On naYana, and why English spelling is the largest unacknowledged tax on human attention
An eight-year-old girl I know writes Hindi in the Roman alphabet. She has no training in transliteration. She has no teacher correcting her spelling. What she has is an ear, and a set of letters she learned in school. The letters represent sounds, the sounds represent words, and so she writes. Her spellings are her own, and they are perfectly consistent. Mausam, paani, khushi. If you can read English letters and you know Hindi, you can read everything she writes.
A few weeks ago she asked me what I do for a living. I told her I work in philosophy. She wrote it down in her journal: filosafi.
She was, of course, wrong about the spelling. She was also, just as clearly, right about everything else. She had heard a word, identified its sounds, and assigned each sound the nearest letter she knew. This is what every literate child in the world does, except for one population. Children learning to read English do it once or twice in kindergarten and then spend the next several years learning that what they did was wrong. The word is philosophy. The reasons are historical. The teacher will not explain them, because there is no time, and because the teacher does not know them either.
This essay is about what that time costs, why every previous attempt to recover it has failed, and why we believe that, after a century of failures, a different approach is now possible.
English spelling is the single largest learnable inconsistency in human literacy. The numbers are not subtle.
Children learning to read in Finnish, Italian, Spanish, Korean, or Indonesian reach reading fluency in roughly one to two years. Children learning English take three times as long.1 This is not a difference of effort or pedagogy or intelligence; the gap holds across socioeconomic backgrounds, educational systems, and teaching methods. It tracks one variable: the regularity of the orthography — what linguists call orthographic depth.2
Finnish has roughly one spelling per sound and one sound per spelling. Indonesian was rationalized in the twentieth century and is similarly regular.3 Spanish has small irregularities at the edges (the silent h, the b/v merger), and a child can decode 95% of words by the end of first grade.4 Hindi in Devanagari is fully phonetic; once a child knows the consonants and the vowel marks, every new word is readable on first sight.
English is none of these. The letter a represents at least seven distinct vowel sounds (cat, father, care, about, late, all, any). The sound /f/ is spelled at least four ways (fish, phone, enough, off). The letter sequence ough is pronounced seven different ways (though, through, tough, cough, bough, thorough, hiccough).5 There are no rules that resolve these. The child must memorize each word. A literate English speaker has spent years of childhood building a mental lookup table.
The cost of that lookup table is not just instructional. It is opportunity.
Consider what a Finnish child does in the two years that an English child is still learning to spell. She reads stories. She writes her own. She participates in the literate culture of her community as a full member at the age of seven. Her English counterpart, at the same age, is still being told that write and right are different words even though they sound identical, that knight contains a silent letter she has no way to predict, that island and isle contain an s that nobody pronounces because of a sixteenth-century Latin misattribution.6
By age eleven, the Finnish child has read perhaps three hundred books. The English child has read fewer. By age eighteen, the gap has narrowed only because the English child has grimly memorized the lookup table. But the years of reading lost cannot be reclaimed. The cultural participation she could have had — the early entry into the conversations of literate humans, the formation of identity through text — was deferred. For some children, who could not bear the opacity of English spelling and were diagnosed as dyslexic7 or simply gave up, it was deferred permanently.
The numbers are catastrophic. English is the first language of about 400 million people and the second language of perhaps another 1.5 billion.8 If even one year of childhood is lost per child to spelling mastery — a conservative estimate — and that year contains roughly 1,000 hours of school instruction, the total annual cost across the world is on the order of two trillion hours.9 The dollar cost, valued at the prevailing wage of teaching time alone, runs into the hundreds of billions per year. The energy cost, valued in the electricity to light the classrooms and the food to fuel the children, is its own catastrophe, multiplied across decades.
But the cost that matters is none of these. The cost that matters is the creative time — the time a child of seven could be spending on stories, on music, on building things, on absorbing the literate culture she was born into — that she instead spends decoding the spelling history of a language that has neglected its own writing system for five hundred years.
We are asking children to pay, in childhood, for the failures of dead typesetters and pedants. The children pay. We do not even count.
This problem is not new. It has been seen, named, and attacked repeatedly. Each attempt has failed for reasons worth understanding, because their reasons remain instructive.
George Bernard Shaw, in his will, left a substantial sum for the design of a new English alphabet — what became the Shavian alphabet, published in 1962.10 Shavian was technically sound: a phonetic script with distinct, simple shapes, designed by a typographer who knew what he was doing. It was used to print exactly one book: a parallel-text edition of Androcles and the Lion. It was never adopted by anyone. It had no place to live. The technology of 1962 could not deliver a typeface to a reader without printing presses, and printing presses do not adopt new alphabets on a whim.
The Deseret alphabet of the nineteenth-century Mormon settlers was a similar effort, similarly doomed by its dependence on infrastructure that would not be repurposed.11 Type was cast. Books were printed. A generation of children was taught to read it. Then the church moved on, the type was melted, and the alphabet went extinct within fifty years.
The Initial Teaching Alphabet (ITA) of 1960s Britain came closer.12 ITA was used in primary schools to teach early reading: forty-four characters, each representing one sound. Children learned to read ITA quickly — much faster than traditional spelling — and then were transitioned to standard English. The program ran for years and was eventually abandoned. The problem turned out to be the transition. Children who learned to read ITA had a harder time, not easier, switching to standard English than children who had been taught traditionally from the start.13 The crutch became an obstacle. The new alphabet had not failed to teach reading; it had failed to let go.
More recent efforts — SoundSpel, Cut Spelling, SaypU, Unifon14 — have all foundered on the same rock: the network effect. Any new spelling exists in a world where every existing book, sign, contract, website, and trained reader is in the old spelling. To adopt the new script is to lose access to everything written before, and to be unreadable to everyone who has not adopted it. The benefit is small and individual; the cost is large and immediate. Rational individuals do not adopt. The reform dies.
There is a deeper pattern across these failures. They all asked the reader to choose between two scripts at a moment in time — the old or the new. They all required commitment before the reform proved itself. They all underestimated how much existing literacy was a sunk cost the reader had no incentive to abandon. And they all operated in a physical world of paper and type, where every reform required the cooperation of printers, publishers, schools, and governments — all conservative by structure.
These are the failures we are trying not to repeat.
Four things have changed since the last serious attempt. None of them is small, and together they are decisive.
The substrate is now digital. A reader’s text is no longer captured on paper. It is rendered by a font, served by a browser, or displayed by an application. The same underlying text — the same sequence of Unicode characters — can be displayed differently on different devices, for different readers, at different times, with no change to the source. A reform of the rendering layer does not require the cooperation of any printer, publisher, school, or government. It requires only that the reader install a font.
This is the change that makes everything else possible. Shaw could not have done what we can do, because Shaw lived in a world where the appearance of a text was fixed at the moment of printing. We do not live in that world. We live in a world where every text is remade, pixel by pixel, every time it is read. The remaking is under software control. Software is malleable.
The reform can be phased. No previous attempt allowed a reader to move gradually from the old script to the new, learning one substitution rule at a time, with the previous month’s progress preserved and the next month’s introduced only when the reader is ready. Every previous attempt asked the reader to commit fully or not at all. The phased approach inverts this: at each stage, the reader has learned a small, named, well-defined thing. After one phase, ph becomes f. After another, ck becomes k. After many phases, the reader is reading a phonetically faithful script without ever having confronted a wall of unfamiliar text.
This solves the cold-start problem that killed every previous reform. There is no moment at which the reader cannot read. The script the reader sees at every stage is a script the reader can already mostly read, with one small new rule applied.
The reform is reversible. A reader who has installed the font and the engine can, at any moment, turn it off. The original text is preserved underneath. The new spelling sits as a rendering of the original, not a replacement of it. This means there is no risk to the reader of being cut off from existing literature. Every book ever written remains available in its original form by toggling a switch. The reader keeps full backward compatibility with five centuries of English text.
This is the answer to the ITA problem. ITA failed because the transition back was hard. Our system has no transition back to fail at, because the original is never erased. It is one click away at all times.
The encoding is IPA. Under the surface, the script the reader eventually arrives at is not a new invention. It is the International Phonetic Alphabet — the standard used by every linguist, every language-learning textbook, every speech-recognition system, every text-to-speech engine on earth.15 The Unicode codepoints we use are not ours. They are the codepoints assigned by Unicode to IPA, and they have been stable for decades and will be stable for decades more.16
This means three things. It means that a reader who learns naYana can also read IPA — that the script they learn is not a private language but a passport to every dictionary and language resource in the world. It means that text written in naYana is automatically compatible with speech synthesizers and language tools, because those tools already accept IPA. And it means that the script can be extended to other languages, because IPA is universal — any sound in any language is already represented in it.
The novelty in naYana is not the encoding. The encoding is borrowed from a hundred years of careful linguistic work. The novelty is the glyph design: the shapes that the IPA characters take when rendered. IPA was designed by linguists for transcription, not by educators for learning. Its shapes are scientifically precise but cognitively expensive — mirror-image pairs, near-duplicates, marks that depend on tiny distinctions.17 naYana redraws each IPA character with a shape chosen for learnability: distinct from every other shape, easy to write by hand, easy to recognize at small sizes, free of mirror confusion. The underlying Unicode is IPA. The visible script is naYana.
We have built the first phases of this system. They work. A child can install a font on her browser, slide a control from 0 to 1, and see philosophy appear as filosofy. Slide it again, and cat becomes kat, city becomes sity. Each slide is small. Each is named. Each is reversible. Over months, the slide can move from 0 to 20, and by the end, the child is reading a phonetic script in which every sound has exactly one spelling and every spelling has exactly one sound. The script she is reading is not English. It is also not a new invention. It is the actual sounds of English, written down at last with the consistency that every other major language in the world has long enjoyed.
We do not believe this will be easy. We do not believe everyone will adopt it. We do not even believe most people will adopt it. What we believe is that the option should exist — that a child should not be required, in the twenty-first century, to spend five years memorizing the etymological accidents of receive and believe when she could have been reading stories. The choice should be hers, or her parents’, or her teacher’s. The cost of not offering the choice is the continued, unmeasured loss of trillions of hours of human childhood every year, forever, with no end and no audit.
That is the case for naYana. It is not a manifesto. It is a proposal. The proposal is: let us build the rendering layer that makes phonetic English available to anyone who wants it, without asking them to give up English; let us see whether children take to it; and let us measure whether the time saved is real.
If it works, the saved time is the answer to a question humanity has been ignoring for five hundred years. If it does not, we will have spent some years and some money trying, which is the smallest expense we could have made on a problem this size.
Brave enough to try it? Continue reading the same essay in a partially phased and reformed version, where the spelling reforms are revealed progressively as you scroll. You will see what the early phases actually look like on the page.
The name naYana is a Sanskrit word meaning “eye” or “guidance.”18 It echoes the Nyāya school of Indian logic, founded on the principle that careful seeing precedes correct reasoning. The script is meant to guide the eye, gently, from the spelling it knows to the spelling that matches the world.
The name is also a phonetic palindrome. Read it forward: n-a-Y-a-n-a. Read it backward: n-a-Y-a-n-a. The capitalized Y in the middle is the axis of symmetry. This was not an accident. A script that works for English must, in principle, also work for languages that read right-to-left like Arabic and Hebrew, or top-to-bottom like classical Chinese and Japanese. The palindrome marks the script’s intent: writing is a recording of speech, and speech does not have a preferred direction. The orthography should not impose one.
We are starting with English because English is the language with the largest mismatch between its sounds and its writing — and because English, as the global second language, exports that mismatch to every country where it is taught. A child in Maharashtra learning English does not just learn English; she learns to spend years of her childhood on the same fossilized irregularities that English-speaking children spend years on. The cost is global. The fix would be global.
But there is no reason the script needs to stop at English. The IPA encoding underneath naYana is universal. Any language ever transcribed in IPA — French, Mandarin, Arabic, Yoruba, Tamil — can be rendered in naYana shapes. A child who learns naYana for English has also, without knowing it, learned the script for every other language she might encounter. The literacy is transferable the way no orthography in history has been.
This is the larger vision, stated as carefully as we can state it: that the time stolen from billions of children by inconsistent writing systems is recoverable; that the recovery requires no new laws, no government adoption, no schoolboard approval — only a font, an engine, and the consent of one reader at a time; and that the human capacity for creative work, for cultural participation, for early entry into the literate world, can be restored to its natural age of seven instead of being deferred to fourteen.
What that early entry would mean, at scale, we do not know. We have never had it. Every literate adult alive today was educated under the old system. Whole generations of children, in every English-speaking country, have spent the most plastic years of their lives on a problem we could have removed.
It is time to remove it.
naYana is an open project. The code, the font, and the engine are public.19 The script itself is built on IPA and is unowned. We invite contributors — typographers, linguists, educators, software developers, parents, and children — to use it, test it, and improve it. Particularly children. They will tell us, faster than any study, whether what we have built is worth using.
If a child can write filosafi and be understood, we have done nothing for her. If a child can write filosafi and be right, we have given her back her childhood.
— for the gnowledge lab project, naYana for universal literacy
The most-cited cross-linguistic comparison is Seymour, Aro, & Erskine (2003), “Foundation literacy acquisition in European orthographies,” British Journal of Psychology 94: 143–174. They found Finnish, Greek, Italian, and Spanish first-graders reading at near-ceiling accuracy on familiar words, while English first-graders averaged around 40%. English children took roughly 2.5 times as long to reach equivalent decoding fluency. The “three times” figure in the text is approximate; the gap depends on how fluency is operationalized. ↩
The term orthographic depth was introduced by Katz & Frost (1992) in “The reading process is different for different orthographies: The orthographic depth hypothesis,” in Orthography, Phonology, Morphology, and Meaning (Frost & Katz, eds.). “Shallow” orthographies have transparent grapheme-to-phoneme mappings; “deep” orthographies, like English, encode etymological and morphological information at the cost of phonemic transparency. ↩
Indonesian’s spelling was rationalized in stages: the Ejaan Republik (1947) and the Ejaan yang Disempurnakan (1972) brought it close to fully phonetic. See Sneddon, The Indonesian Language: Its History and Role in Modern Society (2003). ↩
Defior, Martos, & Cary (2002), “Differences in reading acquisition development in two shallow orthographies: Portuguese and Spanish,” Applied Psycholinguistics 23: 135–148. ↩
The ough example is a popular illustration but worth a note: hiccough is a now-rare spelling of hiccup preserved chiefly in older British usage. The other six are current. For a careful catalogue of English phoneme-grapheme correspondences see Edward Carney, A Survey of English Spelling (Routledge, 1994), or the empirical re-analysis of Hanna et al.’s 17,310-word corpus by Sherman & Lott (2014). ↩
The silent s in island was introduced in the 16th century by scholars who incorrectly believed the word derived from Latin insula. The native English word is ī(g)land, with no historical s at all. The Latinate spelling stuck, and English children continue to pay the bill. See the Oxford English Dictionary entry for island. ↩
Rates of developmental dyslexia diagnosis correlate with orthographic depth. Italian, with a transparent orthography, has dyslexia rates roughly half those reported in English-speaking populations, despite the underlying neurological condition being identical. See Paulesu et al. (2001), “Dyslexia: Cultural diversity and biological unity,” Science 291: 2165–2167. ↩
Estimates vary by source. Ethnologue (2024) gives roughly 380 million L1 speakers and 1.1 billion L2. The British Council’s higher estimate of 1.5 billion L2 speakers includes “users with any working command.” The order of magnitude is what matters. ↩
This estimate is an order-of-magnitude calculation, not a peer-reviewed finding. 1.5 billion learners × 1 year × 1,000 instructional hours ≈ 1.5 × 10¹² hours. The piece says “approximately two trillion” to round generously upward and account for non-instructional time spent on spelling (homework, reading practice, testing). The point of the figure is the scale, not the precision; the smallest defensible estimate is still hundreds of billions of hours. ↩
For the will, see The Last Will and Testament of George Bernard Shaw (1950). The Public Trustee held an alphabet design competition; the winning entry by Kingsley Read was published as Androcles and the Lion: An Old Fable Renovated, With a Parallel Text in Shaw’s Alphabet (Penguin, 1962). The Shavian script is encoded in Unicode at U+10450–U+1047F. ↩
The Deseret alphabet was developed by the regents of the University of Deseret in the 1850s under Brigham Young’s direction. See Beesley & Elzinga (1992), “An 1860 English-Hopi vocabulary written in the Deseret alphabet,” Utah Historical Quarterly 60: 252–273. Encoded in Unicode at U+10400–U+1044F. ↩
Sir James Pitman’s ITA was first published in 1959 and widely used in British primary schools from 1961 until the late 1970s. For longitudinal evaluation see Warburton & Southgate, I.T.A.: An Independent Evaluation (Murray and Chambers, 1969). ↩
The longitudinal follow-up by Downing (1967) and subsequent meta-analyses showed that ITA-taught children initially outperformed traditional-orthography peers but lost the advantage during the transition to standard English. Some studies showed worse spelling performance among former ITA students years later. The transition problem is now a standard cautionary tale in literacy research. ↩
For SoundSpel, see the work of the American Literacy Council (sound-spel.com). Cut Spelling was developed by the Simplified Spelling Society, now the English Spelling Society. SaypU (Spell As You Pronounce Universally) was a 2014 web project by Jaber George Jabbour. Unifon was designed by John R. Malone for the Bendix Corporation in the 1950s. Each has its devotees; none has crossed into general use. ↩
The International Phonetic Alphabet was first published by the International Phonetic Association in 1888 and has been periodically revised. See the IPA Handbook (Cambridge University Press, 1999). ↩
IPA characters are encoded principally in Unicode’s “IPA Extensions” block (U+0250–U+02AF) and “Spacing Modifier Letters” (U+02B0–U+02FF). The block has been stable since Unicode 1.1 (1993). ↩
A clear discussion of IPA’s cognitive load for non-linguists appears in Ladefoged & Disner, Vowels and Consonants (3rd ed., 2012), particularly the chapters on phonetic notation. The IPA was explicitly designed as a transcription tool — to record observed speech — not as a literacy script. Its visual choices reflect that purpose. ↩
From Sanskrit नयन (nayana), “the eye,” from the verbal root nī, “to lead, to guide.” The same root underlies the name of the Nyāya (न्याय) school of classical Indian philosophy, which holds that valid perception (pratyakṣa) is the first source of knowledge. See Matilal, The Character of Logic in India (SUNY Press, 1998). ↩
The project repository, font sources, preprocessor engine, and documentation are at https://www.gnowledge.org/projects/naYana. Contributions welcome. ↩