Objects That Remember

Core Concepts

Semasiography: a third category

Western classification of writing has long been organized around a single axis: does this system encode the sounds of spoken language? Systems that do—whether alphabetically (phonemes), syllabically (syllables), or through morphosyllabic combination like Chinese (meaning + pronunciation)—qualify as "full writing." Everything else falls below the line. Scholars including DeFrancis and Sampson established phonetic encoding capacity as the primary dividing criterion between writing and proto-writing, and this framework has dominated academic typology for decades. The consequence: mnemonic object systems like khipu, wampum, and winter counts were systematically classified as less-than.

Frank Salomon's work on Andean khipu proposed a way out of this binary. He argues that khipu constitute a semasiographic system—a non-phonetic symbol system analogous to musical notation, mathematical formulas, or chemical diagrams. Semasiography conveys meaning without being tethered to the speech sounds of any particular language. This frames khipu not as failed writing but as a third category: distinct from both phonetic script and mere mnemonic device.

The practical implication of semasiography is significant. A semasiographic system can serve multilingual populations (as the Inca empire required), survives translation, and encodes meaning through inherently non-verbal codes—materials, colors, weaves, spatial arrangements—that function independently of any spoken tongue.

The question is not whether a system is writing. The question is what kind of information technology it is, and what it was designed to do.

The paradigm shift

The treatment of material mnemonic systems as "insufficient" for real documentation was not a neutral finding. It was a theoretical commitment rooted in what Western scholarship was epistemologically prepared to recognize as writing. From 1900 to 1970, khipus were consistently dismissed as mnemonic devices—at best, accounting tools. The theoretical space to reconsider this did not exist until the 1986-onward revaluation of non-phonetic writing systems by theorists like Roy Harris and Stephen Sampson, which created room for plural classification frameworks beyond alphabetic supremacy.

This opened the door for the post-1990s archival recovery of Andean materials and more recent AI-assisted pattern analysis, which have begun to frame some khipus as possessing narrative recording capacity—not just accounting data. The historiographic lesson is pointed: new evidence alone rarely produces a paradigm shift. The shift required new frameworks that made the evidence intelligible.

Distributed cognition: memory beyond the skull

The dominant Western model of memory treats it as something that happens inside an individual mind. Mnemonic object systems force a different model. The concept of distributed cognition holds that cognitive processes—including memory—can be distributed across a person, an artifact, and a context. The object does not supplement memory; it is part of the memory system.

The Luba lukasa (memory board) of Central Africa makes this concrete. The lukasa is read not by silent visual inspection but by fingertip contact: the trained practitioner (a member of the Mbudye society, called a bana balute) runs their fingers across the surface while reciting. Tactile contact activates recall. Meaning is stored in the interaction between the board's bead configurations, the trained body of the practitioner, and the ritual context of performance.

This means the cognitive system has three components, none of which is sufficient alone:

• The material object (bead configurations, spatial arrangement on the board)
• The trained practitioner (embodied kinesthetic knowledge, memorized associations)
• The performative context (the recitation setting that activates and structures the recall)

Remove any one element and the system fails. The bana balute without the board cannot access what the board holds. The board without the bana balute is mute. A lukasa held by someone untrained is literally illegible.

A single lukasa, through this three-part system, can enable recitation of approximately 200 years of historical detail—genealogies, king lists, migration routes, territorial boundaries, court protocols—that would be nearly impossible to retrieve without the mnemonic interface. The board encodes both temporal history and spatial geography simultaneously: bead configurations mark locations of royal compounds, clan migration routes, and territorial boundaries, making it function as both historical record and cognitive map.

The coupled oral-object model

If distributed cognition describes the cognitive architecture, the coupled oral-object model describes the communication architecture. In most mnemonic object systems, the object and the oral performance it generates are not separable. Neither alone constitutes the full record.

Australian message sticks illustrate this with precision. The carved markings on a message stick did not contain the full message. They served as mnemonic anchors that prompted the messenger's memorized oral statement. The messenger-object pairing was the complete communication technology. The authority of the stick (its material authentication) and the authority of the memorized oral content were inseparable components of a single message.

The same structure appears across the Luba lukasa: Luba royal history is performed orally, not read silently. The board anchors and triggers the performance; the performance is the record. This is not a failure to achieve alphabetic literacy. It is, as scholars have argued, an intentional technological choice aligned with Luba cultural values of oral transmission and ritual knowledge-making.

The coupled model has a corollary: when you separate the object from the oral tradition that activates it, meaning is not merely reduced—it is destroyed. This is why so many mnemonic objects in museum collections remain "undeciphered." They were never designed to be read by outsiders without institutional access to the oral component.

Keeper traditions

Every mnemonic object system depends on what we can call a keeper tradition: a structured social role (or set of roles) responsible for maintaining the institutional knowledge needed to activate the system. The keeper is not a passive archivist. They are a living component of the information technology.

The khipu example is instructive here too. Frank Salomon's 1994 observation in Tupicocha, Peru, and Sabine Hyland's collaborative access to khipus maintained in San Juan de Collata both demonstrate that khipus remain actively preserved and used as civic and ritual objects by contemporary Andean communities. These are not fossils. The keeper traditions survived colonialism, and Indigenous communities remain the authoritative interpreters of their own knowledge systems. External decipherment projects that treat khipus as archaeological puzzles to be solved without Indigenous collaboration miss the living half of the coupled system.

Wampum belts among the Haudenosaunee involve a parallel structure. The standardized bead patterns encode treaty terms, historical narratives, and relational obligations according to a system comprehensible to those trained in wampum language. Indigenous scholar Angela Haas has characterized wampum as an early form of hypertext literacy—a sophisticated, non-linear information system where embedded symbols enable multiple interpretive pathways and connect related concepts. This reframing positions wampum not as an ethnographic curiosity but as a complex literacy technology with specific cognitive affordances distinct from alphabetic writing.

Keeper traditions are also materially embedded. Lakota winter counts were recorded on buffalo hide until the late 19th century, when keepers transitioned to muslin and ledger paper as buffalo populations collapsed under colonial hunting pressure. The material shift was not stylistic preference; it was adaptation to ecological destruction. Colonial history is written into the physical form of the records themselves.

Analogy Bridge

Thinking of a mnemonic object as a kind of "primitive book" will mislead you. A better analogy: think of a musical score.

A score does not contain the music. It is not music at all in isolation. The score encodes enough information for a trained performer—someone with years of embodied practice—to reconstruct and perform the work. Without a trained performer, the score is marks on paper. Without the score, the performer improvises (or relies on memory alone, which degrades). Together, score and performer produce something neither can produce alone.

Now add one more feature: the score is not designed to be read silently. It is designed to be performed—and the performance is the point. The score-as-artifact and the performance-as-event are both components of the same communication technology. Neither is the "real" thing. Both are required.

Mnemonic objects work the same way. They are not degraded writing. They are score-like technologies optimized for oral performance, distributed memory, and keeper-mediated transmission. Evaluating them against alphabetic literacy standards is like evaluating a musical score by asking whether it reads well as a novel.

Compare & Contrast

The classification column should be read critically: it reflects the Western typological framework, not an objective hierarchy of sophistication. A wampum belt encoding treaty obligations across generations in a standardized, institutionally-maintained system is not less sophisticated than alphabetic writing. It is differently sophisticated, optimized for different conditions and values.

Note also that classification is not always cleanly resolved. Some Australian message sticks appear to have employed the rebus principle—encoding linguistic sound in addition to semantics—which would satisfy strict definitions of writing systems rather than merely notation systems. Whether a given object qualifies as "writing" depends not only on the object's encoding strategy but on which definition of writing you apply.