J.Konstapel, Leiden, 9-10-2026.

Jump to the book here

From Animal Call to Alphabet

A Structured History of Language — an essay on J. Konstapel, The History of Language: From Animal to Alphabet (Leiden, 2026)

1. Introduction

This essay summarises and evaluates a reference work on the development of language. The book was published in Leiden in October 2026. Its author is J. Konstapel of Constable Research. The work covers 157 pages, six parts and 28 chapters. It traces language from the alarm call of the vervet monkey to the 22 letters of the ancient alphabet.

The book has a distinctive method. It treats every form of language as a rung on a ladder. Each rung stands on the one below it. Every claim is backed by a count. The counts come from opened primary sources and from fresh computations on the PHOIBLE phonological database. Each chapter closes with a status section. That section separates what was counted, what was read, what rests on secondary sources, and what remains open.

Four results carry the book. First, the click inventories of southern African hunter languages are nearly full grids of articulation place times accompaniment. Second, every click series with two or more accompaniments has its parent in the same language: 77 of 77. Third, the ǀXam narrators gave each animal the human language with one fixed change on the click. Fourth, the road to the alphabet runs from full to thin — from a click that does several things at once to letters set one after another, and from 161 sounds in one language to 22 letters.

This essay follows the book’s own sequence. It states the theory first. It then reviews the animal evidence, the click evidence, spoken language, writing, and the alternative branches. It closes with an assessment and an annotated reference list.

2. The Theoretical Frame: Vacuum.Net

The book reads every fact through one theory, the Vacuum.Net theory. The theory is stated in one image, six terms and one rule.

The image is a fishing net without an edge. The net is one rope. Where the rope crosses itself, it forms a knot. Between the knots lie the meshes. A knot holds by its shape and needs no glue. A mesh is an opening with a form and a measure.

The basic unit is the trit. A crossing of the strand has three states: over (+1), under (−1), or no crossing (0). The zero is a lawful state. It stands for rest or passage, not for absence. One step of +1, 0 or −1 is called a turn.

The core rule is closure. Only what returns onto itself in phase persists. In the register of trits, a window of three consecutive crossings closes when its three values sum to zero modulo 3. Of 27 possible windows, 9 close.

From this follows the arithmetic of rings. Ring n is the set of all routes of n turns and holds 3ⁿ rooms. Ring 3 has 27 rooms. Ring 4 has 81. Every finite thing has exactly one address, computed from its content and never assigned. A text in a 27-sign alphabet is three trits per sign. A text of L signs has depth 3L.

The central instrument of the book is the parent rule: a form exists only on top of its parent. Each new form is the parent plus one turn. Turn zero is the parent itself. The step is expressed as x ↦ 3x + s. The child contains the address of the parent.

A second law runs beside this: the two-layer law. Every closure has a frozen address layer that counts in threes, and a tension layer where anything happens, counted in twos. The book applies these terms to whales, clicks, scripts and alphabets. Two cautions apply. The net sections state what is found. They do not claim causal arrows. And the parent rule is counted only for sound systems; for scripts it remains a reading.

3. Part I — The Animals

Language did not begin with people. The book sets out the animal forms as rungs, each supported by playback experiments.

The call. Vervet monkeys give acoustically distinct alarm calls for leopards, eagles and snakes. Playback tests from 1980 show that the sound alone produces the escape response. Infants overgeneralise at first. The classification sharpens with age. Gunnison’s prairie dogs add a shading: statistical sorting of calls placed the intruder’s shirt colour correctly in 54.9% of cases, against a 33.3% chance level.

The affix. Campbell’s monkeys keep six loud call types: boom, krak, krak-oo, hok, hok-oo and wak-oo. The ending “-oo” widens the meaning. Krak means leopard. Krak-oo means general alert. The study rests on about 2,000 hours of observation and 1,067 analysed calls. One cell of the 3×2 grid is empty: bare wak does not occur. Its parent exists only inside the child.

Order and sequence. Putty-nosed monkeys combine pyows and hacks into a sequence that means neither leopard nor eagle. It means “the troop moves”. It works only from the group’s own male. Japanese tits join the call ABC (scan for danger) with D (approach). Listeners do both. Reversed as D–ABC, the effect largely disappears. Chimpanzees produce 390 ordered sequences from 12 units. Bonobos show extensive compositionality.

Grids, names, songs. The sperm whale coda is the strongest case. Sharma and colleagues analysed 8,719 codas from clan EC-1 in the Eastern Caribbean. A coda is not one item in a list of types. It is a point in a grid of four features: rhythm (18 states), tempo (5), rubato (3) and ornamentation (2). The grid holds 540 cells. At least 143 are in frequent use — about 26%. Other work finds vowel-like categories inside the single click.

Names exist without words. Bottlenose dolphins carry a signature whistle for life. Others copy it to address the bearer. African elephants address one another with individually specific rumbles that are not imitations. Marmoset families share vocal labels for the same receiver, across 53,993 recorded calls.

Song adds depth and learning. Humpback song nests over five levels: unit, sub-phrase, phrase, theme, song. When cut by the method human infants use to find words, its pieces show Zipf’s law and Zipf’s law of brevity. Nightingales hold about 186 song types per male, kept for years. Bengalese finches need roughly twice as many states as syllables to describe song order: the same sound stands at more than one place.

Map and bridge. The honeybee waggle dance gives direction and distance at once. The angle to vertical on the comb equals the angle to the sun. The waggle duration encodes distance, with a measured bend near one kilometre. The dance stands for something absent. It is the first map. The honeyguide gives the first bridge between species. The Yao call “brrr-hm” raises the chance of being guided from about 33% to about 66%. Birds answer the local human signal, not the foreign one. The exchange is cultural on the human side and fitted to it, region by region.

4. Part II — The Hunter and the Click

The click as a sound. A click has two closures in the mouth. The enclosed air is thinned, and the front closure opens. The sound needs no air from the lungs. Its airstream is lingual and ingressive. There are five places of articulation: bilabial, dental, lateral, alveolar and palatal. The four tongue clicks can be laid out on two two-way distinctions — sharp or noisy release, high or low sound. Four places are 2×2. The lip click stands apart. Five is four plus one.

The grid. A click language holds a grid of places times accompaniment series. The book computed fresh counts on PHOIBLE 2.0 (3,020 inventories, downloaded 8 October 2026). Clicks occur in 21 inventories covering 19 languages. The four largest “ph” inventories give the headline result:

Language	Places × Series	Cells	Filled	Share
!Xóõ	5 × 17	85	83	98%
!Xun	4 × 13	52	48	92%
Xhosa	3 × 7	21	18	86%
Sandawe	3 × 5	15	15	100%

A !Xóõ speaker does not learn 83 separate sounds. The speaker learns five places and seventeen releases: 22 choices, not 83. The full inventory of 161 segments is nearly five times the world median of 33.

The parent rule tested. The rule is concrete for clicks. Take a series with two or more accompaniment marks. Remove one mark. The parent must stand in the same language, at the same place. The computed test on PHOIBLE found no exception: 77 of 77 series with two or more marks have their parent present, against a chance level of 87.9%. The chance level is high because the grid is nearly full. The book states this plainly. The margin is twelve points, and the series are not fully independent.

The stronger evidence comes from languages without clicks. Across 3,001 inventories, a sound with two modifier signs has its one-sign parent in its own inventory in 68.7% of cases. In a random other inventory the figure is 18.5%. At three or more signs the figures are 65.7% against 1.7% — nearly forty times chance. The share at home stays near two thirds at every depth. The deeper the sound, the less its parent can be found by accident.

The speech of animals. In the 1870s, ǀXam narrators dictated stories to Wilhelm Bleek and Lucy Lloyd in which animals speak ǀXam. Each animal has one fixed change on the clicks. The Tortoise turns clicks into labials. The Ichneumon turns them into sibilants. The Moon, the Hare and the Anteater replace all tongue clicks with one click and keep the lip click. The Blue Crane keeps the clicks and adds a “tt” after the first syllable. This is the parent rule performed in the 1870s: the words stay, and one turn marks the speaker. Bleek inferred from the Jackal’s “strange labial click” that the language once had more than five clicks. In the book’s terms: rooms that daily speech no longer inhabits are inhabited by the animals.

How old, and where the clicks went. The book is disciplined on dating. “Khoisan” is not a proven family. Genetic studies give dates for populations, not for sounds: the deepest African divergences run to about 300,000 years. Three positions on click age stand side by side. Knight and colleagues hold that clicks are inherited from early speech, and note themselves that the dating is the most tentative part. Güldemann holds that clicks spread by contact, often through marked speech registers — Nguni avoidance speech is his example. Xhosa today has 15 or 18 clicks in a family whose ancestor had none. Du Plessis holds that clicks are innovations from ordinary consonants. The sources record no dated step from click speech to non-click speech. The rung is structural, not dated.

5. Part III — Speech, Tone, and Language Without the Mouth

The body of speech. The descended-larynx theory of Lieberman has been contested on three fronts. Deer pull the larynx down to roar. The hyoid bone does not indicate larynx position. Baboons produce vowel-like contrasts. The Kebara 2 hyoid, about 60,000 years old, shows inner bone stress consistent with speech. Its own authors do not claim proof. Proposed dates for speech differ by a factor of four hundred across living authors. The book prints the range and names each holder.

The sounds of the world. In WALS, consonant inventories run from 6 (Rotokas) to 122 (!Xóõ); the median is 21. In PHOIBLE the median whole inventory is 33 segments; 72.3% of inventories fall between 14 and 40 segments. Only 21 of 3,020 exceed 81. Inventories are built from few features, each used many times — Clements’ feature economy. A fresh count shows that the number of features needed to keep all segments apart stays near three times log₃N, from 33 segments to 157.

Tone. In the WALS sample, 41.8% of languages are tonal. Maddieson himself states this underrepresents the true share. Tone adds a second line to the sound. Three tones multiply the syllable by three. In PHOIBLE, 636 inventories list tone — a lower bound, not a share.

Language without the mouth. Whistled languages transpose ordinary speech onto one line of pitch. A language without tone whistles its vowels; a tonal language whistles its tones. The whistle travels over a kilometre, against about 40 metres for ordinary speech. Bora drum language reverses the usual account. With two drum tones, rhythm — not tone — carries most of the distinction: 7 rhythm patterns against 2 tone patterns in 18 drummed names. Kele drumming replaces each word with a fixed long phrase, roughly eight drum words for one spoken word. Sign languages are languages in their own right, with parts — Stokoe’s cheremes — and emergent structure in new communities.

The analysed language. Sanskrit closes the spoken rungs. Pāṇini’s Aṣṭādhyāyī holds about 4,000 rules, recitable in about two hours. Its sound order is a 5×5 grid of places times manners, all 25 cells filled. Staal compared it to the periodic table. The Ṛgveda — 1,028 hymns, 10,600 verses — was carried by voice for centuries, through eleven recitation patterns, several of them reversing the word order. Sound symbolism is not wholly arbitrary either: “bouba” matches the round shape at 72% across 25 languages, and certain sound-meaning pairings recur independently across word lists covering two thirds of the world’s languages.

6. Part IV — From the Mark to the Alphabet

Marks before writing. Engraved ochre from Blombos Cave dates to between 100,000 and 72,000 years ago. In Ice Age Europe, 32 geometric signs recur across hundreds of caves over 30,000 years — a small, stable stock. Bacon and colleagues read line, dot and Y as a phenological calendar: 862 sequences, none longer than 13 marks. Whether it is writing is left open. Counting tokens in the Near East precede the first tablets. The numerical half of Schmandt-Besserat’s theory is accepted. The visual half is not: Zimansky’s review found 58% of token subtypes attested once or not at all.

Sumer. Proto-cuneiform appears at Uruk around 3350–3000 BC. The corpus holds about 6,700 artefacts. The sign count depends on the counting rule: 1,990 sign names, 1,299 after merging variants, 705 after splitting compounds. Of Uruk III texts, 85.6% are administrative and 14.4% are lexical lists — the school texts that passed the script on. The language of the first tablets is not proven. A list of things can be read in any language. The rebus changed that: the arrow sign, ti, came to write “life”, til, and then the syllable ti anywhere. With the sound, the reader’s tongue was fixed. The stock of signs then fell from about 1,500 to about 600, with about 400 in working use. Fewer signs, longer rows.

Egypt. The Abydos labels of tomb U-j (about 3300–3200 BC) suggest Egyptian writing used sound signs from its first known day. The script holds three kinds of sign at once: the picture, the consonant sign (one, two or three consonants), and the silent determinative that marks the class. The house sign writes “house” (pr), the sound pr in “go out”, and takes walking legs as its class sign. The 24 uniliteral signs stood inside the large script for three millennia before anyone used such signs alone. Between 1% and 5% of the population could read.

The first alphabet. The alphabet was made once. Semitic speakers, working for Egyptians, took a few dozen Egyptian picture signs and read each by the first sound of its name in their own tongue — acrophony. The house sign, pr in Egyptian, is bayt in Semitic; the makers kept the b. Twelve everyday things made twelve consonants. The date is disputed across six centuries, from the 19th to the 13th century BC. The count is 22 or at least 27. Goldwasser holds the makers were Canaanite miners who could not read Egyptian; Rainey holds they were sophisticated scribes writing on lost papyrus. New finds — the Lachish sherd of the 15th century BC and the 17-letter ivory comb — narrow the gap. On the ring reading, an alphabet of 27 signs fills ring 3 exactly: 27 rooms, three trits per sign.

Consonants to vowels. Phoenician fixed the consonant alphabet at 22 letters. The Greeks took the letters around the early 8th century BC and gave the Semitic throat letters to vowels. Greek is the first script that writes vowels as a rule. The Latin alphabet grew from 21 to 26 letters over some 2,300 years; W, U and J are each a parent letter plus one turn. The letter names and the letter order are fossils: ox, house, throw-stick, door, hand — still in sequence after some 3,400 years, frozen as address. Side branches (runes, Ogham, the Aegean syllabaries) and the numerals stayed on their own roads. The digit 5 is read in every language and spelled in none: the number sign never took the turns from image to sound.

7. Part V — The Other Branches

China. Chinese writing is the image-sound rung, deepened. The oracle bones of Anyang (about 1250–1050 BC) hold about 150,000 inscriptions and about 4,000 characters, of which half or more cannot be read today. Nearly all later growth took one form: a semantic part joined to a phonetic part. The Kangxi dictionary of 1716 files about 47,000 characters under 214 radicals. Modern lists cap use far lower: 8,105 characters in the 2013 standard, 3,500 in tier 1. Mandarin has about 1,300 tonal syllables: 3,500 characters against 1,277 syllables is 2.7 characters per spoken syllable. The script keeps apart in the eye what the ear hears as the same. DeFrancis’ term is morphosyllabic: a sound system, not an idea system.

Korea. Hangul was made on purpose and by a known person. King Sejong’s script was promulgated in 1446. The consonant letters are drawings of the speech organs; stronger sounds add strokes. The letters stack into 11,172 possible syllable blocks, of which about 2,000 are in common use — an 18% inhabited grid.

India. The Indus signs (about 2,905 objects, 417 distinct signs, average length 4.6) may or may not be writing; Farmer and Rao disagree on the evidence. Brahmi and Kharosthi appear in Ashoka’s edicts, about 1,650 years later. The abugida principle keeps the syllable whole: each consonant carries a built-in vowel, other vowels are marks, and the virama removes the vowel. From Brahmi descend nearly all scripts of South and Southeast Asia.

The Americas. Mesoamerica invented writing independently. Maya is the only American script that is read. Its syllabary is a grid: about 200 signs fill about 80 consonant-vowel cells — 2.5 signs per syllable. The decipherment turned on Knorozov’s 1952 paper, which showed that Landa’s “alphabet” was a syllabary. In the Andes, khipus record in knots and coloured fibres; the Collata khipus, with 95 symbols, are the strongest case that cords could record names. Rongorongo remains unread; the wood of one tablet dates to 1493–1509.

Scripts made in one lifetime. Six modern cases show what people reach for first. Sequoyah, illiterate, made an 86-sign syllabary for Cherokee in 1821. Eight illiterate Vai men made a syllabary of about 200 signs; over 171 years its letters grew steadily simpler, repeating the evolution of older scripts. King Njoya of Bamum went from 465 word-signs to 80 syllable signs in 21 years. Only the literate makers reached for the single sound: Kanté’s N’Ko alphabet of 27 letters. No maker without letters reached for the phoneme.

The kinds of writing. Four inventions are accepted as independent: Mesopotamia, Egypt, China, Mesoamerica — with the Indus signs and rongorongo open. Between 175 and 324 scripts are known. About half of the world’s roughly 7,170 living languages are written. The other half is not small: for at least 719 languages it is known that they are not written at all. Speech came first, by a very long span.

8. Part VI — The Whole Tree

Chapter 27 draws the tree. Rungs 2 to 8 lie among the animals: call, affix, order, grid, name, song, map. Rung 9 puts the full grid in one mouth: the click. Rungs 10 to 13 cover speech, tone, mouthless carriage and the mark. Rung 15, image-sound writing, is the last rung before the tree branches. Each branch is a choice of what to keep and what to drop at that rung. Chinese dropped nothing. The syllabaries dropped the image. The abjad dropped the vowel. The alphabet kept every single sound. Hangul drew the movement of the mouth itself. The khipu took neither image nor sound: strand, knot, position.

Four things hold from the first rung to the last. The parent rule: a form exists only on top of its parent, and the parent often stands inside the child. The grid: independent choices, nearly every combination inhabited. The move from at-once to one-after-another: the click does several things in one sound; the letter does one thing, in a row. And the move from full to thin: 161 sounds, then 33, then 22 letters.

9. Assessment

The book’s strengths are clear.

• Discipline about evidence. Every chapter separates counted, read, secondary and open. Where sources disagree, the range is printed with the name of each holder. The author does not hide the weaknesses of his own headline results, such as the high chance level in the click parent test.
• Fresh computation. The PHOIBLE counts — the click grids, the parent test on 3,001 inventories, the tone distribution, the feature economy — are new work, documented in appendices.
• Synthesis. Setting the whale coda grid beside the click grid beside the Maya syllable grid beside the Sanskrit 5×5 table is a genuinely illuminating comparison.
• Human attention. The “people” sections name narrators, speakers, scribes and last speakers. Katrina Esau, born 1933, is the last first-language speaker of Nluu. The book counts people as well as sounds.

The limits should be stated with equal candour.

• The Vacuum.Net theory does most of the interpretive work. Readers who do not accept the net, the trit and the ring will still find the counts useful. But the “in the net” readings are exactly that — readings, not findings.
• The parent rule is counted for sound systems only. For scripts it is a reading. A count over script sign lists has not been made.
• Some figures rest on secondary sources, chiefly Wikipedia, and the book says so itself. A reader who needs a figure for further work must go to the primary source named.
• The book treats its rungs as structure, not as dated events. The step from click speech to breath speech, in particular, has no dated evidence anywhere in the sources. The book says this plainly. It is to its credit.

As a business-style summary: the work is a well-documented, quantified survey of language history, with a strong original dataset on click phonology and a consistent, if unproven, theoretical overlay. Its reference apparatus is unusually honest. Its central empirical claim — that sound systems build every marked form on a simpler parent in the same language — is supported by a large-scale test at depths where chance cannot explain the result.

Annotated References

A selection of the sources used in the book, with the reason each is worth reading. The annotations follow the book’s own reference style.

1. Seyfarth, R. M., Cheney, D. L. and Marler, P. (1980). “Monkey responses to three different alarm calls.” Science 210: 801–803. The founding playback study. Three calls, three responses, and infants who learn the classes. Every later animal study repeats this method.
2. Ouattara, K., Lemasson, A. and Zuberbühler, K. (2009). “Campbell’s monkeys use affixation to alter call meaning.” PLoS ONE 4(11): e7808. The six calls and the ending “-oo”. Shows affixation in a wild primate, with the telling absence of bare wak.
3. Arnold, K. and Zuberbühler, K. (2006). “Semantic combinations in primate calls.” Nature 441: 303. One page. The first report that two calls in sequence mean something other than either call alone.
4. Suzuki, T. N., Wheatcroft, D. and Griesser, M. (2016). “Experimental evidence for compositional syntax in bird calls.” Nature Communications 7: 10986. The Japanese tit’s ABC–D against D–ABC. Order as meaning, tested by playback.
5. Sharma, P. et al. (2024). “Contextual and combinatorial structure in sperm whale vocalisations.” Nature Communications 15: 3617. Turns the list of coda types into a grid of four features. The book’s whale-grid chapter rests on it. Open access.
6. Oren, G. et al. (2024). “Vocal labeling of others by nonhuman primates.” Science 385: 996–1003. Names in a monkey: 53,993 marmoset calls, with labels shared inside the family.
7. Pardo, M. A. et al. (2024). “African elephants address one another with individually specific name-like calls.” Nature Ecology & Evolution 8: 1353–1364. Names that are not imitations of the bearer. Numbers are from the preprint; a correction appeared in July 2025.
8. Kohl, P. L. and Rutschmann, B. (2021). “Honey bees communicate distance via non-linear waggle duration functions.” PeerJ 9: e11187. Fresh measurements of the dance at nine distances. The relation of duration to distance bends.
9. Spottiswoode, C. N., Begg, K. S. and Begg, C. M. (2016). “Reciprocal signaling in honeyguide–human mutualism.” Science 353: 387–389. The test of the Yao honey call. Doubles the chance of being guided.
10. Knight, J. et al. (2003). “African Y chromosome and mtDNA variation provides clues about the history of click speakers.” Paper with nine co-authors. The genetic comparison of Hadzabe and Ju|’hoansi. The basis, and the weakest point, of the “clicks are ancient” position.
11. Güldemann, T. (2007). Work on the spread of clicks by contact. The position that clicks entered Dahalo and the Bantu languages through contact, often through marked speech registers. Read together with Knight et al.
12. Bleek, W. H. I. (1875). A Brief Account of Bushman Folklore and Other Texts. The printed source for the speech of animals: seven speakers, each with one fixed change on the clicks, and the inference that ǀXam once had more than five clicks.
13. Moran, S. and McCloy, D. (eds.) (2019). PHOIBLE 2.0. Max Planck Institute for the Science of Human History. The database on which the book’s own counts were computed: 3,020 inventories, 105,484 segments.
14. Clements, G. N. (2003). “Feature economy in sound systems.” Phonology 20(3): 287–333. Inventories are built from few features, each used many times. The book extends this with its own feature-needed counts.
15. Seifart, F. et al. (2018). “The Bora drummed speech allows low-error message transmission over long distances.” Paper on Bora manguaré drumming. The best-measured drum language: 169 messages, and rhythm — not tone — carries most of the distinction.
16. Henshilwood, C. S. et al. (2002). “Emergence of Modern Human Behavior: Middle Stone Age Engravings from South Africa.” Science 295: 1278–1280. The first two engraved ochres from Blombos, about 75,000 years old. Marks before writing.
17. Bacon, B. et al. (2023). “An Upper Palaeolithic proto-writing system and phenological calendar.” Cambridge Archaeological Journal 33(3): 371–389. The reading of line, dot and Y as a calendar of animal life. Read the authors’ own caveats.
18. Zimansky, P. (1993). Review of Schmandt-Besserat, Before Writing. Journal of Field Archaeology 20(4): 513–517. Five pages that test the token theory by counting. Essential against the theory’s numerical half.
19. Born, L. and Kelley, K. (2021). “A Quantitative Analysis of Proto-Cuneiform Sign Use.” Cuneiform Digital Library Bulletin 2021:6. Shows on one dataset how the sign count moves from 1,990 to 705 with the counting rule.
20. Goldwasser, O. (2010). “How the Alphabet Was Born from Hieroglyphs.” Biblical Archaeology Review 36(2): 40–53. The main statement of the miners’ view of the alphabet’s origin. Read beside Rainey’s rebuttal.
21. Höflmayer, F. et al. (2021). “Early alphabetic writing in the ancient Near East: the ‘missing link’ from Tel Lachish.” Antiquity 95(381): 705–719. The Lachish sherd in its radiocarbon context. Narrows the gap between Sinai and the later Levant.
22. DeFrancis, J. (1989). Visible Speech: The Diverse Oneness of Writing Systems. University of Hawaii Press. Source of the character-structure counts and the term “morphosyllabic”. The case that Chinese characters write sound, not ideas.
23. Farmer, S., Sproat, R. and Witzel, M. (2004). “The Collapse of the Indus-Script Thesis.” Electronic Journal of Vedic Studies 11(2): 19–57. The case that the Indus signs are not writing. Read against Rao et al. (2009), who measure language-like entropy in the same corpus.
24. Kettunen, H. and Helmke, C. (2011). Introduction to Maya Hieroglyphs. Wayeb. The Maya sign counts and the syllable grid. Freely available.
25. Hyland, S. (2017). “Writing with Twisted Cords: The Inscriptive Capacity of Andean Khipus.” Current Anthropology 58(3): 412–419. The strongest case that khipus could record names: two khipus from Collata with 95 symbols.
26. Kelly, P., Winters, J., Miton, H. and Morin, O. (2021). “The Predictable Evolution of Letter Shapes.” Current Anthropology 62(6): 669–691. A measured 171-year history of the Vai script: visual complexity falls steadily. New scripts repeat the evolution of old ones.
27. Konstapel, J. (2026). The Vacuum.Net Theory: Foundational Paper, edition 7. constable.blog. The book’s theoretical source: the five ground rules and the unique-address theorem. Read the ground rules N1–N5 and theorem T7.

Note on sourcing: the book itself distinguishes primary works it opened, works read in abstract only, and facts taken from secondary sources such as Wikipedia articles that name their primary references. The annotations above preserve those distinctions where they matter. A reader who needs a figure for further work should consult the primary source named in the book’s reference sections for each part.

From Animal Call to Alphabet