeric arming scene illustrates the principle across a whole poem. Paris arms for his duel with Menelaus, Agamemnon for his great day of fighting, Patroclus for his fatal sortie, and Achilles for his return to battle, and each arming follows the same order: greaves, corselet, sword, shield, helmet, spear. The scene is elaborated in proportion to its importance, most fully for Achilles, whose new armour is the work of a god. The sequence is fixed; the expansion is free.
These devices can be analysed as means of reducing the combinatorial burden of producing narrative. The performer does not choose every element from scratch. Tradition supplies a structured field of possibilities, in which many choices are already made, so that attention can go where it is needed. This is analogous to the settlement of coordination at L1. What might otherwise require constant L3 effort, sustained and explicit reconstruction of every phrase and every step of every scene, becomes partially sedimented into patterned ease, available without deliberation because it has been done so often, by so many.
Far from being uncreative, as a literary taste formed by print might suppose, a stock scene is a stabilized scaffold that frees recursive effort for variation elsewhere. The hospitality scene can unfold on its familiar rails while the singer concentrates on what is distinctive about this guest, this host, this moment of danger concealed beneath courtesy. The arming of Patroclus in Achilles's armour follows the arming type-scene, and it is exactly the familiarity of the sequence that lets the single departure from it, the spear he cannot lift, carry its weight.
This allows a formulation that bridges the account of oral form and the graded account of recursive coordination: oral-formulaic tradition converts recurrent narrative problems from high-cost explicit reconstruction toward lower-cost patterned ease. A performer can work within deeply sedimented forms while still improvising or adjusting at the edges, and much of the art of oral composition lies in the relation between the two. What matters, then, is less the choice between fixed and improvised than where recursive effort is concentrated. A tradition that sediments its formulas and scenes can invest its performers' living attention in the particular, the dramatic, the adapted, and the new.
VI. The Muse: Memory Explicitly Staged as Interrecursive
The Homeric poems place a relation at the threshold of narration. The Iliad bids a goddess sing; the Odyssey asks the Muse to tell. Before the catalogue of the ships, the Iliad gives the reason: the Muses are goddesses, present at all things and knowing all things, while mortals hear only report and know nothing, and no singer, even with ten tongues and a voice that could not break, could name the multitude unless the Muses recalled it. In Hesiod's Theogony, the Muses are daughters of Zeus and Mnemosyne, Memory herself, and they tell the shepherd-poet on Helicon that they know how to say many false things that resemble truth, and how to speak truth when they wish. Hesiod adds that when a singer, servant of the Muses, sings of the deeds of former men and of the gods, a man with fresh grief in his heart forgets his sorrows at once. The genealogy is precise: memory is the mother of song, song is the child of a divine union, and what song gives its listeners includes a forgetting of their own troubles in the presence of other people's.
The invocation is extraordinary because it refuses the image of the isolated remembering individual. Instead of claiming to possess the past, the singer appeals to another, one who was present when he was not, and the act of narration therefore begins with a relation. It would be a mistake to literalize this into a thesis about Homeric psychology, as though singers experienced recollection as the arrival of a goddess. The invocation is better read as a formal staging of an ontology of memory in which access to the absent past depends on being-with. The singer remembers with another.
The poems extend the same structure to their human singers. When Demodocus sings at the Phaeacian court of the Achaeans' sufferings, Odysseus praises him for singing it as though he had been present himself or had heard it from one who was. Presence, direct or relayed, is the standard of a true song. And when Phemius, the singer in Odysseus's hall, pleads for his life after the slaughter of the suitors, he says that he is self-taught, and that a god has planted all kinds of songs in his mind. The two claims, self-taught and god-given, sit side by side without contradiction. The singer's competence is his own, and it is also received; it is formed in him, and it comes from beyond him.
Ordinary remembering has the same shape whenever more than one person takes part. "Do you remember when...?" One contributes; another corrects; another adds. Memory emerges relationally, and the version that emerges may belong to no one singly. Research on conversational remembering has shown how such joint recall reshapes what each participant subsequently remembers, aligning memories in some cases and silencing what went unsaid in others (Hirst and Echterhoff 2012), and work on distributed remembering describes joint recall as coordination of cognitive, bodily, social and material resources across several timescales (Bietti and Sutton 2015; Sutton 2026).
Performance widens the relation further. The performer remembers with the audience, with inherited formulas, with teachers, with earlier performances, with expectations about what the story must contain. The Muse offers no evidence that Homer anticipated the theory of distributed cognition, but she is an extraordinary literary enactment of what a multimediated account predicts: epic memory is represented as something too large for an isolated self.
VII. The Mahabharata: Memory Through Relay
The Mahabharata offers an equally strong case, built on a different principle. Where the Homeric singer appeals upward to a goddess, the Indian epic arranges its remembering as a chain of human and near-human relays. The epic calls itself itihasa, a word conventionally glossed as "thus indeed it was", and it belongs, in the classical scheme, to smriti, "what is remembered", as against the shruti, "what is heard", of the Veda. Its category is already a category of memory.
Its outermost frame presents the bard Ugrashravas, son of the bard Lomaharshana, arriving at the twelve-year sacrifice of Shaunaka in the Naimisha forest and reciting to the assembled sages what he heard Vaishampayana recite at the snake sacrifice of King Janamejaya. Vaishampayana recited it at the bidding of his teacher, Vyasa, who composed it and who was present at the sacrifice. Within the war books, the battle itself reaches its first audience through a further relay. Vyasa offers the blind king Dhritarashtra sight so that he may watch the war; the king declines, and Vyasa grants divine vision instead to his charioteer and counsellor Sanjaya, who reports what he sees. When Sanjaya first returns from the field, it is to announce that Bhishma has fallen, and only then does he narrate the battle from its beginning. The story of the war is told backwards from its first great consequence.
The striking point is that none of these relays is neutral. Each is interrecursive and positioned. Vyasa is far more than the author: he is the biological father of Dhritarashtra and Pandu, the grandfather of the warring cousins, a participant who intervenes at critical moments, a witness, and the arranger of the whole. Janamejaya, far from being a neutral audience, is Arjuna's great-grandson, and he hears his ancestral past during a sacrifice undertaken to destroy the serpents in revenge for the death of his father Parikshit, killed by the serpent Takshaka. Sanjaya narrates a battle to a father whose sons are dying in it, and the father's questions, laments and self-justifications are part of what the narration records. Even the scene of writing that later tradition added to the epic, in which Vyasa dictates the poem to Ganesha on condition that the god-scribe never pause, while Ganesha must understand each verse before he writes it, stages transmission as a contest between two parties with conditions of their own. The critical edition relegates the passage to its apparatus, and its later arrival is itself telling: when writing enters the epic's image of itself, it enters as another relation.
Memory, in this epic, is transmitted through stake-bearing relations. This offers a strong contrast with modern archival assumptions. Modern archives aspire, at least ideally, to separate preservation from personal involvement: the archivist is meant to keep the record without being part of it, and the record's authority depends in part on that separation. The Mahabharata repeatedly does the opposite. Its chain of transmission carries relational stakes with it, and those stakes are displayed, not concealed. The epic famously claims that whatever is here may be found elsewhere, and whatever is not here is nowhere, and it makes the claim through voices that are themselves inside the history they encompass.
This yields one of the article's central formulations: in the Mahabharata, remembering is more than an activity directed at a consequential past; the act of remembering is itself part of the consequential past's continuing life. The telling is another event in the chain. Transmission, then, belongs to narrative history rather than standing outside it, and the listener who asks to hear it is already one of its descendants, already acting within what it tells.
VIII. Audience as Mnemonic Infrastructure
Modern literacy encourages the image of the audience as recipient: the reader alone with a book, the spectator in a darkened room, taking in what has already been fixed elsewhere. For oral traditions this image is far too passive. An audience can recognise, anticipate, correct, demand repetition, notice omission, respond emotionally, supply cues and confirm that a performance belongs to the tradition. Lord described singers performing in the coffeehouses of Bosnia during the nights of Ramadan, before listeners who came and went, drank, talked and called for particular songs, and whose attention or restlessness shaped how long a song became and how richly it was ornamented (Lord 1960). A singer who told a story wrongly before such an audience would hear about it.
The audience therefore partly stabilizes what counts as continuity. This gives a proposition as important as any about performers: the audience does not merely receive narrative memory; it helps constitute the conditions under which narrative remains rememberable. A tradition survives not only because singers can sing it but because listeners can tell a good performance from a bad one, and because their recognition is a constraint the singer must satisfy.
The point is visible even where writing is present. At the great Ramlila of Ramnagar, near Varanasi, where the story of Rama is enacted over a month, devoted spectators follow the performance with their own copies of Tulsidas's Ramcharitmanas, reading along as the verses are chanted. Written text, performance and audience knowledge interlock, and the audience's books are part of the apparatus that keeps the performance true. The book does not replace the relational memory; it sits inside it.
Institutions can formalize the audience's authority. A tradition preserved in the Platonic dialogue Hipparchus credits the son of Peisistratus with requiring that the Homeric poems be recited at the Panathenaic festival in order, each rhapsode taking up where the previous one had stopped. Whatever its historical accuracy, the report describes a transmission in which no single performer carries the whole and in which the sequence is guaranteed by succession before a festival audience that knows what should come next. The relay is itself a mnemonic device, and the audience is the witness that it has been run correctly.
This connects directly to being-with, and it explains why narrative transmission is social rather than merely cognitive. Several regimes of remembering can be distinguished by how widely the labour is distributed. Private recall is carried by one person, with whatever bodily and material resources that person has. Dyadic recall is carried between two, each correcting and cueing the other. Group reconstruction draws on several participants, none of whom holds the whole. Traditional performance extends this further still, distributing mnemonic authority across a whole social field: the performer, the performer's teachers and their teachers, the present audience, and the absent community whose standards the audience embodies. In the last regime no one person, including the performer, is the final authority on what the story is. The story is what the field recognises.
IX. Narrative Form as Metabolic Economy
Long narrative is expensive. It consumes attention, training, repetition, performance time, social support and the labour of memory itself. A singer must be trained for years; an audience must be free to listen for hours or nights; patrons must feed and house performers; the whole apparatus must be maintained from generation to generation. Why, under these costs, do oral traditions develop such elaborately patterned forms?
The answer is that form can economize the cost of reconstruction. Repeated structures stabilize what would otherwise have to be actively reinvented. Meter lowers search. Formula lowers search. Genealogy organises relation, so that the question of who belongs where in the story need not be solved afresh. Type-scenes regularize sequence. Audience expectation reduces uncertainty about what is required. Performance setting narrows relevance, so that the story appropriate to this occasion is already partly selected by the occasion. Each device transfers some of the cost of remembering from the living effort of the performer to a structure that has been built up over generations and can be drawn on without being rebuilt.
This makes it possible to speak of oral-formulaic narrative as a cultivation of remembering. Culture, understood as activity rather than inventory, is the cumulative cultivation of embodiment, dwelling and being-with through multimateriality and multisymbolization, undertaken because coordination beyond the reach of embodied bonding is a recurring and unavoidable stake in human life. Oral epic can be seen as one highly specialized cultivation of transgenerational being-with through narratively stabilized memory. It cultivates bodies, in the training of voice and rhythm. It cultivates relations, in the bonds of teacher and pupil and in the knowledgeable audience. It cultivates occasions and places, in the festivals and nights reserved for telling. It cultivates material supports, in instruments, halls and objects. And it cultivates symbolic form, in meter, formula, scene and genealogy. Narrative traditions, in short, cultivate their own conditions of persistence.
The framing also clarifies the difference between epic and Vedic transmission. The Vedic schools chose to fix the exact sound of their texts, and they paid for that choice with an extraordinary investment: years of daily recitation beginning in childhood, the multiple braided patterns in which each hymn was learned, hand movements marking the pitch accents, and lineages of teachers whose whole social identity was organised around the preservation. Many oral epic traditions, by contrast, have tolerated substantially greater verbal variation while strongly protecting character, episode, sequence and relational structure. Neither pattern is more primitive than the other, since each allocates metabolic cost differently, according to what the tradition holds must survive unchanged.
The contrast yields a more general principle: fidelity itself has dimensions and costs. A tradition can invest heavily in exact phonological fidelity, in verbal fidelity, in metrical fidelity, in the fidelity of episodes, of genealogies, of causal sequence, or of ritual performance, and no tradition can afford to maximize all of them at once. Vedic transmission protects exact sound at enormous cost. Oral epic can tolerate lexical variation while holding its characters, episodes, sequences and relations remarkably steady. Genealogical recitation may protect names and order while letting everything else move. Writing, as the following sections show, drastically lowers the cost of verbal fidelity, and thereby changes which dimensions a tradition must spend its own living labour to protect. The question to ask of any transmission is therefore less how faithful it is than what the tradition spends its resources on keeping invariant.
The metabolic framing also explains why the forms are so stable. A cultivating structure that has succeeded in lowering the cost of a recurring task is difficult to abandon, because abandoning it would return the task to its full cost. Formulas outlive the circumstances that produced them; epithets persist after their meaning has become obscure; archaic words survive in the meter because the meter needs their shape. What looks like conservatism in oral tradition is often the inertia of a successful economy.
X. Writing Redistributes Memory Rather Than Inventing It
The major historical pivot is the arrival of writing, and it has too often been told as a Great Divide: before writing, societies without history, without archives, without the capacity for analytical thought; after writing, memory, history and reason. The strong versions of this story, associated with Jack Goody, Eric Havelock and Walter Ong, drew attention to real transformations (Goody 1977; Ong 1982), and their critics, among them Ruth Finnegan, showed how badly the dichotomy misdescribes the variety of oral traditions and their constant interaction with writing (Finnegan 1988). The account offered here sides with the critics in one respect and with the Great Divide theorists in another. Writing does not create cultural memory. Nonliterate societies are not societies without archives; their archives are differently distributed, carried in embodied performance, social transmission, repeated occasions, mnemonic form and collective recognition. Writing does, however, change something fundamental, and what it changes can be specified.
Writing moves more of the burden of persistence into durable objects that are at once material and symbolic. The exact sequence of words can now survive without continuous human rehearsal. This changes the burden entirely. A story no longer has to be actively reproduced in order not to disappear. In an oral tradition, a story that no one tells for three generations is likely to be lost; a written story can lie unread for a thousand years and still be there to be read.
This can be named mnemonic redistribution: a shift in which mediations carry the burden of making a past retrievable. With writing, less embodied repetition may be required, and less dependence on a particular performer. Less dependence on co-present audiences follows, since a text can be read by one person alone, far from any community that knows the story. Greater exactness becomes possible, since the words themselves can be checked, and verbal fidelity, which oral traditions could secure only at the extraordinary cost the Vedic schools paid, becomes cheap. Greater temporal separation from the originating community becomes possible, since a text can outlive the language community that produced it and be recovered by strangers.
There are costs as well, and a theory of redistribution must count them. The text can survive after the performance tradition dies, and when it does, much of what the tradition carried in bodies, relations and occasions is lost even though the words remain. Meaning can detach from shared embodied context, so that readers encounter formulas without the rhythms that gave them life and allusions without the audience that caught them. Interpretation must compensate for lost co-presence, and commentary, philology and scholarship arise in part to supply, by explicit reconstruction, what a living audience once supplied by recognition. Plato's Phaedrus already articulates this cost. In the myth Socrates tells, the Egyptian king Thamus rejects the god Theuth's claim that writing is a remedy for memory. It is a remedy only for reminding, he says, and it will make those who learn it forgetful, since they will rely on external marks rather than remembering from within; and a written text, Socrates adds, cannot answer when questioned, but always says the same thing. The complaint is, in effect, that writing breaks the interrecursive relation: the text cannot respond to its reader's response.
The trade-off is exactly the kind a multimediated account should be able to describe. Nothing is simply gained or simply lost. Burdens move from embodiment and being-with toward multimateriality and multisymbolization, and the move changes what each of the mediations must do.
Redistribution, however, describes only half of what happens. The other half may be called mediational recombination: the continuing availability of earlier mnemonic mediations, which combine with newer ones and can become newly salient under changed conditions. Mnemonic regimes accumulate and recombine; they do not simply supersede one another. Writing did not abolish meter, which went on organising written poetry for millennia. Print did not abolish oral storytelling, which continued in households, pulpits, taverns and nurseries alongside the book. The Ramnagar spectators following the chanted verses in their own printed copies show performance, text and audience knowledge operating together in a single scene. Redistribution tells where more of the burden moves; recombination makes clear that the mediations it moves away from remain available, and can return to the centre. Holding the two together is what separates the present account from the Great Divide story. Writing shifts the relative distribution of mnemonic labour; it does not inaugurate an entirely new cognitive epoch in which the older mediations cease to work.
XI. What Writing Lets Narrative Afford
The redistribution suggests a historical hypothesis. Once external material and symbolic supports carry more of the mnemonic burden, narrative can devote more of its organisation to features that oral transmission makes expensive. Among these might be longer-range cross-reference, since a reader can turn back and a writer can plan a detail on the first page to pay off on the last; greater lexical exactness, since a precise word can be relied on to survive; more intricate temporal dislocation, since a reader with a book can hold complex sequences that a listener could not; silent interiority, since a text read alone can dwell in a mind without having to be performed aloud; more sustained selfrecursivity, the long following of one consciousness through its own changes; denser symbolic patterning; greater tolerance for passages that would be hard to remember orally; and more unstable or idiosyncratic form.
The hypothesis must be held with care, and three cautions should be explicit. It does not claim that writing simply causes the novel. It does not present oral traditions as formally simple; the architecture of the Mahabharata's frames or the ring-compositions of Homeric speech are as intricate as anything in written literature. And it does not describe a teleology, a march from oral simplicity to written sophistication. The claim is more modest and more defensible: different mnemonic infrastructures alter the cost structure of narrative form. Some forms become cheaper and others more expensive, and over long periods, what is cheaper tends to be done more.
The cautions are borne out by the history itself. Augustine records his surprise at seeing Ambrose read without moving his lips, a moment often cited as a milestone in the history of silent reading, and it shows that silent interiority in reading was noticeable, not universal, in late antiquity. Murasaki Shikibu's Tale of Genji, composed around the year 1000 and circulated in manuscript, sustains an attention to shifting inner states over an enormous span, long before print. The relation between writing and interiority is real, and it is neither sudden nor confined to one tradition.
It becomes possible, then, to pose a question rather than to settle it. Did the externalization of narrative memory help create the conditions under which sustained selfrecursive fiction could become more prominent? Emma Bovary's slow erosion, followed through the free indirect style that lets the narration slip in and out of her consciousness, or the feverish self-interrogations of Dostoevsky's characters, would be very costly to carry in an oral tradition, where each performance must rebuild the whole from what performer and audience can hold. In writing, the text holds it, and the reader can return. If selfrecursive narrative is, as it appears to be, a marked departure from the interrecursive baseline of most narrative traditions, the redistribution of memory into books may be one of the conditions that lowered its cost enough for it to flourish.
XII. From Manuscript to Print: Exact Repetition Becomes Cheap
Writing stabilizes; print changes scale. Printing produces many materially identical copies, greater reproducibility, broader audiences and reduced dependence on any particular line of manuscript copying. Elizabeth Eisenstein argued that this "fixity" transformed European learning (Eisenstein 1979), and Adrian Johns showed that fixity was itself an achievement, laboriously secured against piracy, error and variant editions rather than given by the technology (Johns 1998). Both positions are compatible with the account here. Print redistributes mnemonic labour again, and the redistribution had to be cultivated like any other.
Some of the tools that print generalized were invented in manuscript culture. Medieval scholars divided the Bible into chapters and compiled concordances that let a reader find every occurrence of a word, and indexes, tables and running heads developed in manuscripts before they became standard in printed books. What print did was to make these tools cheap and uniform. Quotation becomes more exact, since every copy has the same words. Page references become possible across copies, since every copy has the same pagination. Narrative can be compared against itself, and readers can return to exact earlier wording.
That last point is especially important. Oral narrative relies heavily on recognitional returnability: a listener recognises an episode, a formula or a character when it recurs, but cannot go back to the exact words of an earlier performance. Print permits literal symbolic returnability. A reader can reopen exactly the sentence encountered fifty pages ago and read it again, word for word, and so can every other reader of the same edition.
The redistribution was never complete, and its gaps are instructive. The Victorian novels published in monthly or weekly parts, read over a year or more by audiences who could not easily turn back to an installment several months old, reintroduced characters with identifying traits, recapitulated earlier events and relied on recurring tags of speech and manner. Devices that look like survivals of oral technique returned where the material archive was, for practical purposes, out of reach. Where readers could not return, narrative again had to help them remember.
This changes memory again. The reader no longer needs to remember the whole text, because the book remembers part of it for the reader. Hence a formulation that captures much of the transformation: external archives permit internal forgetting. One can afford to forget because one can return. A reader of a long novel can let the details of the first chapter fade, confident that they are there to be recovered, and a writer can plant a detail that no reader will retain on first reading, confident that a second reading will find it.
XIII. Digital Narrative: When Forgetting Becomes the Scarce Capacity
Digital media intensify external memory radically. They offer cheap and potentially indefinite duplication, search across enormous bodies of text, replay of audio and video, automatic transcription, hyperlinks, version histories and archives held on distant servers. The redistribution of mnemonic labour here is so extensive that the problem itself changes shape. In oral traditions, the question was how enough could be retained. With writing and print, it was how enough could be stabilized, fixed and made reliably retrievable. With digital media, the question becomes how anything can be selected from what can all be retained.
Cheap retention should not be mistaken for durable preservation. Digital systems are spectacular at copying and poor at guaranteeing that anything will last. Platforms close, accounts vanish, formats become unreadable, links rot, storage fails and companies delete what they no longer find profitable to keep. What distinguishes the digital condition is cheap, potentially indefinite duplication combined with unprecedented searchable abundance, not perfect persistence. The paradox is sharper for that: digital systems make retention astonishingly cheap while making durable preservation, selection and recursive relevance into separate problems. Oral traditions faced all three as one problem, since what was told was kept and what was kept was told. Digital media pull them apart.
This suggests a historical reversal: the scarce resource shifts from retention to recursive relevance. When everything can remain available, the problem is deciding what deserves reopening. Viktor Mayer-Schönberger argued that for most of human history forgetting was the default and remembering the costly exception, and that digital technology has inverted the relation, so that forgetting now requires effort (Mayer-Schönberger 2009). The legal recognition in Europe of a limited right to have certain personal information delisted from search results is one sign of the inversion: forgetting has become something that must be secured.
An effectively limitless archive creates a new burden of selection. Search becomes a technology of recursive relevance, the means by which a present question reaches into the archive and brings back what bears on it. Recommendation systems become technologies of selective reopening, deciding which parts of the past a person will encounter again. Platforms that return a user's photographs and posts from years ago, on the anniversary of their making, have automated the reopening of personal pasts, so that what comes back is chosen by a system rather than by the one who remembers. Generative AI introduces a further transformation, because the retrieval, recombination and rearticulation of archived material become partly delegated to systems that can produce new narrative from the patterns of old narrative. A limited structural analogy with oral composition can be drawn, and it should be marked as limited. Both the singer and the computational system generate present outputs under constraints derived from enormous bodies of previous form. The mechanisms, however, are radically different, and the contrast is more instructive than the likeness. The living tradition has embodied stakes, teachers, audiences, recognition and ongoing interrecursivity: the singer answers to listeners who know the story and whose response shapes the next line, and the performance matters to people who are present. The computational system draws on a stock that is no one's tradition in particular, and no audience of knowers stands by, as a condition of its producing anything, to say whether the performance is true.
Digital media also display recombination vividly. They bring back voice, video, live performance and collective commentary, and in some respects they make narrative more oral again: podcasts and audiobooks restore the listening ear, livestreams restore co-present performance at a distance, spoken-word video restores the performer's body and timing, and serial narratives followed collectively online restore something like the knowledgeable audience that anticipates, corrects and demands. The newest mnemonic infrastructure revives some of the oldest.
The point of this section is limited, and should not be allowed to take over the argument. It is only that narrative memory has entered a phase in which the material and symbolic archive can exceed human mnemonic horizons by orders of magnitude. The labour that oral traditions spent on retention, and that literate cultures spent on stabilization, now has to be spent on durable preservation and above all on relevance: on deciding, among everything that persists, what should be made to matter again.
XIV. Narrative Memory and the Difference Between Preservation and Life
The history just traced could be read as a story of steady improvement in which each new medium preserves more, more exactly, for longer. That reading would be a technological determinism, and it would miss the distinction on which the whole account depends. Perfect preservation is not the same as living memory.
A forgotten file is technically retained and recursively dead. The clay tablets of the Mycenaean palaces survived because the fires that destroyed the palaces baked them hard, and for three thousand years they preserved their contents perfectly without anyone being able to read a word. The tablets of the Epic of Gilgamesh lay in the ruins of Nineveh for more than two thousand years before they were excavated, and when George Smith announced in 1872 that one of them told of a flood, a story that had been preserved all along became, in a moment, consequential again, for scholarship, for religion and for the public that crowded to hear of it. By contrast, a story told imperfectly in every generation may be materially unstable and recursively alive throughout.
The distinction can be formulated precisely: preservation is persistence without necessary reopening; memory is persistence capable of recursive reactivation. A narrative survives socially not because information about it still exists somewhere, but because somebody can make it matter again. Archive and memory, storage and remembering, availability and relevance, persistence and recursive life: in each pair the first term names a condition of the second, and never its substitute.
The singers Parry and Lord recorded offer a closer case. Their performances survive, on aluminium discs and later tapes, in the collection at Harvard that bears Parry's name, and they can be heard today with a fidelity no ancient listener could have imagined. The tradition that produced them, however, has largely ceased to be sung in the coffeehouses where it once lived. What has been preserved with great exactness is the trace of a living memory, and the living memory itself has thinned to near silence. The archive is invaluable, and it is not the tradition.
This reconnects the argument to oral tradition in an unexpected way. Oral traditions may have low exact textual persistence, since their words change with every telling, and extraordinarily high rates of active recursive reactivation, since a story that is not told regularly does not survive at all. Oral traditions impose a much tighter coupling between persistence and recurrent reactivation than external archives do, even though they too can carry formulas, epithets and archaic words whose original meanings have become obscure. Digital archives may loosen the coupling almost entirely: cheap retention of vast quantities, reactivation of a small fraction, and no guarantee that the rest will endure. The medium that makes retention cheapest may also be the one in which the largest proportion of what is kept is kept dead.
XV. Conclusion: The Story Remembers by Being Remembered
The argument began from a paradox: a narrative can carry a past only if the narrative itself remains carryable. In oral traditions, that carrying is distributed. Bodies carry rhythm, in voices trained to the shape of a line. Relationships carry authority and correction, in teachers, fellow performers and audiences who know the story. Places and seasons carry occasions, calling up the stories that belong to them. Objects carry cues, in instruments, halls, monuments and boards of beads. Symbols carry sequence, in formulas, scenes, genealogies and meters. No single mediation is sufficient, and their alignment is what makes a tradition robust.
This explains why meter, formula, audience, performance and transmission should not be treated as packaging around a narrative content that could exist without them. They are part of what made large narrative possible in the first place. The Odyssey and the Mahabharata could not have carried the pasts they carry had they not first been carried themselves, by an apparatus as elaborate in its way as the stories it sustained.
The history of that apparatus is a history of redistribution. Writing moved part of the burden of persistence into durable marks, so that stories could survive without continuous performance, at the cost of separating them from the bodies and relations that once kept them alive. Print made exact recurrence cheap, and allowed readers to forget what they could return to. Digital media have made retention cheap and relevance scarce, while separating both from durable preservation, so that the work of remembering now lies above all in choosing what to keep and what to reopen. None of these redistributions abolished the mediations that came before; each recombined them, and the oldest supports of memory, the voice, the audience, the occasion, keep returning to the centre. Through every transformation, the underlying process remains the same: narrative memory is constrained reconstruction, and a past persists only insofar as present configurations, bearing constraints left by what came before, allow it to become available again.
Narrative remembers by being rememberable. Oral traditions solved that problem by distributing remembering across bodies, relations, places, objects and symbols. Writing did not replace that architecture but redistributed its burdens, allowing stories to survive without continuous performance and thereby changing what narrative could afford to become. The history of narrative media is therefore less a history of forms for recording stories than a history of where societies place the work required to make absent worlds return.
Selected References
Arend, Walter. 1933. Die typischen Scenen bei Homer. Berlin: Weidmann.
Assmann, Jan. 2011. Cultural Memory and Early Civilization: Writing, Remembrance, and Political Imagination. Cambridge: Cambridge University Press.
Bartlett, Frederic C. 1932. Remembering: A Study in Experimental and Social Psychology. Cambridge: Cambridge University Press.
Basso, Keith H. 1996. Wisdom Sits in Places: Landscape and Language among the Western Apache. Albuquerque: University of New Mexico Press.
Bietti, Lucas M., and John Sutton. 2015. "Interacting to Remember at Multiple Timescales: Coordination, Collaboration, Cooperation and Culture in Joint Remembering." Interaction Studies 16 (3): 419–450.
Eisenstein, Elizabeth L. 1979. The Printing Press as an Agent of Change. 2 vols. Cambridge: Cambridge University Press.
Finnegan, Ruth. 1988. Literacy and Orality: Studies in the Technology of Communication. Oxford: Blackwell.
Goody, Jack. 1977. The Domestication of the Savage Mind. Cambridge: Cambridge University Press.
Hirst, William, and Gerald Echterhoff. 2012. "Remembering in Conversations: The Social Sharing and Reshaping of Memories." Annual Review of Psychology 63: 55–79.
Johns, Adrian. 1998. The Nature of the Book: Print and Knowledge in the Making. Chicago: University of Chicago Press.
Lord, Albert B. 1960. The Singer of Tales. Cambridge, MA: Harvard University Press.
Mayer-Schönberger, Viktor. 2009. Delete: The Virtue of Forgetting in the Digital Age. Princeton: Princeton University Press.
Ong, Walter J. 1982. Orality and Literacy: The Technologizing of the Word. London: Methuen.
Parry, Milman. 1971. The Making of Homeric Verse: The Collected Papers of Milman Parry. Edited by Adam Parry. Oxford: Clarendon Press.
Roberts, Mary Nooter, and Allen F. Roberts, eds. 1996. Memory: Luba Art and the Making of History. New York: Museum for African Art.
Rubin, David C. 1995. Memory in Oral Traditions: The Cognitive Psychology of Epic, Ballads, and Counting-out Rhymes. New York: Oxford University Press.
Sutton, John. 2026. "Distributed Remembering and Cognitive Philosophy." In Cognition, Culture, and Political Momentum: Breaking Down the Silos in Collective Memory Research, edited by Astrid Erll and William Hirst, 27–43. Oxford: Oxford University Press.