Spelling Reform: An Engineering Approach

Kamran Ossia (kamran@euspell.org)

Download the PDF — original fonts and typesetting — or preview it online

Introducing Euspell, a conservative spelling reform system which includes: a 205000-word lexicon where each word is tagged with its common parts of speech, a 35000-word PLS lexicon of new spellings with their IPA pronunciations for use in speech synthesis, and a 46000-word lexicon listing the parts of speech of the new spellings. Also provided are browser extensions and word processor add-ins, PDF and epub readers, and in-browser dictation. The new spellings are chosen to remain within the spirit of English pronunciation and to make human and machine comprehension easier. No diacritics or new letters are introduced, and minimal changes are made to scientific and technical terms. Most new spellings involve removing unneeded letters, changing the verb ending -s to -z for third person singular, and -ed to -d for past tenses when the e is silent, all designed to improve consistency and to make reading easier. Two new forms qh and uw are introduced to disambiguate the ch and ow sounds. Source code is released under GPL-3 and lexicons are released under CC BY-SA 4.0, available at github.com/ossiak/ .

Nomenclature

The Main Lexicon: A 205000-entry csv file containing English words, their parts of speech, three-digit codes indicating the reason for spelling change, and the new spellings, if any.

PoS: Part of speech, e.g. noun, verb, adjective.

CLAWS-7: University of Lancaster PoS tagset [1].

NLP: Natural language processing. For our purpose, NLP involves mainly text classification.

Grapheme: The smallest unit in a system of writing a language that can express a difference in sound or meaning.

Homophone: Words that sound the same but have different spellings, e.g. so and sew.

Homograph: Words that have the same spelling but different sounds or meanings, e.g. live and bears, some of which may split a homograph into two or more words, e.g. tear → tear / taer.

Spurious letter: one whose removal or replacement brings the word more in line with common English spelling rules, e.g. the final e in have, give, examine (→ hav, giv, examin), the o in lose, move, flood (→ luse, moov, fludd).

Disambiguation: Changing the spelling of a word so that its meaning, PoS, or pronunciation is not ambiguous, either by removing or replacing spurious letters, or by changing grammatical endings.

PLS: Pronunciation Lexicon Specification, an XML (extensible markup language) file describing the phonetic IPA (International Phonetic Association) pronunciation of each word.

Introduction

English became the international language of science, commerce through a unique combination of attributes: expressivity with a vast and expandable vocabulary, a simple grammar free from gender and case endings, the use of the Latin alphabet without diacritics, a comprehensive set of verb tenses, absence of diglossia, and a systematic way of generating new technical and scientific terms. The main drawback of English is its inconsistent spelling. We believe that we have found the sweet spot of maintaining readability, improving comprehension, and reducing spelling inconsistencies and ambiguities by creating an entire ecosystem for the new spelling, including spelling and grammar check lexicons for word processors, browser extensions to convert web pages in real time, plugins for text-to-speech and dictation apps, and PDF/epub/text converters for desktop and mobile platforms.

History

The English language does not have a ‘regulatory agency’ similar to the French Academy or the Council for German Orthography. Spelling changes in English generally happen through mass adoption. While Old English was largely phonetic when it first adopted the Latin alphabet, linguistic evolution and the arrival of Norman scribes in the 11th century made its orthography highly irregular. This inconsistency was worsened by distinct regional dialects and the misguided ‘Latinization’ of certain words. The invention of the printing press during the Great Vowel Shift exacerbated the situation. Spellings were chosen almost at random from existing variations. Additionally, early printers added or removed letters simply to justify text margins rather than adjusting the spacing, which further delayed the establishment of a consistent spelling system.

One of the early attempts at standardizing English spelling came from Richard Mulcaster [4]. Taking a more moderate approach, he sought to standardize existing spellings, establishing conventions like using a silent e to indicate long vowels. Many attempts over centuries have had mixed results. On one extreme was George Bernard Shaw who invented an entirely new alphabet [5], while others have promoted phonetic spelling or adding diaciritical marks. The Americanized spelling promoted by Noah Webster [6] made lasting changes including -ise → -ize for Greek-derived verbs (which disambiguates a few words, e.g. analyses) and dropping unneeded letters as in colour → color. The majority of Euspell changes are a continuation of this approach, as explained in the guidelines below. In the early 1900’s, Andrew Carnegie financed efforts to create a simplified spelling system, published by Paine [7]. Although quite conservative in its reform, it included a large set of rules which were not always consistent, plus a set of case-by-case changes. Many of Euspell changes coincide with those of Paine. Upward [8] has proposed ‘Cut Spelling’ where redundant letters and many unstressed vowels are removed. However, there are inconsistencies int their new spelling, e.g. burlesque becomes burlesq but brusque remains unchanged. Disambiguating diatones is done for -ate forms but not for the ow and ch graphemes or the -s ending. Collisions are even introduced, e.g. can’t → cant, guild → gild, write → rite, weather/whether → wethr, etc. Some pronunciations are also changed, e.g. bear → ber, could → cud, and putt → put. Proper nouns and names of people and places are also respelled, creating potential legal and administrative issues.

Motivation

The cost of inconsistent spelling falls on two groups. For human readers, the burden is heaviest on learners and on those with reading difficulties. Memorizing irregular spellings on a case-by-case basis requires a different level of effort than remembering consistent rules such as when a vowel is long or short. The size of that difference has been measured. Seymour et al. [9] tested foundation literacy in thirteen European orthographies using the same three tasks in each: letter knowledge, reading familiar high-frequency words, and reading simple nonwords. At the end of the first school year, children reading Finnish scored 98% on familiar words, Greek 98%, German 98%, Italian 95%, Spanish 95%. Children reading English scored 34%. On nonwords — the test of decoding proper, uncontaminated by whole-word memory — the same children scored 29%, against 95% for Finnish and 94% for German. Reading speed showed the same break: most languages reached fluency at under 2 seconds per word within the first year, while the English-language sample took 7.8 seconds per word and was classified as an outlier. The authors’ summary is that “the rate of development in English is more than twice as slow as in the shallow orthographies”, and that 7.5 years or more of reading is needed before English readers match the European levels reached in the first school year.

The study points out three findings. Letter-sound knowledge was not the problem: the English-language sample scored 94% on letters, comparable to every other group, so the deficit is in the orthography rather than in the alphabet or in how it was taught. The comparison used the higher-socioeconomic sample, whose reading ages were already ahead of expectation. The effect persists into the second school year, where the same children reached 76% on familiar words — still below what every shallow orthography achieves a year earlier. Seymour et al. attribute the delay to two independent causes: syllabic complexity, which affects decoding, and orthographic depth, which affects both word and nonword reading. English is unusual in being extreme on both. Respelling addresses only the second. A reform of the orthography cannot make English syllables simpler, and no claim is made here that it would close the whole gap, only that it removes one of the two documented causes, and the one that is amenable to change.

The same pattern appears in the clinical literature. Landerl et al. [10] found that German and English dyslexic children share the same underlying phonological deficit but differ substantially in how severely it shows up in reading, the German children reading nonwords far more accurately than their English counterparts. Paulesu and colleagues [11] reached the same conclusion from neuroimaging across Italian, French and English readers: one biological condition, expressed more or less disablingly according to the depth of the orthography the reader’s native language.

The KPMG Foundation [12] estimated that the pupils leaving primary school in England and Wales each year unable to read and write properly — 6% of eleven-year-olds, about 38700 children — will each cost the public purse between £44797 and £53098 by the age of 37, through the associated effects on education, health, employment and crime: between £1.73 billion and £2.05 billion for every year’s cohort. That figure is for one jurisdiction and covers literacy failure from all causes, not orthography alone.

For machines, the same irregularity is a direct obstacle: text-to-speech engines mispronounce homographs, and PoS taggers must guess the tags where the orthography gives no signal. A wider collection of research on the subject is maintained by the English Spelling Society [14].

The reason this has remained hard is that the interesting cases are not simple substitutions. Whether read rhymes with reed or red, whether the -s in a word marks a plural or a third-person verb, none of these can be settled by a context-free rule, yet each demands a different spelling if the reform is to reduce ambiguity. A reform that ignores such cases leaves the hardest part of the problem untouched.

Euspell is therefore built as an engineering system rather than a set of prescriptions. Conversion is performed by a deterministic pipeline over a lexicon, so the same input always yields the same output with no human judgment at conversion time; the genuinely ambiguous cases are resolved by classification over local context, treating disambiguation as the machine-learning problem it is. Because the reform is expressed as software, it can be delivered where reading and writing actually happen: browsers, document readers, word processors, and text-to-speech and dictation, and applied reversibly, so that a reader loses nothing by trying it and can recover the original text.

The Lexicons

euspell_lexicon.csv is a comprehensive lexicon of English with about 205000 entries in comma-separated value format with four fields: Word (the original English word), PoS (all common parts of speech for each word), Encoding (a three-digit number indicating if and how the spelling will be changed, see Appendix D), and euspelling (the new spelling(s) of the word, if any). All different parts of speech for nouns (singular, plural, or uncountable), four verb forms, adjectives, and adverbs, etc. are included in the Lexicon. The PoS tags are adapted from CLAWS7 [1] which is the most granular tagset with 138 tags (See Appendix B). The following lines are typical:

barrelled,JJ|VVD|VVN,021,barreld 

barrelling,JJ|VVG,631,barreling 

barrels,NN2|VVZ,012,barrels|barrelz 

barren,JJ,000,[]

euspell_tts.pls. This XML pronunciation lexicon specification file contains over 35000 words with new spellings and IPA pronunciations [3] (see Appendix C), to be used by speech synthesis software. The following lines are typical:

<lexeme><grapheme>buwz</grapheme><phoneme>baʊz</phoneme></lexeme> <!-- bows -->

<lexeme><grapheme>buyz</grapheme><phoneme>baɪz</phoneme></lexeme> <!-- buys -->

<lexeme><grapheme>buzzd</grapheme><phoneme>bʌzd</phoneme></lexeme> <!-- buzzed -->

<lexeme><grapheme>buzzez</grapheme><phoneme>ˈbʌzɪz</phoneme></lexeme> <!-- buzzes -->

euspell_pos.tsv. This tab-separated file with three fields (word, lemma, PoS) is a list of over 46000 entries for the new spellings in the Penn Treebank (PTB) [2] format required by grammar checkers. Each word has as many entries as its parts of speech, for example the word barreld has three entries:

barreld barreld JJ

barreld barreld VBD

barreld barreld VBN

Rules

1. No diacritical marks or new apostrophes will be added.

2. The new spelling should not coincide with an existing one.

3. The third-person singular present tense ending -s will be changed to -z.

4. The past tense verb ending will be -d unless the e is needed for correct pronunciation.

5. The silent gh will be reduced to h, with the preceding vowels adjusted as needed.

6. The f-sounding gh will be changed to ff, with the preceding vowels adjusted as needed.

7. The string ch will be changed to qh only when pronounced tʃ as in church.

8. The string ow will be changed to uw only when pronounced aʊ as in cow.

9. The French strings ch and oi pronounced sh and wa will be spelled as such.

10. The string ou pronounced uː will be replaced with u or oo when appropriate.

11. Spurious letters will be removed or replaced to disambiguate meaning or pronunciation.

12. Capitalized common nouns and adjectives may be changed; otherwise proper nouns and names of people and places will be left alone. Transliterated foreign names may be respelled.

Guidelines

The main goal of Euspell is for an English speaker to be able to read the text with relative ease; Euspell words are in almost all cases readily recognizable by the reader, differing by only one letter.

Familiarity. The new text must be close enough to traditional text that reader familiar with either spelling should be able to read the other text comfortably. This is the case for all shortened words. The words with the new forms qh and uw (described below) may take a few days to get used to.

Disambiguation. Euspell attempts to disambiguate homographs as much as practical, improving both human and machine comprehension. The largest category belongs to the -s ending, retained for plural nouns and changed to -z for third-person-singular verbs. The NN2|VVZ disambiguation algorithm is based on support vector machine (SVM) classification and has an accuracy of 94%. It struggles with minimalist headlines and words with insufficient context. The next largest category is JJ|VV0 words ending in -ate, such as graduate, where the JJ/NN1 form will lose the final e. A three-way disambiguation involves the words leads, showers, and winds, and the variants of slough (except sloughs). A four-way disambiguation involves the words bows, fillets, rows, sloughs, and tears.

Consistency. Another goal of Euspell is removing or replacing spurious letters which violate the spirit of English spelling: live (verb) should be spelled liv, examine should be examin, and so forth, because the spurious final e signals a long preceding vowel which is not the case here. Nearly 2000 words have their final e dropped to conform, including adjectives ending in -ive and -ate, e.g. activ and fortunat. About 150 words ending in -ate[s] have two pronunciations, e.g. graduate is ɡrædʒueɪt as verb and ɡrædʒuɪt as noun/adjective, thus splitting into graduate/graduat.

Avoiding collisions. Creating collisions with existing words goes against the philosophy of Euspell. Of the nearly 35000 words with new spellings, only a few have resulted in collisions with existing words, e.g. heart → hart (male red deer) and weather → wether (neutered ram). These words can be distinguished from context. Until a solution is found to distinguish the two short u sounds ʊ and ʌ (put versus putt), the words could, crowed, should, and would will be spelled coodd, crowdd,shoodd, and woodd. Without the doubled d’s these would result in collisions. Other past tenses avoiding collision are words like barred, gelled, mannd which will be barrd, gelld, and mannd respectively. Euspell standardizes on American spelling, so technically the converted British spellings result in collisions. A set of ‘good’ collisions are words reverting to their former spellings, as shown in Table 1.

Table 1. Words reverting to former spellings

Reversibility. The rarity of collisions means that Euspell is reversible, i.e. the original document can be recovered, albeit with American spelling. Note: Do not convert this paper or other documents with mixed traditional and Euspell content.

Treatment of silent letters. Silent letters fall into a few categories. In some cases such as hymn and bomb, pronouncing them requires additional effort, so they tend to be dropped in normal speech, but they can appear in derived forms such as hymnal and bombardment (but not bombing). These silent letters are left alone. Another category is where silent letters indicate a preceding long vowel, as in rogue, which are dealt with on a case-by-case basis to follow the relevant rules. Spurious silent letters as the ue in colleague are removed.

Disambiguation of the -s ending. The largest single change in Euspell is to distinguish the -s ending of plural nouns from third-person-singular verbs (VVZ) by changing the latter ending to -z. This single change will be of great help in human and machine comprehension. Although the s in is is not an ending, it will also be changed to -z to be consistent with other VVZ forms and to distinguish it from the possessive ending -’s: Jim’s home (the house of Jim) and Jim’z home (Jim has returned).

It’s not it’s any more. The confusion of its and it’s has become widespread in recent years, seen even in advertising and published articles. The word it’s, meaning it is, will no longer exist in the new spelling, replaced with it’z. Its will remain unchanged.

Past tense ending. It has been argued that adding -ed as the past tense ending is a simple rule that should not be changed. That is not entirely accurate, as additional rules exist to double the preceding consonant in some cases, and adding only -d in other cases. It may even create ambiguities such as secreted (past tense of both secret and secrete). The new rule is to add only -d, unless the e in -ed is needed for correct pronunciation: dated, pitted, added, learned (JJ) and dogged (JJ).

Disambiguation of gh. The silent gh will be shortened to h. The preceding o or u will be dropped to maintain correct pronunciation. The h is retained to preserve vowel length as in night → niht, and to retain familiarity with the older forms. The f-sounding gh will be replaced by ff to preserve vowel lengths, e.g. cough, coughing, coughed will be coff, coffing, coffd. Slough will have three spellings: slouh, sluh, sluff. The latter two are already the Americanized spellings for slough meaning marsh/pond and shed/cast off, respectively. As shown later, gh will used to indicate a hard g.

Disambiguation of ea. The string ea can represent four sounds. In case of a single-syllable ea, the sounds are ɛ, iː, and eɪ. There is no apparent rule for its pronunciation, and it is present in four homographs: cleanly, read, lead, and tear, each with two pronunciations. The new rule is that all ea forms will keep the iː sound, and other sounds will be represented by e, eh, or ae, as needed, e.g. hedd, erth, behr, lether, graet. Homographs will be split as needed, and in the case of tears there will be four spellings: tears, tearz, taers, taerz. For bear, the animal will be baer (same as the German spelling) and carry/tolerate will be behr, with a single-letter change.

Disambiguation of o. There are many commonly used words where o is used for the ʊ or uː sound: to, do, lose, move, prove, etc. These words will be modified on a case-by-case basis.

Disambiguation of ou. The ou form is present in about 9000 words in the Lexicon. These fall into three categories: native English words, adjective endings -ous, and French or Greek loan words. It can have four different pronunciations: aʊ (out), uː (group), oʊ (soul), and ə (nimbous). The schwa (ə) almost always occurs in the unstressed adjective ending -ous, where changing -ous to -us will create dozens of collisions (e.g. callous, pious, venous), so it remains unchanged. The oʊ sounds are derivatives of soul, and will be represented with oh. The uː sounding cases, the majority of which are in Greek and French loan words, will be mostly replaced by oo or u depending on the context, except for oup which is unchanged because the sound aʊp does not exist in English. Four, tour, and pour will be forr, toor, and porr. Couth and youth will become cuth and yuth, consistent with Ruth and truth. Words ending in -outh pronounced -aʊθ will remain unchanged.

The liberation of q. The restriction of q having to be followed by u will be lifted. In cases where the u in qu is silent, dropping it solves more than one problem: it allows the q to act as an unambiguous k sound, while preserving both familiarity and the length of the preceding vowel. Hence we will have macaq, opaqe, picturesq, techniqe, torq, etc. Now all qu forms will have the kw sound. The other use of q is to use qh as the tʃ sound as in church, discussed below.

The doubling of v. The letter v seems to be treated differently from other consonants: first, there is a reluctance to end a word in v, resulting in vowel-length violations as in active, give and have. The spurious vowels in these words will be removed. Second, the rule of doubling a consonant to indicate a preceding short vowel is not applied to v, except for a few more recent words like chivvy, divvy, savvy. A possible reason may have been to avoid confusing vv with w. This means that the vowel length before a v has to be memorized: cover/over, seven/even, devil/evil, gavel/navel, etc. Unfortunately, there are more than 500 words in this category, so the change has been restricted to eav → evv, only to disambiguate ea. Similar doublings in general may not be sound-neutral for some consonants, so acid will remain unchanged, as will tether, cosher and other words with two-letter consonants.

The new grapheme qh. The ambiguity of ch in English is most annoying, something which is absent in other languages. The Greek chi poses no problem for Spanish which writes it as c/qu, or other European languages which pronounce it as their native ch. In English the Greek chi is pronounced k with a few exceptions like arch- which can be ark or artʃ . In Euspell, words borrowed from Greek will retain the ch as the k-sound, and the tʃ sound will be represented by qh. The reason for this is to keep scientific words, where most ch forms represent the Greek chi, unchanged. Also, many other languages have their own ch pronunciation, so the English qh will be the unambiguous universal tʃ sound. The French ch will become sh as discussed later.

The new grapheme uw. There is currently no unambiguous way to represent the sound aʊ as in cow in English. The form uw, inspired by Dutch ouw, is introduced to remedy this, replacing the ambiguous ow, but not the ou (as in noun). Instead, the uː sounding ou, almost entirely in French and Greek loan words, will be adjusted based on the context, as mentioned earlier. Some words will end up with two spellings, e.g. bow/buw, sow/suw. The Lexicon has four traditional words with the uw string: juwansa, lauwine, rauwolfia, and vrouw. In juwansa, uw is not a grapheme (the w belongs to the next syllable), and the pronunciation of uw in the last three actually matches the new form.

The soft g. Soft g’s are mostly of French origin, origin, also extended to Latin and Greek words. The rule is straightforward: g is soft before soft vowels e, i, and y, and hard before a, o, and u. Changing these to j will affect too many scientific terms – these are left alone. Words ending in -inge may drop the final e in their gerunds, spelled both as binging/bingeing or hinging/hingeing, but only singeing, because singing would be ambiguous. Words ending in -geon will lose the o and gain a d if needed, e.g. bludgen, dundgen, pidgen, because the pronunciations of dge and dgi are not ambiguous.

The hard g. Euspell maintains gh as hard g, in some cases introducing an h to enforce this, for example rogue → roghe, guide → ghide. This follows the unambiguous Italian convention (as does the now-consistently k-sounding ch) which we believe is more appropriate for English than the French/Spanish gu, because the u is pronounced in many words such as distinguish, segue, etc.

The silent b. The following common words in the Lexicon end in mb where the b is silent: bomb, catacomb, climb, comb, crumb, dumb, jamb, lamb, limb, numb, plumb, recumb, rhumb, succumb, thumb, tomb, and womb. In particular, comb, the only word with an oʊ sound, has no reasonable alternative spelling than cohm. Other words are changed based on the preceding vowel or whether the b is pronounced in derived words. So we have clym, catacohm, cohm, dumm, jamm, lamm, limm, numm, thumm, toom, and woom. Six words are unchanged because some of their derivatives have a non-silent b: bomb, crumb, plumb, recumb, rhumb, and succumb. The b in debt was inserted by Renaissance scribes imitating Latin debitum. Debt is reverted to dett as shown in Table 1.

The useful h. In some languages, a static h (one not followed by a vowel) is pronounced nevertheless, but not in English. This allows us to use this silent h to replace other silent letters while keeping vowel lengths intact. Fruit, suit, recruit, comb, and sew will be fruht, suht, recruht, cohm, and soh. Many French words will have their silent t, s, and x replaced by h.

Words ending in -ost. The o in these words can be short as in cost and lost, or long as in host and most. The natural way of indicating a long o in these words is to add e to the end, as the old spellings of most and ripost are actually moste and riposte. Since post is also a prefix with over 700 words in the Lexicon – many of which are scientific terms – these are left alone, and only the root forms are changed, e.g. poste, outposte. All forms of ghost and host which are not already followed by a vowel will be changed to goste and hoste.

Doubling consonants. Some verb endings trigger the doubling of the last letter in British spelling, but not in American: travelled/traveled, modelling/modeling, etc. In case of the past tense, this is a moot point as both forms collapse into one: traveld, canceld, etc. For the VVG form, these will be made consistent using the following rule: if the final syllable is stressed, double the last letter before -ing: compelling, referring but traveling, pivoting. One caveat is that this rule applies to the base word, even if it receives a stressed prefix, e.g. backfitting, typesetting. Doubling letters in the stem was considered to indicate a preceding short vowel, affecting about 1400 words. Only a subset of these words are changed for other reasons, e.g. heavy → hevvy (to disambiguate ea), with others left alone to avoid inconsistency with derived forms, e.g. study/studious.

The e before -able. The suffix -able is currently added inconsistently to some words: likable/likeable, sizable/sizeable, etc. These will be made consistent with the following rule: if the -ing form drops the e, so should the -able form. This rule does not apply to words ending in -ce or -ge, to avoid hardening the consonants: changeable, noticeable, etc.

The final th. Words ending in th/the have a consistent rule in voicing: -th is voiceless and -the is voiced, e.g. cloth/clothe. The two exceptions are betroth and smooth, which will be betrothe and smoothe. The change in betrothe also fixes the long o problem.

French words. A large number of French words need to be respelled to conform with English. The strings ch and oi are pronounced ʃ and wa in French respectively, and will be spelled as sh and wa: noir → nwar, machine → mashine. The gue and gui forms will be ghe and ghi as mentioned earlier. Boeuf in particular has been left alone because it is pronounced three ways: buːf, boʊf, and bʊf.

Foreign names. The general guideline for foreign names is to apply the Euspell rules only if they are transliterated from a different writing system. Latin-alphabet names are unchanged.

Singletons. Three words have a characteristic not present in any other English word. I is the only always-cap word in English, confused with the Roman numeral one and the letter i as intended. I will be normal-case like other words and spelled ih. Next is of, the only word with a voiced f. It will be spelled ov. The third is shall, the only word ending in -all but rhyming with pal (excluding unstressed syllables). It will be spelled shal.

Apps, extension, and add-ins

No spelling system can be of much use without aids to help in learning and usage. Euspell comes with a comprehensive set of software tools for reading and writing. Table 2 shows the platforms and software availability of Euspell. See Appendix A for installation and additional details.

Table 2. Platform availability

ჱ The Euspell logo is Unicode character 0x10F1, the Georgian letter ‘archaic he’ in blue (RGB 0000ff)

Browser extensions. The Euspell browser extensions convert web pages in real time, with the option of turning it on and off globally. Extensions are available for Chrome, Edge, Firefox, and Safari on desktops.

PDF. The PDF viewer is integrated into the browser extension on Windows, Mac and Linux. On the desktop and mobile devices, the eupub app also opens and converts PDFs.

Epub and text. A standalone app called eupub is created for reading text, epub, and PDF documents displayed in Euspell. It is available for Windows, Mac, Linux, Android, and iOS. PDF files are only shown converted.

Spelling checker. Spelling checker functionality is included in the word processor add-ins. A custom dictionary for new spellings and an exclusion dictionary are available.

Grammar checker. A grammar checker’s first step is POS tagging, and its hardest tagging cases are exactly the homographs Euspell reforms. The main Lexicon contains about 35000 words with new spellings. Many of these words can have more than one PoS, totaling about 46000, contained in the lexicon euspell_pos.tsv to incorporate into grammar checkers.

Speech synthesis. The euspell_tts.pls lexicon specifies the pronunciations of over 35000 words with new spellings in the IPA format, all unambiguous, allowing text-to-speech engines to pronounce new words. See more details at github.com/ossiak/euspell/blob/master/docs/ssml-lexicon.md .

Dictation. Lets a user speak and have their words inserted already spelled in Euspell — an authoring tool, the inverse of the reader. The recognizer produces traditional English; the existing Euspell converter respells it, one sentence at a time. For a detailed discussion and analysis of Euspell dictation, visit github.com/ossiak/euspell/blob/master/docs/dictation.md .

Game. A text-based game to practice your Euspell conversion. Instructions are shown on the web page.

How to support the project

The adoption of Euspell depends on a grassroots effort and support at multiple levels. Here are some ways to help the project:

• Install the software and get used to the new spelling – it is matter of days

• Share this paper and #euspell with others

• If you are a student, ask your instructors to accept your assignments in Euspell

• If you are the parent of a student, ask the school and the school district to recognize Euspell

• If you are an influencer, spread the word, and guide your followers to euspell.org

• If you are an NLP developer, incorporate Euspell in your apps.

• Lobby government agencies to adopt Euspell

• Wear the Euspell T-shirt and hoodie: shop.euspell.org .

Conclusion

The rules of English spelling are not hard to learn, were it not for the numerous exceptions. The main goal of Euspell is to minimize these exceptions and make spelling as consistent as possible within the current rules and spirit of English spelling, and to reduce ambiguities to make reading and comprehension easier with minimum change. The majority of new spellings are only one letter different from the original. A lexicon of around 205000 words, each carrying its parts of speech and its reformed spelling, already provides this consistency. Four fifths of the dictionary is untouched, no new letters or diacritics are introduced, and proper nouns are left alone. A classifier reading the surrounding words settles the entries that genuinely require context, resolving the dominant noun/verb ambiguity correctly in 94% of instances. Euspell was created in the hope of setting the groundwork for a grassroots effort to reform English spelling, without any destructive effect on existing documents. Billions of article, papers, and books are released electronically every year, and being able to read these through the Euspell filter allows everyone to judge it for themselves. This filter works in real time and can be switched on and off easily. Installation of the software is straightforward, and all lexicons, documentation, and code are open-sourced.

Future Work

A few issues are still unresolved. First is distinguishing the two short u sounds ʊ and ʌ (put versus putt). The second case is the of spelling the hard and soft g sounds. Phonetic spelling of these, e.g. get → ghet, gist → jist is too disruptive, with hundreds of scientific words being affected. In both cases the phonetically correct answer costs more in legibility than it returns, and a partial fix would be worse than none. The only gh forms enforcing the hard g are those where h replaces u for the purpose of disambiguation. A third issue is statistical: the noun/verb classifier degrades on minimalist headlines and short phrases, where there is too little surrounding text to read. Beyond the reform itself, the machine-readable outputs are not explored yet. The pronunciation lexicon (about 35000 entries carrying IPA) and the part-of-speech lexicon (about 46000 entries in Penn Treebank tags) are both generated and awaiting integration into apps: the first exists so that a speech synthesizer can read converted text correctly, and the second is intended to teach grammar checkers like Harper [13] and LanguageTool [14], a vocabulary they do not currently possess, rather than to justify building a checker of our own. Real-time spelling and grammar checking inside word processors remains blocked by the host applications. A fourth issue is vowel length consistency: hundreds of words could have been respelled to indicate short vowels which currently may be seen as long, e.g. seven → sevven, honor → honnor, epic → eppic. This change would be too disruptive to implement at this time.

AI assistance. The reform architecture and the main lexicon are created by the author. Anthropic Claude [15] Opus 4.8/5 and Fable 5 assisted with software development, revisions to sections of this paper, installation instructions, website design, and video production. Machine learning is used only to decide which of two or more spellings, when applicable, is chosen in a given context.

Trademark. EUSPELL is a trademark of Kamran Ossia; United States registration is pending. The mark identifies this project and its software. Using the word for the spelling itself — describing text as being in euspell, or writing about the reform — needs no permission and is not restricted.

References

[1] University of Lancaster PoS tagset: ucrel.lancs.ac.uk/claws7tags.html

[2] Penn Treebank: www.ling.upenn.edu/courses/Fall_2003/ling001/penn_treebank_pos.html

[3] International Phonetic Alphabet: www.internationalphoneticassociation.org/content/ipa-chart

[4] Mulcaster, R. “The First Part of the Elementarie”, 1582. ota.bodleian.ox.ac.uk/repository/xmlui/handle/20.500.12024/3176

[5] George Bernard Shaw’s Shavian alphabet, en.wikipedia.org/wiki/Shavian_alphabet

[6] Webster, N. “A Compendious Dictionary of the English Language”, 1806. archive.org/details/compendiousdictionaryoftheenglishlanguage1806

[7] Paine, H.G. “Handbook of Simplified Spelling”, 1920.

www.spellingsociety.org/files/handbook-of-simplified-spelling-1546879176.pdf

[8] Upward, C. “Cut Spelling: A Handbook”, 1996. www.spellingsociety.org/uploaded_books/cs1intro.pdf

[9] Seymour, P.H.K., Aro, M., Erskine, J.M. “Foundation literacy acquisition in European orthographies”, British Journal of Psychology 94, 2003, pp. 143–174.

[10] Landerl, K., Wimmer, H., Frith, U. “The impact of orthographic consistency on dyslexia: A German-English comparison”, Cognition 63, 1997, pp. 315–334.

[11] Paulesu, E. et al. “Dyslexia: Cultural diversity and biological unity”, Science 291, 2001, pp. 2165–2167.

[12] KPMG Foundation. “The long term costs of literacy difficulties”, December 2006.

[13] Harper grammar checker: writewithharper.com

[14] LanguageTool: languagetool.org

[15] Claude: claude.ai

Appendix A. Installing the tools

Everything described in this paper is free software under GPL-3.0 or later for code and Creative Commons Attribution-ShareAlike 4.0 for the lexicons. Everything runs on the reader’s own device. There is no account, no server, and no telemetry: each tool carries its own copy of the lexicon, and every conversion is performed locally.

A.1 Downloads

euspell.org carries the current download links for the tools which have reached a store or a release. The browser extension is on the Chrome web store, also covering Edge, Brave, Opera and other Chromium browsers: chromewebstore.google.com/detail/euspell/jijbndkdmbmomfmgblomkkejjgdnemja .

The software and its documentation are in two public repositories:

A.2 Instructions

Each tool has one install document, kept current with the authoritative version of the code it describes. Where they disagree with this paper, they prevail.

All the builds themselves are available on a single page euspell.org/downloads on the website. For the browser extension: Firefox xpi, the notarized Safari dmg, and a link to the Chrome Web Store listing. For Eupub, a signed Windows installer, a notarized macOS dmg, a Linux AppImage, a signed Android APK, and a link to the App Store listing for iPhone. The same files are available from the github repository.

A.3 Current status

As of August 2026, here is the status of the tool set:

• The browser extension is on the Chrome Web Store, which also serves Edge, Brave, Opera and other Chromium browsers. Safari has no store listing but does have a signed, notarized app to download, drag to Applications, and switch on in Safari's settings. Firefox has neither a listing nor a build step: its Mozilla-signed .xpi is published with the releases, and opening that file in Firefox installs it permanently.

• The word-processor add-ins have no marketplace listing on any of the four platforms. Three are built from source; Word can instead be pointed at a hosted manifest and needs no build at all. They are one-pass converters rather than live spell-checkers: a command rewrites the document, and nothing is underlined as you type.

• Eupub ships prebuilt for the three desktop platforms: Windows (.exe), macOS (.dmg), and Linux (.AppImage). For Android, the .apk is signed with the release key and installs directly. For iOS 17 or later, search the App Store.

• The conversion game is the one item with nothing to install: a single web page that asks the reader to rewrite a passage in Euspell by hand and scores what they produce. It needs no server, no build, and no network, and it is the quickest way to find out how easy the reform is to learn.

A.4 If a link has moved

The repositories are the durable items, and each is self-sufficient: the instructions live in the docs/ directory of the repository they describe, so a clone or a downloaded ZIP carries the software and its documentation together. A reader who can reach either repository needs nothing else from this appendix, and one who can reach neither will find the project’s current address through a search for euspell.

Appendix B. CLAWS7 [1] and Penn TreeBank [2] part-of-speech tagsets

Appendix C. International Phonetic Alphabet (IPA) for English [3]

Stress markers applied before syllables

Appendix D. Euspell encodings and their distribution in the Lexicon