WORK IN PROGRESS - NON-EXHAUSTIVE - NOT THE FINAL PRODUCT
AI Compute and Educational Access: Living Research Notebook
Living notebook updated 2026-09-09 · Current readable paper: v0.11
This is an ongoing research notebook about where AI-assisted educational translation and adaptation could create useful access. It is not the final study, an exhaustive list of needs, or a completed funding-priority ranking.
The 100 C001-C100 entries on this landing page are a supporting, nonordinal research catalogue. Separate ranked Top-100 comparisons report potential readers under stated assumptions and potential reach per workload for a common package. The catalogue identifiers are not ranks. Missing from the catalogue is not a finding of lower need; missing evidence remains unfinished research.
Every educational stage remains eligible: early learning and primary school, secondary school, vocational and practical learning, university, and postgraduate or research use. The same language may have several different educational opportunities.
The study concerns worldwide educational needs. Files, translations or previous spending in one local programme do not reduce a population's need or determine its place in the comparison.
GitHub is the continually evolving version. The current v0.11 reading links above present the dated working-paper checkpoint. The original 4 September 2026 hypothesis catalogue remains available for comparison; later evidence may change findings and recommendations.
Zenodo mirror for v0.11: pending. Read the current v0.11 edition on GitHub and GitHub Pages. The preceding v0.10 edition remains on Zenodo; that DOI does not identify v0.11.
This integrated edition examines where AI-assisted educational resources could help and how limited production budgets can be compared. New sections examine Thai, Malay and Khmer resource findings and prerequisite-expanded learning packages. They distinguish observed resource evidence from conditional allocation examples. The two worldwide conditional Top-100 comparisons remain unchanged. Token workload, physical compute and achieved learning are different measurements.
Original analysis and presentation are CC BY 4.0. Included OpenStax source material retains CC BY-NC-SA 4.0 and its original licence and credits; the analysis licence does not replace or relicense that material.
Current ranked comparisons
These are two explicitly ranked, conditional comparisons. The first orders language-and-script editions by potential readers under stated assumptions. The second compares potential reach per workload when every edition receives the same 100,000-source-token educational package. Neither numerator counts people observed to have learned, and token events do not measure physical compute.
These tables are conditional research results, not the finished funding order. The final answer still has to bind the educational package, stage, independently usable local supply, additional learning effect, delivery and accessibility costs, overlapping beneficiaries, reusable work, and standardized physical compute. The stable C001–C100 profile catalogue below is supporting research and is not a ranking.
Potential readers under stated assumptions: current first ten
Simplified Standard Written Chinese profile — 1,151,960,048 potential readers under stated assumptions at the point scenario (900,233,133–1,273,186,860 interval).
Hindi, Devanagari-script profile — 519,812,678 potential readers under stated assumptions at the point scenario (406,222,938–574,515,298 interval).
Spanish, Latin-script profile — 454,675,495 potential readers under stated assumptions at the point scenario (355,319,567–502,523,387 interval).
Arabic, Arabic-script profile — 339,298,442 potential readers under stated assumptions at the point scenario (265,154,768–375,004,601 interval).
French, Latin-script profile — 298,241,289 potential readers under stated assumptions at the point scenario (233,069,446–329,626,787 interval).
Urdu, Arabic-script profile — 280,449,183 potential readers under stated assumptions at the point scenario (219,165,280–309,962,324 interval).
Bangla, Bangla-script profile — 250,711,118 potential readers under stated assumptions at the point scenario (195,925,593–277,094,767 interval).
Portuguese, Latin-script profile — 223,454,037 potential readers under stated assumptions at the point scenario (174,624,743–246,969,280 interval).
Indonesian, Latin-script profile — 222,592,226 potential readers under stated assumptions at the point scenario (126,363,708–246,016,776 interval).
Russian, Cyrillic-script profile — 180,178,793 potential readers under stated assumptions at the point scenario (140,806,028–199,139,954 interval).
Potential reach per workload for one package: current first ten
Simplified Standard Written Chinese profile — 858,078,700 potential readers under stated assumptions per million standardized fresh-equivalent token events at the point scenario.
Hindi, Devanagari-script profile — 347,929,450 potential readers under stated assumptions per million standardized fresh-equivalent token events at the point scenario.
Spanish, Latin-script profile — 330,708,224 potential readers under stated assumptions per million standardized fresh-equivalent token events at the point scenario.
Arabic, Arabic-script profile — 215,448,628 potential readers under stated assumptions per million standardized fresh-equivalent token events at the point scenario.
French, Latin-script profile — 213,167,471 potential readers under stated assumptions per million standardized fresh-equivalent token events at the point scenario.
Urdu, Arabic-script profile — 183,122,705 potential readers under stated assumptions per million standardized fresh-equivalent token events at the point scenario.
Portuguese, Latin-script profile — 167,824,727 potential readers under stated assumptions per million standardized fresh-equivalent token events at the point scenario.
Indonesian, Latin-script profile — 165,773,561 potential readers under stated assumptions per million standardized fresh-equivalent token events at the point scenario.
Bangla, Bangla-script profile — 160,790,872 potential readers under stated assumptions per million standardized fresh-equivalent token events at the point scenario.
Russian, Cyrillic-script profile — 126,609,405 potential readers under stated assumptions per million standardized fresh-equivalent token events at the point scenario.
The catalogue and dated research checkpoints below are supporting and historical material. Current ranked comparisons and the v0.11 working paper are linked above.
Practical-business education: resources, budgets and benefit thresholds
Read the 5-page supplement: original analysis of specific vocational resources and two primary studies, separate production and delivery budgets, and conditional marginal-benefit thresholds. This is not a final global ranking or measured AI effect. Working paper and supplements on Zenodo. Original analysis is CC BY 4.0; inherited OpenStax material retains CC BY-NC-SA 4.0.
Focused learning packages: what can shorter materials save?
Read the four-page study of matched-outcome benefit and workload, actual OpenStax learning supports and four primary studies. The results are conditional, not a final language ranking. Working paper and both supplements on Zenodo. Original analysis is CC BY 4.0; included OpenStax source retains CC BY-NC-SA 4.0.
New: when can shared editions improve a limited budget?
The two-page Hindi-Urdu study compares a bridge with the best affordable native-edition mix. Shared preparation and audience overlap can change the answer. These are conditional thresholds, not measured learning rates. Working paper and supplement on Zenodo.
New research: what would actually help?
Nineteen further African, Arabic/Iranic and signed-language profiles now have concrete full-course and smaller-package comparisons. They cover early learning, scientific reading, interrupted-schooling continuity, practical data, mathematics and computing.
Read the evidence and specific proposals. The current calculation has designs for 44 forecast profiles and 112 of 600 stage cells. These are investigation counts, not ranks.
Ongoing and non-exhaustive. The v0.5 PDF and Zenodo version incorporate the current checkpoint; the global study remains unfinished.
If your language is not shown
Absence is not a judgment of lower need. This forecast is non-exhaustive. Communities outside it remain eligible; research order is not priority order.
How to read this notebook
Forecast
An expectation to test, not a conclusion or a list of the only eligible communities.
Package proposal
A specified educational resource to compare. Writing a proposal does not establish its effectiveness.
Not yet investigated
Work remains. It does not mean low need, low priority, or no useful resource could be made.
Research coverage
How far this investigation has progressed, not how well a population is served by education.
Different educational stages
School provision cannot settle a language's university, research, practical-learning or accessibility needs.
Related communities
A need identified in one community prompts investigation of comparable communities. It does not justify copying their scores or assuming identical needs.
Twelve hypotheses to test
Written before the next comparison, but after an earlier analysis. These are not an original preregistration. Evidence can support, weaken or overturn them.
H01
Large language communities with substantial stage-specific unmet need should repeatedly appear among high-reach candidates. Indonesian, Hindi, Bangla, Simplified Written Chinese and Urdu are strong initial expectations for investigation, not guaranteed winners.
What would support or weaken it?
Compare affected learner counts and usable local provision at the same stage. High local provision or small additional benefit can weaken a particular package case; national wealth or local project completion cannot substitute for that evidence.
What an unexpected absence triggers
Trace the candidate through every inclusion, population, overlap and scoring step. Missing data cannot settle the comparison against it.
H02
Where a regional language supports schooling but advanced study moves to another language, bridge-to-tertiary resources may have especially high utility.
What would support or weaken it?
Need evidence for the actual transition, learners' capabilities and subject gaps. NPTEL explicitly identifies this mechanism in India. Strong advanced local resources or comfortable academic-language access would weaken the predicted benefit.
What an unexpected absence triggers
Compare Tamil, Telugu, Marathi, Bangla, Odia and other relevant communities; do not infer one language's provision from another.
H03
A need mechanism found in one community should generate hypotheses for related educational systems and neighboring communities. Russian-language research or self-study gaps imply checking Ukrainian, Belarusian, Polish, Bulgarian and other relevant communities.
What would support or weaken it?
Shared educational histories, academic registers or resource routes make an analogy worth testing. Conflict, displacement, different publication markets and different language policies may increase or decrease the gap. Genealogical relationship alone supplies neither an equal score nor a population estimate.
What an unexpected absence triggers
If Russian is present and Ukrainian absent, investigate source coverage, population definitions and implementation before interpreting the result. Repeat this test globally.
H04
Large national or census language categories may conceal distinct communities with different access needs.
What would support or weaken it?
Compare the category definition with exact mother-tongue rows, scripts, schooling and bilingualism. Examples include Hindi-group varieties and Bengali categories containing Sylheti and Chatgaya. Shared classification is not measured comprehension.
What an unexpected absence triggers
Retain the group observation and its possible constituent populations. Resolve overlap or use bounds; never silently discard a constituent or credit it with the whole group.
H05
Educational-resource shortages differ by stage. A language can have extensive school resources and substantial postgraduate or research gaps.
What would support or weaken it?
Compare curriculum coverage and source availability separately for school, vocational, university and research use. Japanese, Chinese, Russian, English, French and other established academic languages remain candidates for specific missing materials.
What an unexpected absence triggers
Investigate the proposed advanced package, rather than excluding the entire language as already served.
H06
A modest number of well-designed interlanguage editions could sometimes serve more learners per compute than separate editions. The advantage may be greater in formula-rich mathematics for readers with adequate prerequisites.
What would support or weaken it?
Compare exact reader communities, scripts, mathematical background, prose burden, semantic-error patterns and standardized production cost. Evaluate definitions, negation and proof dependencies as well as formulas. Existing literature, source analysis and model/mechanical tests can support provisional scenarios without waiting for new participants.
What an unexpected absence triggers
Investigate bridge design and domain fit. Missing a comprehension estimate is not evidence of zero reach; family population is not evidence of full reach.
H07
Availability of some translated resources will often leave important curriculum and delivery gaps.
What would support or weaken it?
Compare actual subjects, prerequisite sequences, exercises, answers, offline availability, accessibility and licensing. One catalogue is not a census of all provision; a listed file is not measured learner access.
What an unexpected absence triggers
Ask what is missing and what would add value, rather than treating either a catalogue entry or its absence as a final need estimate.
H08
Signed-language and accessibility interventions can create substantial access that written-language counts miss.
What would support or weaken it?
Use sign-language-specific populations and educational evidence, not total deafness counts. Compare text, signed explanation, captions, MathML, audio and tactile routes by endpoint.
What an unexpected absence triggers
An entirely spoken/written result set requires an explicit modality audit. A missing written tokenizer is not a reason to remove signed learners.
H09
Offline, phone-usable, exercise-rich resources can benefit motivated learners without teachers, while classroom resources can help a different or overlapping audience.
What would support or weaken it?
Estimate autonomous study, classroom use and research/reference use separately. Test file size, navigation, prerequisite support, answers and device requirements. Do not assume every person owns a smartphone or that a download guarantees learning.
What an unexpected absence triggers
Investigate independent-learning routes whenever a model implicitly requires teachers or continuous internet.
H10
Conflict, displacement, statelessness and cross-border language use can change both need and delivery routes without changing the language's registry homeland.
What would support or weaken it?
Use dated displacement and education evidence, separate current location from language identity and avoid double-counting people across countries.
What an unexpected absence triggers
Census-country absence triggers a route search, not exclusion. Ukraine is one test; other displaced communities require the same treatment.
H11
The best source packages may differ materially across communities: foundational learning, science, computing, statistics, finance, management and advanced research may each dominate somewhere.
What would support or weaken it?
Link package recommendations to local educational or practical needs and independent existing provision. OpenStax and Open Logic provide comparison baselines, not an exclusive subject universe.
What an unexpected absence triggers
A uniform proposal to translate the same book everywhere requires justification.
H12
Raw population, tokenization and missingness can each dominate a numerical model for the wrong reason.
What would support or weaken it?
Run factor-removal, source-removal, cost-standardization, common-prior sensitivity and counterfactual missing-data tests. Keep gross, cached and fresh tokens distinct.
What an unexpected absence triggers
Investigate any ranking that is nearly a population table or that drops a community after evidence is removed.
Supporting nonordinal 100-profile forecast
C001–C100 are identifiers, not ranks. The groups are research themes, not funding bands. Initial stage leads do not exclude any other stage.
F: foundational and primary; S: secondary; V: vocational and practical; U: university; R: postgraduate and research. These are frozen initial leads. Preschool (P) is now explicit in the comparison; all six stages remain eligible for every profile.
100 forecast candidates shown. This is not the full eligible universe.
A. Broad-reach and major educational-language candidates
Candidate identifiers are not ranks. Stage key: F = foundational/primary; S = secondary; V = vocational/practical; U = university; R = postgraduate/research.
ID
Edition or community profile
Boundaries to preserve
Initial stage leads
C001
Indonesian
Latin; Indonesia; L1 and additional-language users separately
F/S/V/U/R
C002
Hindi
Devanagari; exact Hindi versus census-group mother tongues
F/S/V/U/R
C003
Bangla
Bengali script; Bangladesh and India separately
F/S/V/U/R
C004
Standard Written Chinese, Simplified
Mainland written-literacy profiles; not all Chinese varieties or all residents
S/V/U/R
C005
Urdu
Perso-Arabic/Nastaliq; Pakistan and Indian communities separately
F/S/V/U/R
C006
Telugu
Telugu script; regional school-to-tertiary routes
S/V/U/R
C007
Tamil
Tamil script; India and Sri Lanka separately
S/V/U/R
C008
Marathi
Devanagari; Maharashtra learning routes
S/V/U/R
C009
Vietnamese
Vietnamese Latin orthography
S/V/U/R
C010
Hausa
Latin first profile; Ajami separately; Nigeria and Niger separately
F/S/V/U
C011
Standard Kiswahili
Latin; national standards and L1/L2 learning routes separately
F/S/V/U
C012
Modern Standard Arabic
Learned written register; country, schooling and vernacular profiles separately
S/V/U/R
C013
Brazilian Portuguese
Brazil; independent provision and disadvantaged learners
F/S/V/U/R
C014
Spanish, Latin American educational profiles
Country-specific terminology and underserved routes, not a continent-wide count
F/S/V/U/R
C015
English, underserved learner profiles
Territory and capability cells; targeted open/offline/accessible gaps
F/S/V/U/R
C016
French, underserved learner profiles
African and other contexts separately; French ability cannot cover all residents
S/V/U/R
C017
Turkish
Latin; Türkiye and diaspora separately
S/V/U/R
C018
Iranian Persian
Perso-Arabic; distinguish Dari and Tajik
S/V/U/R
C019
Thai
Thai script; standard versus other Thai-area languages
S/V/U/R
C020
Burmese
Myanmar script; distinguish other Myanmar languages
F/S/V/U
C021
Filipino
Latin standard; Tagalog and other Philippine language access separately
S/V/U/R
C022
Standard Malay, Malaysia
Latin profile; Jawi and other regional standards separately
S/V/U/R
C023
Gujarati
Gujarati script
S/V/U/R
C024
Kannada
Kannada script
S/V/U/R
C025
Odia
Odia script; standard versus census-group scope
S/V/U/R
C026
Malayalam
Malayalam script
S/V/U/R
C027
Eastern Punjabi
Gurmukhi; do not substitute Western Punjabi/Shahmukhi counts
F/S/V/U
C028
Russian
Cyrillic; resident and cross-border academic users separately
S/V/U/R
C029
Japanese
Japanese writing; targeted subject and advanced-source gaps
U/R
C030
Korean, South Korean standard
Hangul; North Korean and diaspora profiles separately
U/R
B. Regional and language-transition candidates
Candidate identifiers are not ranks. Stage key: F = foundational/primary; S = secondary; V = vocational/practical; U = university; R = postgraduate/research.
ID
Edition or community profile
Boundaries to preserve
Initial stage leads
C031
Amharic
Ethiopic; do not substitute Ethiopia's total population
F/S/V/U
C032
West-Central Oromo educational profile
Latin; other Oromo varieties separately
F/S/V/U
C033
Somali
Latin standard; regional and displaced learners separately
F/S/V/U
C034
Yoruba
Standard Latin with full marks; national and diaspora routes separately
F/S/V/U
C035
Igbo
Standard Latin with tone/orthographic choices specified
F/S/V/U
C036
Kinyarwanda
Latin; independent local supply audit
F/S/V/U
C037
Kirundi
Latin; do not copy Kinyarwanda provision or reach
F/S/V/U
C038
Lingala
School/written and urban spoken registers separately
F/S/V/U
C039
Central Kanuri
Latin and Ajami routes separately
F/S/V
C040
Wolof
Latin first profile; Wolofal and Garay separately
F/S/V/U
C041
Bambara
Latin and N'Ko routes separately
F/S/V/U
C042
Pulaar
Latin and Adlam profiles; Senegal/Mauritania and other contexts separately
F/S/V
C043
Nigerian Fulfulde
Exact regional profile; no whole-Fulani population assignment
F/S/V
C044
Zulu
Latin; school and higher-education transition
F/S/V/U
C045
Northern Sotho / Sepedi standard
Latin; census and standard identity boundaries retained
F/S/V/U
C046
Haitian Creole
Standard Haitian orthography; Haiti and displaced communities separately
F/S/V/U
C047
Nepali
Devanagari; native and additional-language profiles separately
F/S/V/U
C048
Sinhala
Sinhala script; specific Sri Lankan educational routes
S/V/U/R
C049
Northern Pashto
Perso-Arabic; Afghanistan/Pakistan and Southern Pashto separately
F/S/V/U
C050
Dari
Perso-Arabic; Afghanistan and displaced learners separately
Khmer script; Cambodia and other communities separately
F/S/V/U
C064
Lao
Lao script; no automatic Thai coverage
F/S/V/U
C065
Northern Uzbek / Uzbekistan standard
Latin and Cyrillic profiles; Southern Uzbek separately
S/V/U
C066
Kazakh
Cyrillic first profile; actual script cohorts separately
S/V/U/R
C067
Tajik
Cyrillic; do not assign Persian/Dari literacy automatically
S/V/U
C068
Uyghur
Perso-Arabic first profile; diaspora script routes separately
F/S/V/U
C069
Central Kurdish / Sorani
Perso-Arabic; country and educational systems separately
F/S/V/U
C070
Northern Kurdish / Kurmanji
Latin first profile; other scripts and regional varieties separately
F/S/V/U
C. Comparators, research access and regional-depth candidates
Candidate identifiers are not ranks. Stage key: F = foundational/primary; S = secondary; V = vocational/practical; U = university; R = postgraduate/research.
ID
Edition or community profile
Boundaries to preserve
Initial stage leads
C071
Ukrainian
Cyrillic; residents and displaced learners separately
F/S/V/U/R
C072
Polish
Latin; advanced gaps and refugee-learning routes separately
S/V/U/R
C073
Romanian
Latin; Moldova and Romania separately
S/V/U/R
C074
Bulgarian
Cyrillic; no Russian or Interslavic coverage assumption
S/V/U/R
C075
Belarusian
Cyrillic first profile; actual preferred/usable educational language
S/V/U/R
C076
Serbian standard
Cyrillic and Latin; other BCMS standards separately
S/V/U/R
C077
Tigrinya
Ethiopic; Ethiopia, Eritrea and displaced learners separately
F/S/V/U
C078
Kabyle
Latin first profile; Arabic/French overlap separately
F/S/V/U
C079
Central Atlas Tamazight
Exact variety; Tifinagh/Latin routes and Moroccan standard separately
Specified regional educational orthography; not all Quechua
F/S/V
C083
Paraguayan Guarani
Standard orthography; bilingual and predominantly Guarani profiles
F/S/V/U
C084
Central Nahuatl
Exact variety and local orthographic route; not all Nahuatl
F/S/V
C085
K'iche'
Standard plus regional comprehension profiles
F/S/V
C086
Tok Pisin
Standard written register; L1/L2 and schooling routes separately
F/S/V/U
C087
Bislama
Standard written register; not all Vanuatu language users
F/S/V
C088
Kalaallisut / West Greenlandic
Greenlandic standard; other Inuit languages separately
S/V/U/R
D. Signed-language hypotheses
Candidate identifiers are not ranks. Stage key: F = foundational/primary; S = secondary; V = vocational/practical; U = university; R = postgraduate/research.
ID
Edition or community profile
Boundaries to preserve
Initial stage leads
C089
American Sign Language
US and other actual communities separately; not universal signing
F/S/V/U/R
C090
Brazilian Sign Language / Libras
Regional signing variation and educational stages
F/S/V/U
C091
Indian Sign Language
Regional variation and actual sign-user population; no national deafness substitution
F/S/V/U
C092
Kenyan Sign Language
Actual Kenyan signing communities, not Kiswahili coverage
F/S/V/U
C093
South African Sign Language
Regional variation; English/Zulu written access separately
F/S/V/U
E. Shared-register and interlanguage hypotheses
Candidate identifiers are not ranks. Stage key: F = foundational/primary; S = secondary; V = vocational/practical; U = university; R = postgraduate/research.
Devanagari/Nastaliq; register and literacy-specific reach
S/V/U
C096
Malay-Indonesian shared technical profile
Exact national terminology choices; no wholesale Austronesian coverage
S/V/U
C097
Persian-Dari-Tajik shared technical profile
Perso-Arabic/Cyrillic; lexical and schooling differences explicit
S/V/U/R
C098
Oghuz-focused Inter-Turkic design hypothesis
Turkish/Azerbaijani/Turkmen profiles; not all Turkic languages
U/R
C099
N'Ko literary Manding educational profile
Exact Manding communities and N'Ko literacy; no family-wide reader count
F/S/V/U
C100
Romance mathematical intercomprehension profile
Exact Portuguese/Spanish/French/Italian/Romanian routes; natural-text and constructed alternatives compared
U/R
No match in this forecast. That is not an exclusion or a low-need finding. Check the wider register below; aliases and exact varieties may differ.
Research progress, not a need score
The earlier v0.1 snapshot contained routes for 25 profiles across 42 of 500 stage cells. The linked working update now contains designs for 44 profiles across 112 of 600 cells, including explicit preschool. Those counts describe this calculation only: they omit separate shared-register analyses and do not measure the full research record, educational provision or need. The remaining cells are unfinished comparisons, not rejected populations. The wider retained register contains 8,618 language-level entries from Glottolog 5.3, including historical and bookkeeping entries labelled in the source. That register is an identity starting point, not an exhaustive inventory of every community, variety, script, diaspora or accessible format.
Expand comparable population, stage, subject, supply and delivery evidence beyond the initial forecast.
Test useful full courses and smaller additions on the same basis, including OpenStax, Open Logic, other open resources and advanced research materials.
Compare native-language, interlanguage, signing and accessible-format options without assuming complete overlap or zero comprehension.
Investigate unexpected absences and possible high-value candidates outside the forecast; test the responsible rule across the wider universe.
Develop evidence-supported recommendations with uncertainty, rather than converting research order or missing measurements into ranks.