English spelling looks arbitrary because it isn't one system — it's four or five layers of borrowing, each written down by scribes using the conventions of a different donor language, and then frozen in place by printing around 1500. Which means a word's spelling is not a fact about the word. It's a fact about where the word came in.
That's the whole method, and it runs in one direction:
The example words are links — follow one for the full story of how that particular word ended up spelled that way.
Three kinds of evidence, none of which need the spelling. They overlap, and they should agree; when two of them point the same way, that's usually enough to go on.
The strongest of the three. Each layer entered English through a particular trade — conquest, the church, the laboratory, the docks — and the vocabulary mostly stayed where it landed.
| the body, the farm, the weather, the family, the hundred commonest words | Old Englishknee, sheep, wind, daughter |
| blunt everyday things — and a few words too basic to look borrowed | Old Norsesky, knife, window, law |
| court, church, cuisine, fashion, art, anything a ruler did | Norman Frenchvillage, patience, judgment, princess |
| abstraction, law, administration — words written before they're spoken | Latinnation, gravity, various, accept |
| science, medicine, technical coinage | Greekphotograph, chemistry, system, pneumonia |
| sailing, trade, and painting | Dutchdeck, sloop, yacht, easel |
| music, food, and architecture from Italy | Italianpiano, opera, spaghetti, ghetto |
| the Americas — ranching, weather, food | Spanishtornado, canyon, chocolate, tomato |
| mathematics, chemistry, the stars | Arabic, through Spanish or Latinalgebra, algorithm, alcohol, almanac |
Weaker, and deliberately short: most sounds are written more than one way, which is the problem this page exists to solve. These are the few that place a word on their own.
| one syllable, or two with the stress up front — blunt and concrete | Old Englishwolf, water, knee, bridge |
| a hard /sk/ at the start | Old Norsesky, skin, skirt, skull |
| three or more syllables, stress in the middle, audibly assembled from parts | Latinimmediate, obstruct, various |
| a medical or technical word starting with a bare /n/, /t/, /s/, or /z/ | Greek — the letter you can't hear comes firstpneumonia, pterodactyl, psalm, xylophone |
| the stress on the last syllable, and the word simply stops there | Modern Frenchballet, debut, gourmet |
| /ʒ/ — the sound in beige and genre | French, every time. English has no native /ʒ/genre, beige |
| /ʃ/ where an older loan would have said /tʃ/ | Modern French, borrowed after the sound shiftedmachine, chef |
| an unstressed vowel on the end, in food, music, or weather | Italian or Spanishpiano, opera, tornado, cargo |
English often kept all three words for one idea, and they sort by layer: the native one is blunt and concrete, the French one is polite, the Latin one is formal or technical. Find the rung your word sits on and you have found its layer. As a rule of thumb, the shorter and plainer a word feels, the older and more native it usually is.
| Old English | French | Latin |
|---|---|---|
| kingly | royal | regal |
| ask | question | interrogate |
| fire | flame | conflagration |
| holy | sacred | consecrated |
| rise | mount | ascend |
| help | aid | assist |
| house | mansion | domicile |
| go | depart | exit |
One section per layer: a sound you can hear on the left, and the letters that layer writes it with on the right. Each also folds away a table running the other direction — what that layer looks like once it's on the page — which is what step 3 checks against.
the native core, c. 450–1100
Not borrowed at all — this is the language the rest was borrowed into. Almost every one of the hundred most frequent English words is native: the pronouns, the prepositions, be, have, do. The spellings look strange because they are the oldest ones, recording sounds English stopped making 600 years ago.
Writing it
| /f/ | f, the plain letter — never phfish, fire, wolf |
| /ʃ/ | shship, sheep, shirt |
| /tʃ/ after a short vowel | -tchwatch, fetch, ditch |
| /dʒ/ after a short vowel | -dgeedge, bridge |
| /n/ at the start | kn- or gn-. The first letter is silent now and was not in 1400knee, knight, gnaw, gnat |
| /r/ at the start | wr-write, wrist |
| /w/ at the start of a question word | wh-what, when, whale |
| a vowel that runs longer than it should, before nothing you can hear | gh — silent, or said as flight, night, daughter, enough |
| an abstract noun built on a plain word | -ness, -hood, -dom, -shipkindness, childhood, freedom |
| kn-, gn-, wr- | Silent now, fully pronounced in 1400. The letters are a recording of a lost pronunciation.knight, gnat, wrist |
| wh- | Old English spelled it hw-; the scribes flipped the pair.what, when, whale |
| gh | Stands for a throat sound (like the ch in Bach) that English later dropped or turned into f.light, thought, plough, enough |
| sh | The regular modern shape of Old English sc-.ship, fish, shall |
| -dge, -tch | English-only spellings for /dʒ/ and /tʃ/ after a short vowel.edge, bridge, ditch, fetch |
| -ness, -hood, -ship, -dom, -less, -ful | Native suffixes. A Latin root wearing one of these has been naturalised.kindness, childhood, freedom |
| compounds instead of borrowings | The native strategy for new words: glue two old ones together.handbook, sheepdog, bookworm |
| vowel-change plurals and past tenses | A Germanic inheritance older than writing.mouse/mice, goose/geese, sing/sang/sung |
| w- at the start | Latin and Greek roots essentially never begin with w.water, wind, wolf |
Viking settlement, c. 800–1100
The hardest layer to spot, because Norse and Old English were close cousins — the borrowings are short, blunt, everyday words that feel completely native. The giveaway is usually a sound that native words had already softened out of.
Writing it
| /sk/ at the start | sk-, never sc- or sh-sky, skin, skirt, skull |
| hard /g/ or /k/ before e or i | plain g or k. A native word would have softened here, so the hard sound is the loanget, give, gift, kid |
| sk- | Native sc- had already softened to sh-, so a hard sk- came from somewhere else.sky, skin, skirt, skill, skull |
| hard g or k before e/i | Native words palatalised here — the same root gives yield, yell.get, give, gift, kid |
| a doublet beside a native word | Both survived with split meanings. This is the cleanest proof.shirt/skirt, shatter/scatter, whole/hale |
| core vocabulary that isn't Old English | Borrowing pronouns is rare and marks deep contact, not fashion.they, them, their, are, take, want, die, knife, window, law |
after the Conquest, 1066–c. 1400
Three centuries of French-speaking rulers. The vocabulary clusters where power sat: law, government, church, cuisine, fashion, art. Norman scribes also rewrote the spelling of native words they found awkward, which is why some very old words look French.
Writing it
| /s/ before e, i, or y | c — the soft-c convention is Romance, and Old English had no use for itcity, cell, pencil |
| /tʃ/ | chchamber, charge, chase, chief |
| /dʒ/ ending a noun | -gevillage, charge |
| /kw/ | qu-, even in native words the scribes got hold of: queen was cwenqueen, quick, quell |
| hard /g/ before e or i | gu-, with a u wedged in to keep the g hardguard, guest, guide, guile |
| the noun endings | -age, -ance/-ence, -ment, -ery, -ess, -etvillage, patience, judgment, bakery, princess, pocket |
| -age, -ance/-ence, -ment, -ery, -ess, -et | The workhorse French endings.village, patience, judgment, bakery, princess, pocket |
| c said as /s/ before e, i, y | The soft-c convention is Romance; Old English had no use for it.city, cell, cycle, pencil |
| qu- where you'd expect cw- | Pure scribal respelling of native words: queen was cwen, quick was cwic.queen, quick, quell |
| gu- | A u inserted to keep the g hard before a front vowel.guard, guest, guide, guile |
| ch said as /tʃ/ | Early French loans; later French loans say /ʃ/ instead (see below).chamber, chase, charge, chief |
| w- / gu- doublets | The same word borrowed twice — once from Norman, once from Paris.warden/guardian, warranty/guarantee, catch/chase, cattle/chattel |
| the animal/meat split | Native for the beast in the field, French for the dish on the table.cow/beef, pig/pork, sheep/mutton, calf/veal |
learned borrowing, heaviest 1500–1700
Taken off the page rather than out of anyone's mouth, mostly by writers who wanted an abstract or technical vocabulary English lacked. Latin words tend to be long, formal, and morphologically transparent once you know the prefixes.
Writing it
| the abstract endings | -tion/-sion, -ity, -ous, -ate, -al, -ive, -arynation, gravity, various, create, legal, active |
| /ʃ/ buried inside a long formal noun | ti, not shnation, partial |
| /ər/ on a role or agent noun | -or, where a native agent noun would take -eractor, doctor, professor |
| a consonant that sounds doubled just after a prefix | double it. The prefix assimilated to the root: ad+cept, ad+prove, in+mediate, com+lectaccept, approve, immediate, collect, suffer |
| /kt/ | -ct-, a cluster Latin tolerated and English inherited wholeact, fact, direct, instinct |
| a prefix you can hear on the front | pre-, post-, sub-, trans-, inter-, ex-, re-, de-, ab-, ob-predict, transmit, interject, obstruct |
| -tion/-sion, -ity, -ous, -ate, -al, -ive, -ary | The standard Latin suffix set, usually stacked on a Latin root.nation, gravity, various, create, legal, active |
| a doubled consonant after a short prefix | The prefix assimilated to the root: ad+cept, ad+prove, in+mediate, com+lect.accept, approve, immediate, collect, suffer |
| pre-, post-, sub-, trans-, inter-, ex-, re-, de-, ab-, ob- | Latin prepositions doing duty as prefixes.predict, transmit, interject, obstruct |
| -ct- | A cluster Latin tolerated and English inherited wholesale.act, fact, direct, instinct |
| Latin endings and plurals kept intact | The word was never fully naturalised.alumnus/alumni, datum/data, index/indices |
| silent letters that spoil the pronunciation | Renaissance scribes inserted them to display a Latin ancestor. See the caveats below — this is the one cue that is often a lie.debt, doubt, receipt, salmon |
This is the layer where placing the word right still won't spell it. Latin origin licenses a silent letter but never predicts one — nothing in the sound of debt or receipt says a letter is coming that no one pronounces. Those you have to know.
mostly 1500s onward, usually through Latin
The most reliably detectable layer, because Greek came in through Latin transliteration and the transliteration conventions were fixed and consistent. Concentrated in science, medicine, and abstraction. If you have placed a word here confidently, the spelling follows almost mechanically — this is the layer where the guide pays off best.
Writing it
| /f/ | ph, transliterating φ. English had a perfectly good f and pointedly didn't use itphotograph, philosophy, sphere, amphibian |
| /k/ | ch, transliterating χchemistry, character, architect, echo |
| short /i/ in the middle | y, transliterating υ, where a native word would use imyth, system, syntax, crypt |
| /r/ at the start | rh-rhythm, rhetoric |
| a bare /n/, /t/, /s/, /z/ at the start | put the silent letter back on the front: pn-, pt-, ps-, mn-, x-pneumonia, pterodactyl, psalm, mnemonic, xylophone |
| /iː/ or /ɛ/ in an old learned word | ae or oe — American spelling simplifies both to eaesthetic, amoeba, oesophagus |
| a technical compound you could take apart | the combining forms: auto-, bio-, micro-, tele-, -logy, -graph, -meter, -scopebiography, microscope, telemetry |
| ph | Transliterates φ. English had a perfectly good f and didn't use it — that's the tell.photograph, philosophy, sphere, amphibian |
| ch said as /k/ | Transliterates χ.chorus, character, chemistry, echo, ache |
| y as a vowel in the middle | Transliterates υ. Native words use i here.myth, system, physics, syntax, crypt |
| rh-, -rrh- | Transliterates ρ with its rough breathing.rhythm, rhetoric, hemorrhage, catarrh |
| ps-, pn-, pt-, mn-, x- | Initial clusters English can't actually pronounce, so it silences the first letter.psalm, pneumonia, pterodactyl, mnemonic, xylophone |
| ae, oe | Greek diphthongs through Latin. American spelling often simplifies them to e.aesthetic, amoeba, oesophagus |
| auto-, bio-, micro-, tele-, -logy, -graph, -meter, -scope, -phobia | Combining forms, still productive today.biography, telemetry, microscope |
| neuter plurals in -a | The Greek inflection survived the trip.criterion/criteria, phenomenon/phenomena |
Most Greek-looking words are modern coinages assembled from Greek parts, not words any Greek ever said — and the parts are often mixed. Television is Greek tele- plus Latin visio. For spelling that hardly matters: the parts are still spelled the Greek way. It matters when you go looking for the word's history and find it only has one from 1907.
1600s onward, through trade and empire
Borrowed late enough that English mostly kept the donor language's spelling instead of digesting it. That cuts both ways: easy to recognise, but you are spelling someone else's language, so the rule is to write what that language would write rather than what English would.
Writing it
| a final stressed syllable that just stops | Modern French: write the consonant you can't hear, usually -tballet, gourmet, debut |
| /oʊ/ or /wɑr/ at the end of a French loan | -eau, -oirbureau, memoir |
| /ʃ/ in a modern French loan | chmachine, chef |
| an unstressed vowel ending a word about food or music | Italian or Spanish: -o, -a, -ipiano, opera, spaghetti, tornado |
| hard /g/ before e or i, in an Italian word | ghspaghetti, ghetto |
| /nj/ in a Spanish loan | ny, standing in for ñcanyon |
| a rolled or doubled liquid in a Spanish loan | ll, rrguerrilla, armadillo |
| a technical noun about maths, chemistry, or the stars | Arabic: al- welded onto the front, article and allalgebra, algorithm, alcohol, almanac |
| no consonant clusters, every syllable open | Japanese: write it exactly as it soundstsunami, sushi, karaoke, futon |
| -eau, -oir/-oire, ch said as /ʃ/, silent final consonant | Modern French — borrowed after French itself stopped pronouncing final consonants.bureau, memoir, machine, chef, ballet, debut, gourmet |
| sea, trade, and art vocabulary | Dutch and Low German, from centuries of shipping and painting.skipper, deck, yacht, sloop, dock, easel, landscape, boss |
| -o/-a/-i endings, gh for hard g, cc | Italian — music, food, and architecture.piano, opera, graffiti, spaghetti, ghetto, broccoli, balcony |
| -o/-a endings, ll, rr, ny for ñ | Spanish — much of it from the Americas.tornado, cargo, plaza, armadillo, guerrilla, canyon, mosquito |
| al- fused to the front of the noun | Arabic, usually via Spanish or Latin — al is the article, stuck on permanently.alcohol, algebra, algorithm, alkali, almanac, admiral |
| -tl reshaped into -te | Nahuatl, through Spanish.chocolate, tomato, avocado, coyote, chili |
| no consonant clusters, every syllable open | Japanese.tsunami, sushi, karaoke, futon, tycoon |
| schl-, schm-, chutz- | Yiddish and Hebrew, where ch is a throat sound.schlep, schmaltz, chutzpah, kosher, bagel |
| everyday domestic words from India | Sanskrit-descended languages, via the Raj.shampoo, bungalow, jungle, pyjamas, loot, thug, guru |
Now you have letters, so the letters become evidence. Read your spelling back and see which layer it claims to be from; if that isn't the layer you placed the word in, one of the two is wrong. Three questions get you most of the way:
| ph | Greekphotograph, sphere, philosophy |
| ch said as /k/ | Greekchorus, chemistry, architect |
| y as a vowel inside the word | Greekmyth, system, physics |
| rh-, ps-, pn-, pt-, mn- | Greekrhythm, psalm, pneumatic, pterodactyl, mnemonic |
| kn-, gn-, wr-, wh- | Old Englishknee, gnaw, write, wheel |
| gh | Old Englishnight, though, daughter |
| -dge, -tch | Old Englishbridge, watch |
| sk- | Old Norsesky, skin, skill |
| -tion, -ity, -ous, -ate | Latin (often through French)nation, clarity, curious, create |
| -eau, -oir, silent final consonant | Modern Frenchbureau, memoir, ballet |
| al- fused to the front | Arabicalcohol, algebra, algorithm |
| -o ending, ll, rr | Spanishtornado, armadillo, guerrilla |
| doubled consonant after a short prefix | Latin prefix assimilatedaccept, appear, immediate, collect |
These are signals, not proof. The common failure modes:
Placing the word narrows the spelling; it doesn't pick it. Reading letters off a finished word is a lookup. Running the guide forwards, from origin to letters, is a guess with good odds. Placing a word in Greek means ph is worth trying before f, not that ph is correct — fossil and focus are Latin and take the plain letter. Use step 3 to check what you wrote, and the word's own entry to settle it.
Nothing you can hear predicts a silent letter. The b in debt and doubt was never pronounced in English or in the French the words came from; Renaissance editors added it to point at Latin debitum and dubitare. Island is the embarrassing case — it's native Old English igland, and the s was borrowed by false analogy with isle. Compare the genuine silent letters in knight and gnaw, which really do record lost speech. Either way the ear gets no warning, which is the hard limit of spelling from origin.
Compounds fake foreign digraphs. Shepherd has ph and nothing Greek in it — it is sheep plus herd, and the letters just happen to collide. Same with haphazard, uphill, upheaval. Check for a seam before trusting a digraph, in either direction.
Some markers were applied to words that never earned them. Nephew comes from French neveu, but got a learned ph anyway. Ptarmigan is Gaelic tarmachan with a fake Greek pt- bolted on by someone showing off. The cue records what a scribe believed, which was sometimes wrong.
The same root often arrives twice. Once early through French and once later straight from Latin, ending up as two words with different jobs: fragile/frail, secure/sure, hospital/hostel/hotel. Neither spelling is more correct; they're timestamps.
-ough is not a cue, it's damage. Through, though, thought, tough, cough, plough, thorough — all native, all spelled alike, all said differently. When a spelling explains nothing, that is usually the answer: it froze mid-change.
Vowel letters are the least reliable evidence. Printing fixed English spelling around 1500, while the long vowels were still moving through the Great Vowel Shift. That's why the a in name, the i in mine, and the ou in house don't say what the letters say. Judge origin from consonants, clusters, and endings — not from vowels, and don't expect the layer to tell you which vowel letter to write.