wordLAB · English vocabulary

Routes into English

Where a word came from is a different question from how it got in, and both are different again from how often anyone says it. Count the dictionary and English looks Latin. Count a day’s talking and it looks Anglo-Saxon. Follow either one far enough back and most of it converges on a single unwritten language spoken on the steppe five thousand years ago.

Part one

Where the words came from

Origin, then route

The left column is where a word ultimately comes from. The right column is the language English actually took the form from — the only thing a dictionary etymology records directly. Band width is share of the lexicon.

  • Latin at root 50.3%
  • Germanic at root 27.5%
  • Greek at root 9.9%
  • Other or unrecorded 12.3%

Left-column totals and every band are estimates, reconstructed by asking which of the surveyed routes each origin fed. Right-column totals are the survey’s own figures.

Built on 75,150 dictionary entries — every headword in the Shorter Oxford English Dictionary, 3rd edition, classified by Finkenstaedt & Wolff (1973). The right column is that count; the left column is a reconstruction laid over it, adding no new data.

Route of arrival, broken out

The survey classifies each word by its immediate donor — so the direct-versus-indirect split is already built into it. Latin words that came through Paris sit in the French band, not the Latin one.

Route
%

Sub-rows sum to the heading above them. The residual — languages contributing under 1% each — is the survey’s unallocated remainder.

Built on the same 75,150 dictionary entries. Every figure here is counted, not modelled, except the two rows marked est, which are the survey’s own totals minus its itemised rows.

Part two

How far back they go

How much of English reaches Proto-Indo-European? About three fifths.

Almost every language that gave English a word is itself Indo-European — Latin, French, Greek, Norse, Dutch, Sanskrit, Persian. But arriving through an Indo-European language is not the same as descending from a Proto-Indo-European root: each branch picked up vocabulary of its own from the peoples it displaced, and English coins plenty of words from nothing at all. So the honest answer is nothing like 100%.

≈60%
of dictionary entries descend from a reconstructed Proto-Indo-European root. The bars below are the steps down to that figure — not three separate answers.

Built on the same 75,150 dictionary entries for the first step, which is counted. The second step is modelled, and it is the least firm number on this page: it applies published substrate estimates — about 1,100 pre-Greek words in Beekes’ 7,400-entry etymological dictionary, and a contested quarter to a third of core Germanic vocabulary with no accepted Indo-European source — to the routes above.

The family tree, on a time axis

Every branch that fed English, from the steppe to the present. Solid lines are languages; the marks on them are the moments each one is first written down or first does something that shows up in English. Dates before 1000 BCE are conventional estimates and are argued over.

Built on 6 branches, 6 split dates and 15 dated events. Split dates follow the mainstream comparative chronology; they are inferences from linguistic divergence cross-checked against archaeology, not records, and competing chronologies move several of them by centuries.

The same story on the ground

Solid arrows are people moving. Dashed arrows are words moving without them. Latin reaches Britain three separate times — with the legions, with the church, and with the Renaissance — and Greek makes the trip twice: once through Rome, and once, much later, on its own.

Built on 12 movements between 9 places, spanning roughly 4,500 years. Positions are indicative centres, not borders — the peoples involved did not have any. Coastlines are modern.

The history is on the road signs

Every layer of this story settled somewhere and named it. Nowhere else is the sequence so plainly on show: a Celtic river, a Roman fort beside it, an English farmstead over the hill, a Danish one in the next valley, and a Norman lord’s name bolted on afterwards. Select a place to read its name; select a language to see only its own.

    The outline is a schematic drawn from the coordinates of the headlands. The places are at their real positions. The dashed line is roughly the Danelaw boundary agreed between Alfred and Guthrum, running from London to Chester.

    The endings, and what they meant

    Read the ending and you have the settlement: what kind of place it was, and who named it.

    Built on 26 names and 34 elements, following the standard reference works on English place-names. Place-name etymology is an argued field: several of these have more than one proposed derivation, and the ones given here are the mainstream reading.

    Deeper roots are commoner words

    The words that reach furthest back are the ones used most. Nothing in the top hundred of English is younger than the steppe except they and its Norse companions — and those are Indo-European too, just from a different cousin.

    Built on the frequency bands used throughout this page, with Indo-European traceability modelled onto each. Ranks are dictionary lemmas, not tokens.

    Which is why speech looks older than the page

    Weighted by entries, about 60% of English has a Proto-Indo-European root. Weighted by words actually spoken, the figure is nearer 95% — because the borrowings, coinages and unknowns that dilute the dictionary sit almost entirely in the rare tail.

    The practical version for a classroom: the words a child already owns are the old ones. Foot, tooth, heart, mother, three, night, star, name, new, is — every one of those is a direct inheritance from a language nobody wrote down. The Latin and Greek arriving later in their schooling are the same roots coming back round by a longer road.

    Trace a root

    Pick a root. The top lane is inheritance — an unbroken chain of native speech from Proto-Indo-European to the modern word, with the consonants shifted by Grimm’s Law along the way. The two lanes below are borrowings: the word stayed abroad, kept its consonants, and was imported into English thousands of years later.

    • Unbroken inheritance
    • Borrowed into English

    Why the native word looks nothing like the borrowed one

    Because the Germanic branch, and only the Germanic branch, shifted every one of its stop consonants — a single chain of changes that ran to completion before the first century BCE. Latin and Greek sat it out. Grimm’s Law is what makes foot and pedal unrecognisable as relatives, and knowing it turns most of the unrecognisable pairs in English back into obvious ones.

    The first Germanic sound shift. Read across: one PIE consonant, what Germanic did to it, what Latin and Greek left alone.
    PIEGermanicLatin & GreekRootThe same word, three ways

    Part three

    The story in order

    Fifteen centuries, five eras, twenty turns

    Every proportion on this page is the residue of particular events — an invasion, a conversion, a printing press, an empire. The scale below is linear, which makes the point on its own: for the first seven hundred years almost nothing happens to the vocabulary, and then almost everything does.

    Select a marker for the date, and again to jump to its entry. Or filter by era below.

    When the borrowing happened

    Share of the new words first recorded in each half-century, by where they came from. Two crests: French through the thirteenth and fourteenth centuries, then Latin through the Renaissance. After 1700 English mostly stops borrowing and starts building its own words out of the parts it has collected.

      Built on the shape of Philip Durkin’s OED analysis of loanword first-citation dates by 50-year period — French peaking in the second half of the 14th century, Latin in 1500–1650 and again in the 19th. The curves follow his published findings; the individual percentages are mine, and are indicative rather than counted.

      Part four

      Three ways to count the same language

      The dictionary, the head, and the day

      Everything above counts dictionary entries — one line per word, however rare. Two other counts matter more in a classroom: the words one person knows, and the words one person actually uses. The proportions are different in each, and they move in the same direction.

      The same language, counted three ways

      One gradient, three counts. Germanic stock is a quarter of the record, a third of what a person knows, and four fifths of what they say. Latin and French run the other way.

      Built on three different samples. Dictionary: 75,150 entries, counted. Vocabulary: a 42,000-lemma range from 221,268 test-takers (Brysbaert 2016), with origins modelled onto it. Speech: ~14,000 words a day from 2,200 recorded participants (Mehl), with origins modelled from the 100 commonest lemmas tagged individually plus a 10,000-word corpus (Williams 1975).

      How many words does an adult know?

      The figures below disagree with one another, and all of them are right. What changes is the unit being counted.

      Brysbaert et al. (2016), American English. The spread across individuals is wide: 27,100 lemmas at the low end of 20-year-olds, 56,400 at the high end of 60-year-olds. An educated adult sits near the top of that range.

      Built on 221,268 test-takers returning 265,346 sessions, tested against a list of 61,800 lemmas and 18,269 word families.

      Words spoken in a day

      14,000
      Each square is 1% — about 140 words.

        Mehl’s recording studies: about 15,900 a day in the 2007 sample, about 12,700 by 2019, falling roughly 338 a year. 14,000 is the midpoint.

        Built on 2,200 participants across 22 recording studies for the daily total; the origin split is modelled, not counted — the 100 commonest conversational lemmas tagged one by one, the rest weighted from a 10,000-word corpus.

        Germanic share, by frequency band

        Native stock owns the high-frequency core and thins steadily down the ranks. This single curve is the mechanism behind all three counts — where you take your sample from the curve decides what answer you get.

        Built on 10,000 words of business correspondence analysed by Williams (1975), which anchors the top and bottom of the curve at 83% and 25%; the two middle bands are interpolated.

        The hundred commonest words, every one tagged

        Not a percentage to take on trust — the actual words, so you can check them. This is what the 82% figure above looks like when you can see it. Hover a word for its rank and origin; click a total to pick that group out of the crowd.

          Built on 100 words — the commonest hundred lemmas of written and spoken English, tagged one at a time against standard etymologies. Counted, not modelled. Between them they account for roughly half of everything anyone says.

          Old words come first

          Words placed by roughly the age at which a child acquires them. The gradient is the argument for teaching roots in Years 5 and 6 rather than earlier: that is exactly where the Latin and Greek layer arrives.

          Built on 39 individually placed words. Ages are approximate and illustrative, not measured — the ordering follows published age-of-acquisition norms (Kuperman et al., 30,000 words), but each word here is placed by judgement, so read the gradient and not the individual positions.

          How the two modelled counts were built

          No survey counts spoken or known words by etymology, so those two bars are a model, not a measurement. Both are built from the band curve above. For speech, the bands are weighted by how many words each contributes to a day’s talking — the top hundred lemmas alone carry about half of it, and all but a handful are Old English or Old Norse. For a personal vocabulary, the bands are weighted by how many different words each contributes to a 42,000-lemma range, which is why it lands between the other two.

          The single most under-stated figure in the dictionary table is Old Norse. It supplies 1.8% of entries, but among those entries sit they, them, their, are, take, get, give, want, both, same. Weighted by use it is worth perhaps six or seven per cent of everything said — roughly four times its share of the page. English borrowed pronouns from Norse, which languages almost never do.

          Register climbs as you move down the frequency bands, which is why the ladder is teachable: ask → question → interrogate; rise → mount → ascend; kingly → royal → regal. Native, French, Latin — every time.

          Part five

          The words themselves

          The register ladder

          The same meaning three times over: the native word, the French word, the Latin word. Formality climbs as the source gets further from home, and the pattern is regular enough to teach as a rule.

          Built on 24 sets. Not every meaning has all three rungs, and a few pairs here are near-synonyms rather than exact ones — the ladder is about register, not identity.

          The animal and the meat

          The most quoted example of the ladder, and the one worth handling carefully. In the field the animal keeps its English name; on the table the meat takes a French one.

          What is actually true here. The pairs are real and the pattern is real. The explanation usually attached to them is not quite. It comes from Walter Scott, who put it into the mouth of a swineherd in the opening chapter of Ivanhoe: the Saxon keeps the beast and the Norman eats it. But the French words are not recorded in English until around 1300 and later — a century and more after the scene Scott set — and when they arrive they mean the living animal too. Beef was a cow on the hoof as readily as a joint on the table, and veal could still mean a live calf in the nineteenth century. The tidy division of labour between the two words hardened slowly, across the fourteenth and fifteenth centuries, once French had become the language of the kitchen rather than of the conqueror.

          Why it is still worth teaching. Because the shape of the thing is right even though the date is wrong. English ended up with two words where one would do, and it kept both by giving them different work — which is the register ladder in miniature, and the single clearest case of it in the language.

          Built on 8 pairs, with first-attestation dates and the Scott attribution following T. Őrsi, Cow versus Beef: Terms Denoting Animals and Their Meat in English.

          Route leaves a mark on the word

          One Latin etymon, two arrivals: the learned form taken straight from Latin beside the worn form that came through French. The pairs are the evidence that route is real and not a bookkeeping convention.

          The long chains

          Some words changed hands three or four times. Each transfer reshaped the form, which is why the ancestor is no longer visible in it.

          Part six

          The parts words are made of

          Twenty prefixes and twenty suffixes

          Affixes are not evenly used. Four prefixes account for about three fifths of every prefixed word in English, and four suffixes for about two thirds of every suffixed one. Teach those eight and a pupil can attack most of what they meet.

          Prefixes, by share of all prefixed words

          Built on White, Sowell & Yanagihara’s count of the prefixed and suffixed words in a corpus of school reading material (1989). Percentages are of prefixed words only, not of all words.

          Suffixes, by share of all suffixed words

          Built on the same 1989 count. The top three are inflectional — they change grammar, not meaning — which is why they dominate and why they teach differently from the rest.

          A hundred roots

          Seventy-five from Latin, twenty-five from Greek, chosen for how much English each one unlocks. Roots are given in the form a pupil meets them: a grey e is a final e that belongs to the root but drops before a vowel suffix — scribe but scribing, pose but posing, cede but ceding. Greek roots are given without their connecting o, which belongs to the join rather than to the root. Search a root, a meaning, or any example word.

          Built on 100 roots and 388 example words. Forms are given as they appear in English rather than as full Latin or Greek citation forms — duc/duct rather than dūcere — because that is the shape a pupil actually meets.

          Part seven

          What all this did to spelling

          Reading origin off the spelling

          Every wave of borrowing brought its own spelling conventions and English kept all of them, which is most of why the system looks arbitrary. The compensation is that the spelling usually announces something about where a word has been. Two different things are on show below, and they are worth keeping apart.

          Where the word came from

          Where the spelling came from

          These are French scribal habits, not French words. After 1066 the clerks writing English were trained in French, and they respelled what they heard using the conventions they knew — and invented what they needed where French had nothing to offer. So house, queen, come and love are Old English words in French clothing — the spelling is borrowed even though the word is not. This is the single most useful thing to be able to tell a class about why English spelling is the way it is.

          Built on 40 patterns. These are strong tendencies, not laws — ache is Greek-derived but spelt as if native, and sketch reached English from Dutch by way of Italian rather than from Norse.

          Now read one yourself

          Everything above this point is the evidence. This is the use of it: a word, four possible sources, and the marks in the spelling that give it away. Twenty words, in a different order every time.

           

          Built on 20 words, each with the spelling evidence set out in part seven. Where a word has travelled through more than one language the answer is its immediate source into English, which is the one the spelling records.

          Caveats

          The French–Latin boundary is soft. Dictionary etymologies frequently cannot decide between the two for learned borrowings of the late Middle English period, when both routes were open at once. The two large columns trade words at their edge.

          Neo-Latin is filed as Latin. Much modern scientific vocabulary was minted in English or French from Latin and Greek parts. Those words never passed through Latin at all, but the survey counts them there.

          The unrecorded 12% is not evenly spread. Words with no etymology given, and words from proper names, cluster by register and period rather than scattering across the lexicon.

          Reconstructions are hypotheses, not records. Nobody has heard Proto-Indo-European. The asterisked forms are the linguists’ best account of what must have existed for the attested descendants to look as they do, and handbooks write them differently — you will meet *pod-, *pṓds and *ped- for the same root.

          Similar-looking is not related. Sound correspondence, not resemblance, is the evidence. English have and Latin habere mean the same thing and look identical, and are not cognate — Grimm’s Law forbids it. English have answers to Latin capere, ‘seize’. The method is regular, and a rule that lets you predict beats a resemblance that lets you guess.

          Two of the three counts are modelled. Only the dictionary bar is measured. Treat the vocabulary and speech splits as orders of magnitude — the gradient between the three is solid and well evidenced, the second digit of any one of them is not.

          Sources: Finkenstaedt & Wolff, Ordered Profusion: Studies in Dictionaries and the English Lexicon (1973), as tabulated at Foreign language influences in English; Williams, Origins of the English Language (1975); Mehl et al., Are Women Really More Talkative Than Men? (Science, 2007) and its 2005–2019 replication, reported by the University of Arizona; Brysbaert, Stevens, Mandera & Keuleers, How Many Words Do We Know? (Frontiers in Psychology, 2016). Root forms and cognate sets follow Watkins, The American Heritage Dictionary of Indo-European Roots, and Mallory & Adams, The Oxford Introduction to Proto-Indo-European, cross-checked against the Online Etymology Dictionary; substrate figures from Beekes’ Etymological Dictionary of Greek as summarised at Pre-Greek substrate. Figures marked as estimates or described as modelled are reconstructions, not survey output.