Case Study · Linguistics · Graphs
Etymology Graph Explorer
The idea
Every word is the endpoint of a journey: inherited from a parent language, borrowed from a neighbor, derived from a root, or calqued piece-by-piece from a foreign template. Those journeys form a graph (lexemes as nodes, the historical relationships as typed edges) and etymological dictionaries are, structurally, serializations of that graph into alphabetized prose. This project deserializes them: it draws the graph and lets you travel it, across the three languages I speak; English, Spanish, and Russian.
That trio is not just autobiographical convenience. It spans three branches of Indo-European (Germanic, Romance via Latin, Slavic), which means the deep cognates are genuinely deep; you have to climb all the way to Proto-Indo-European to find the common ancestor; while the borrowing traffic between the branches is dense and layered. The same Latin or Greek root routinely reaches all three languages by completely different routes, and seeing those routes diverge and reconverge is the point of the whole instrument.
The explorer
Searches every family, by word, meaning or language.
English holds this root three ways: inherited, via Latin, via Greek; Russian inherits it as 'take' and builds its word for a cathedral out of it; Spanish knows it only from the classroom.
Hover a word, or focus the figure and use the arrow keys.
- inherited 12
- borrowed 12
- derived 6
- reconstructed 5
Counts are the edges drawn in this family. Select a mark to hide that kind of edge. A starred form is reconstructed: inferred from its descendants, not attested in any text.
14 curated families · 278 words
Travel the graph above first; the sections below are how it is built and what a few of its families mean.
The approach
The architecture question came first, because there are two very different projects hiding under
this title. Version 1, shipping first, is curated: a few hundred hand-verified words
organized into rich families. PIE roots with reflexes across the branches, Latin doublets in
English, Romance and Slavic borrowings; each family checked against the standard etymological
dictionaries (Vasmer for Russian, Corominas for Spanish, the OED’s etymologies for English).
Version 3, planned for someday and shipped sooner than that (see below), is extracted: a
pipeline over Wiktionary’s etymology sections, which helpfully encode relationships in
machine-readable templates (inh for inherited, bor for borrowed, der for derived)
wrapped in decidedly less machine-readable prose.
The discipline is that v1 must be a valid output of the v3 pipeline. Nodes are lexemes tagged with language and period; edges are typed; inherited, borrowed, derived, calque; and carry their sources. The curated dataset is hand-written JSON conforming to exactly the schema the extractor will eventually emit, using Wiktionary’s own language codes so the two datasets are drop-in interchangeable:
{
"nodes": [
{ "id": "ine-pro:*bʰer-", "lemma": "*bʰer-", "lang": "ine-pro", "period": "reconstructed", "gloss": "to carry" },
{ "id": "en:bear", "lemma": "bear", "lang": "en", "period": "modern", "gloss": "to carry; give birth to" }
],
"edges": [
{ "from": "ine-pro:*bʰer-", "to": "en:bear", "type": "inherited",
"via": ["gem-pro:*beraną", "ang:beran"], "source": "OED, s.v. bear, v." }
]
}
Designing for the pipeline from day one costs almost nothing now and buys the only thing that matters later: when the extractor exists, the explorer doesn’t change; only the data does.
A worked family: PIE *bʰer-
The reconstructed root *bʰer- meant “to carry,” and it’s a textbook specimen because it survives everywhere. In Germanic, Grimm’s law turns bʰ into b: Proto-Germanic *beraną → Old English beran → English bear, with burden (that which is carried) and birth (a carrying to term) as derivatives of the same root. In Slavic, the root arrives as Proto-Slavic *bьrati, first person *berǫ, with a semantic shift from carrying to taking: Russian брать / беру “to take,” alongside бремя “burden,” which keeps the older sense. Both of these are inherited edges; unbroken chains of parent-to-child transmission.
Latin turns initial bʰ into f, giving ferō, ferre “to carry”; and here the graph gets interesting, because ferre died out in spoken Romance (Spanish says llevar). Spanish therefore has no inherited reflex of this root as a living verb; what it has instead are learned borrowings taken back from written Latin; transferir, referir, preferir, conferir, fértil; which the schema marks as borrowed edges, not inherited ones. English borrowed the same Latin material by its own route: transfer, refer, confer, fertile. And Greek phérō “I carry,” the same PIE root again, reaches English inside metaphor; literally a “carrying across.”
So one PIE node fans out into a single connected family where English holds the root three ways; inherited (bear), borrowed from Latin (transfer), borrowed via Greek (metaphor). Russian inherits it, and Spanish knows it only through the classroom. That difference between routes is invisible in a dictionary entry and immediate in a graph. The fourth edge type shows up in families like skyscraper, which Spanish and Russian each rebuilt from native parts (rascacielos, небоскрёб) the calque edge: borrowed structure, native material.
The v3 pipeline
The extractor exists now, and the day-one discipline paid out exactly as hoped: the explorer
didn’t change; only the data did. The pipeline (scripts/etymology-v3/ in the site’s
repository) streams the
wiktextract per-language dumps for thirty languages; the three focus
languages, their attested ancestors from Old English to Old Church Slavonic, and the
proto-languages; and compiles Wiktionary’s templates into a graph of five million lexemes
and a million typed edges, restricted to the schema’s language world. A family builder then
walks the graph from a seed (a PIE root, a Latin noun, an Arabic loan-source), keeps every
descendant that lies on an attested chain to a modern English, Spanish, or Russian word, and
emits family JSON that validates against the same schema as the curated tier.
The extracted families live in the explorer under their own label: “machine-extracted;
unreviewed,” and now every node carries the same claim about itself, so the label is
per-word rather than per-family. That matters because a reader arrives at a word rather than at a
family: whatever the site is willing to assert about where a form came from has to travel with the
form, into a hover readout and out of any screenshot somebody takes of it. The label is
load-bearing, because the extractor faithfully
repeats whatever Wiktionary asserts, including assertions no etymological dictionary would
bless. Wiktionary’s {{der}} template (“derived, mechanism unspecified”) maps to a dotted
edge with exactly that note, rather than being upgraded to a claim the source never made.
The review has since run, and what it caught is the better argument for the label than any description of it could be. English starboard had been filed under *h₂stḗr, “star,” when it is the steering side of a boat. Old English hund was glossed “dog” and was quietly dragging hound and greyhound into the family for “hundred,” which is two different Proto-Germanic words that happen to look alike. An Old French word for water was glossed “mare,” which is a different word entirely, and the ancestor of eau. One family turned out to be mostly surname jokes about Mark Zuckerberg. Promoting a family means checking every chain against Vasmer, Corominas, and the OED, and eight families have now made that trip; the rest keep the label until they have. (The extracted data inherits Wiktionary’s CC BY-SA 4.0 license, an exception to the site’s CC BY-NC content license, noted in the data files themselves.)
The families
The curated families are live in the plate at the top of this page, every edge cited in the data files; the plate counts itself, under the legend, so the figure and the prose cannot disagree. Among them: *bʰer- (the worked example), the three-language inherited trio night / noche / ночь, the kinship root *ph₂tḗr → father / padre, where Russian conspicuously drops out (отец is a nursery word) and rejoins only as патриот, the tea-by-sea-cha-by-land split (tea / té vs. чай), the zero/cipher doublet out of Arabic ṣifr, and water’s road to both водка and whisky. The calque families (skyscraper → rascacielos → небоскрёб; hydrogène → водород) demonstrate the fourth edge type. Four more (*méh₂tēr, *ḱerd-, *pṓds, *tréyes) are drafted and deliberately unpublished until every chain in them has been read against the reference dictionaries, which is the same rule the tiering is built on.
Hovering or arrow-keying any word shows what the data can support about it. Where a lemma carries a hand-checked transcription it appears in the mono face beside the word; English falls back to CMUdict and is labelled as machine-derived, so the two tiers of confidence stay visible even down at the phoneme, and a word with no defensible transcription simply shows none rather than an empty pair of slashes. The same readout names the language: for anything that is no longer spoken, it gives what the language was, when it was written, and where, since a node labelled OCS or P-Gmc is otherwise a code and not a fact. The explorer also lives full-width at /etymology.
The other axis: how meanings move
Everything above is about form. An edge in the graph asserts that one shape descended from another, and the glosses hanging off the nodes are labels: they tell you which word you are looking at, they do no work. That leaves out half of what happened. A form and its meaning travel independently, and a word can keep its shape almost perfectly while its sense walks somewhere unrecognizable. Latin nescius “ignorant” to English nice is a trivial piece of phonology and a spectacular piece of semantics, and the explorer above draws only the trivial half.
The second half is harder to draw because it is much less regular, and it is worth being precise about how much less. Sound change is the closest thing historical linguistics has to a law: Grimm’s law applies to every eligible word in Germanic, and the residue of apparent exceptions turned out to be a second law (Verner’s) rather than noise, which is exactly the outcome that earns a generalization the name. Semantic change has nothing of that kind. What it has instead is a taxonomy of recurring shapes, named since Bréal put sémantique into circulation in 1897 and stable ever since: narrowing and broadening, pejoration and amelioration, and bleaching, where the content drains out of a word and leaves a grammatical job behind. These are strong tendencies, and some of them even have a direction; bleaching in particular runs one way, since intensifiers wear down into noise and do not spontaneously reacquire content. But not one of them lets you predict what a given word will do next, and that asymmetry is the whole difference between a law and a pattern.
17 words · pick one
nice the canonical wanderer
A Latin word for ignorant enters English as an insult and leaves as the blandest compliment in the language. Harper's own caution is worth keeping in view: in many sixteenth and seventeenth century examples it is impossible to say which sense the writer meant.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
Classical Latin not knowing, ignorant obsolete
Latin nescius
Latin nescius "unaware, ignorant", transparently ne- "not" plus scire "to know".
Plotted at the turn of the era for convenience. Classical Latin words are not dated by first citation the way English ones are, so this node carries no attestation claim at all.
Etymonline, s.v. nice (adj.)
-
12c. careless, weak, simple, foolish obsolete
Old French nice
Old French nice "careless, clumsy; weak; poor, needy; simple, stupid, silly, foolish". The insult is already fully formed before English borrows it.
Etymonline, s.v. nice (adj.)
-
late 13c. foolish, ignorant, senseless obsolete
The first English sense, and an insult: "foolish, ignorant, frivolous, senseless". A neighboring sense "timid, faint-hearted" is attested before 1300.
Etymonline, s.v. nice (adj.)
-
late 14c. fussy, fastidious; dainty, delicate obsolete
"Fussy, fastidious" by late 14c., "dainty, delicate" by c. 1400. The word has stopped meaning stupid and started meaning particular, which is the hinge of the whole history: to be fastidious is to notice small differences.
Etymonline, s.v. nice (adj.)
-
1500s precise, careful restricted
From noticing small differences to making them: "precise, careful". This is the sense preserved in a nice distinction and, oddly, in nice and early.
Etymonline, s.v. nice (adj.)
-
1769 agreeable, delightful sense used today
The modern sense, and the one that swallowed the rest. By 1926 Fowler was complaining that the word had had all its individuality charmed out of it and had become "a mere diffuser of vague and mild agreeableness".
Etymonline, s.v. nice (adj.), quoting Fowler (1926)
-
1830 kind, thoughtful current
Applied to people rather than things, sixty years after the word learned to mean agreeable at all.
Etymonline, s.v. nice (adj.)
Where it stands nowToday it means agreeable or kind. The old precision survives in one fossil, a nice distinction.
silly blessed to foolish
The textbook pejoration, and it runs entirely downhill: happy to blessed to pious to innocent to harmless to pitiable to weak to foolish. German kept the cognate selig "blessed" exactly where English found it, which is the cleanest possible control: nothing in the word forced the slide.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
Old English happy, fortunate, prosperous obsolete
Old English gesǣlig
Old English gesǣlig "happy, fortuitous, prosperous", from sǣl "happiness". Plotted at a conventional midpoint of the attested Old English period rather than at a citation.
Etymonline, s.v. silly (adj.)
-
c. 1200 blessed, pious, innocent obsolete
form seli
Middle English seli, seely. Harper's chain runs "happy" through "blessed" and "pious" to "innocent" here.
Etymonline, s.v. silly (adj.)
-
late 13c. harmless; pitiable obsolete
The pivot. To be innocent is to be harmless; to be harmless is to be defenseless; to be defenseless is to deserve pity. Each step is small and each step is down.
Etymonline, s.v. silly (adj.)
-
c. 1300 weak, feeble obsolete
Pity has become a judgement about capacity. The sense stayed alive in dialect long after this, used of sheep: a silly sheep is a defenseless one, not a stupid one.
Etymonline, s.v. silly (adj.)
-
1570s feeble in mind, lacking in reason, foolish sense used today
Weakness of body becomes weakness of mind, and the modern sense is in place. Roughly six centuries of attestation from blessed to stupid, with no step in the chain that a speaker would have noticed happening.
Etymonline, s.v. silly (adj.)
-
1886 dazed, stunned, as by a blow current
Knocked silly. A late offshoot of the foolish sense, not a continuation of the old chain. Harper marks it as probable rather than certain.
Etymonline, s.v. silly (adj.)
Where it stands nowToday: foolish, or knocked senseless. The blessing is entirely gone.
deer any beast to one beast
Old English dēor was the ordinary word for animal, the way German Tier still is. The animal we now call a deer was a heorot, a hart. Hunting culture did the rest: the beast worth naming became the only beast the word could name.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
Old English wild animal, beast, any wild quadruped obsolete
Old English dēor
The general Germanic word for a creature as opposed to a person, from Proto-Germanic *deuzam; compare Dutch dier, German Tier, which never narrowed.
Etymonline, s.v. deer (n.)
-
late Old English specifically, the animal of the chase sense used today
Harper: the specialization to a specific animal "began in Old English (the usual Old English word for what we now call a deer was heorot), was common by 15c., and is now complete". The narrowing starts here and then takes four hundred years to finish.
Etymonline, s.v. deer (n.), s.v. hart (n.)
-
early Middle English any animal at all, including ants and fish obsolete
At the very moment the narrow sense is spreading, the wide sense is at its widest: early Middle English applies the word to ants and to fish. Drift is not a queue, and a word does not finish one sense before starting the next.
Etymonline, s.v. deer (n.)
-
15c. the narrow sense is the common one current
Not a new sense but a milestone: by the fifteenth century the specialized reading is the default, and the general one is on its way out.
Etymonline, s.v. deer (n.)
-
1605 the wide sense survives as deliberate archaism obsolete
"But mice and rats, and such small deer, / Have been Tom's food for seven long year", King Lear III.iv. Edgar is talking like an old ballad on purpose, which tells you the sense was already antique on stage.
Shakespeare, King Lear, III.iv (1605)
Where it stands nowToday: one family, Cervidae, and nothing else.
meat all food to one food
The same narrowing as deer, in the same centuries, in the same semantic neighborhood, and for a related reason: when a culture cares enormously about one member of a category, the category name goes to it. Old English mete was food of every kind; flesh was the word for flesh.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
Old English food of any kind; nourishment; fodder restricted
Old English mete
Old English mete "food, nourishment, sustenance", habitually paired with drink, and also "animal food, fodder" and "a meal".
Etymonline, s.v. meat (n.)
-
c. 1300 the flesh of animals used as food sense used today
The narrow sense arrives. Before this the specific idea needed the compound flesh-meat, attested early 12c., literally "flesh-food", which is a fair summary of the whole change: the modifier eventually ate the head.
Etymonline, s.v. meat (n.)
-
15c. compounds keep the wide sense alive restricted
White meat "a dairy food or product" (early 15c.), grene-mete for vegetables (15c.), swete mete for a confection. Compounds are where obsolete senses go to be preserved; they are the amber of semantic history.
Etymonline, s.vv. meat (n.), sweetmeat (n.)
-
1611 the wide sense, fossilized in scripture obsolete
"I have given you every herb bearing seed ... to you it shall be for meat" (Genesis 1:29, King James Version). Herbs are meat. Three hundred years of readers have quietly reinterpreted that line.
King James Bible (1611), Genesis 1:29
-
1901 the essential part, the substance current
The meat of the argument. A figurative extension out of the narrow sense, not a return to the wide one.
Etymonline, s.v. meat (n.)
Where it stands nowToday: animal flesh. The wide sense survives only in compounds and fixed phrases, sweetmeat, meat and drink, one man's meat.
dog one breed to every canine
The mirror image of deer, and the standing mystery of English etymology. Docga is a late, rare Old English word of no known origin, with no cognates anywhere, apparently naming one powerful breed. Within five centuries it had forced out hund, the ordinary Germanic and Indo-European word for the animal, which retreated into hound, a dog that hunts by scent. Broadening and narrowing here are two halves of one event.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
late Old English a powerful breed of dog obsolete
Old English docga
"A late, rare word" in Old English whose origin, Harper writes, "remains one of the great mysteries of English etymology". It has no accepted Germanic ancestry and no relatives.
Etymonline, s.v. dog (n.)
-
c. 1200 in abuse: a mean, worthless fellow current
"By c. 1200 in abuse or contempt as 'a mean, worthless fellow, currish, sneaking scoundrel'." The insult is attested about as early as the animal, which is worth pausing on: the pejorative is not a late corruption of a neutral word but a companion of it from the start.
Etymonline, s.v. dog (n.)
-
by 16c. any canine; forces out Old English hund sense used today
The broadening completes. In the same century the word is picked up by continental languages in its old narrow sense: French dogue, German Dogge, Danish dogge all name a heavy mastiff-type breed. The borrowers froze the sense English was in the act of leaving.
Etymonline, s.v. dog (n.)
-
1610s a rakish man; a sport, a gallant restricted
Harper calls it the "playfully abusive" sense. The same word carrying an insult in one register and a compliment in another, at the same time.
Etymonline, s.v. dog (n.)
-
1936 something poor or mediocre; a failure current
The film was a dog. American slang, and the insult generalized from people to things.
Etymonline, s.v. dog (n.)
Where it stands nowToday: the generic term, and a productive insult.
very true, worn smooth
Bleaching is drift with the meaning drained out rather than moved: a word with content becomes a word with a job. Very arrives from Latin verax "truthful" carrying a real claim, that a thing is genuinely what it is called. Chaucer's verray, parfit gentil knyght (General Prologue, line 72) is a true knight, not a very knight. Within a century the truth claim has evaporated and only the emphasis is left.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
Classical Latin truthful obsolete
Latin verax
Latin verax, from verus "true", from a Proto-Indo-European root *were-o- "true, trustworthy". Plotted at the turn of the era for convenience; no attestation claim is made.
Etymonline, s.v. very (adj.)
-
12c. true, truthful, sincere; right, just, legal obsolete
Old French verai
Anglo-French verrai, Old French verai, through Vulgar Latin *veracus. The legal sense is a reminder that this was a word you could swear by.
Etymonline, s.v. very (adj.)
-
late 13c. true, real, entitled to the name, genuine restricted
form verray
"Entitled to the name" is the exact force: a verray knight is one who deserves the word knight.
Etymonline, s.v. very (adj.)
-
late 14c. actual, sheer restricted
The first slippage. "Actual" and "sheer" are still claims about identity, but weaker ones; the word is starting to point rather than assert. Harper notes it was already working as a pure intensive by late 14c.
Etymonline, s.v. very (adj.)
-
mid-15c. greatly, extremely sense used today
The bleaching, complete: an adverb of degree with no truth claim in it at all. Very tired asserts nothing about truth, and the sentence loses almost nothing if you delete the word.
Etymonline, s.v. very (adj.)
-
1690s the very thing: just what is wanted current
A late idiom that still leans on the old identity sense. Sixteenth century writers had already gone further and coined the comparatives verier and veriest, treating truth as a matter of degree.
Etymonline, s.v. very (adj.)
Where it stands nowToday: a degree word with no content of its own. The old sense survives in verily, veritable, and the Nicene Creed's very God of very God.
knight boy to peer of the realm
Social history doing work that sound change cannot. Old English cniht was a boy or a servant; German kept the cognate Knecht as "farmhand, bondman" to this day. English attached the word to the armed retainers of a lord at exactly the moment those retainers turned into a landed class, and the word rose with them.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
Old English boy, youth; servant, attendant obsolete
Old English cniht
Common West Germanic: Old Frisian kniucht, Dutch knecht, Middle High German kneht. German Knecht still means servant or bondman, which is the control case.
Etymonline, s.v. knight (n.)
-
c. 1100 military follower of a king or other superior obsolete
Still a servant, but an armed one, and attached to somebody who matters. Every later step follows from whose servant he is.
Etymonline, s.v. knight (n.)
-
14c. a man of honorable military rank, bound by a code restricted
Harper says the word "began to be used in a specific military sense in the Hundred Years War" and rose gradually from there. He gives no single citation, so this node is plotted at mid-century and is softer than its neighbors.
Contested placement. The sense is a slow social process, not an event, and any single year here is an editorial choice.
Etymonline, s.v. knight (n.)
-
mid-15c. the horse-headed chess piece current
The piece that moves like a mounted retainer. Other languages named the same piece for the horse (Russian конь) or the jumper (German Springer).
Etymonline, s.v. knight (n.)
-
16c. a rank in the nobility current
The word now names a position in a hierarchy rather than a function in a household.
Etymonline, s.v. knight (n.)
-
modern British an honorary dignity conferred by the sovereign sense used today
Harper describes it as "a social privilege or honorary dignity conferred by a sovereign as a reward, without regard for birth or deeds at arms". The military duty is gone entirely; what is left is the honor. He gives no first citation, so this node is plotted at a conventional nineteenth century point and should be read as a period, not a date.
Undated in the source. Placed by editorial judgement, and the weakest date on this rail.
Etymonline, s.v. knight (n.)
Where it stands nowToday: an honor conferred by the sovereign, with no military duty attached, and a chess piece.
villain farmhand to scoundrel
The exact opposite trip to knight, over the same centuries, driven by the same social arrangement seen from below. Klein's summary of the chain is hard to improve on: "inhabitant of a farm; peasant; churl, boor; clown; miser; knave, scoundrel." A word for where you live becomes a word for what you are.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
Medieval Latin farmhand obsolete
Latin villanus
Medieval Latin villanus "farmhand", from Latin villa "country house, farm". Purely locative: the man attached to the villa. Plotted at a conventional Carolingian point.
Etymonline, s.v. villain (n.)
-
12c. peasant, commoner, churl, yokel obsolete
Old French vilain
Anglo-French and Old French vilain. The contempt is already loaded in French before English takes the word.
Etymonline, s.v. villain (n.)
-
c. 1300 a base or low-born rustic obsolete
The first English sense, and a statement about class rather than character. Attested as a surname a century earlier, late 12c.
Etymonline, s.v. villain (n.)
-
early 14c. a feudal tenant bound to the land restricted
form villein
A spelling that split off and kept the legal meaning. English ended up with two words from one, villein holding the tenure and villain taking the abuse: the orthography is a fossil of the semantic split.
Etymonline, s.vv. villain (n.), villein (n.)
-
mid-16c. a scoundrel, a man capable of gross wickedness sense used today
Class has become morals. The claim is no longer that he farms but that he is capable of anything.
Etymonline, s.v. villain (n.)
-
1822 the character in a story whose evil drives the plot current
A technical sense out of criticism and the stage: not a wicked man but a structural role. Drift can end in a job title.
Etymonline, s.v. villain (n.)
Where it stands nowToday: a wicked person, or the character in a story who supplies the trouble.
awful full of awe to merely bad
Two changes at once: the word gets worse and then gets weaker. What makes it useful here is the control group standing right beside it. The noun awe did not move at all, and the sibling adjective awesome made the same trip in the opposite direction, from "profoundly reverential" (1590s) to "impressive, very good" (by 1961). Same root, same language, three different fates.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
c. 1300 worthy of respect or fear; causing dread restricted
form agheful
From aghe, an earlier form of awe, plus -ful; the Old English word for the idea was egefull. Said of God, of kings, of the sea.
Etymonline, s.v. awful (adj.)
-
1590s profoundly reverential, full of awe obsolete
The other direction of the compound: not causing awe but feeling it. Harper calls the use occasional even at the time. Its loss is why awful could not stay ambiguous.
Etymonline, s.v. awful (adj.)
-
1809 very bad sense used today
The weakening. Dread has become distaste, and the theology has drained out of the word entirely.
Etymonline, s.v. awful (adj.)
-
1818 excessively; very great current
Now purely a matter of degree, and applicable to good things: an awful lot.
Etymonline, s.v. awful (adj.)
-
c. 1830 awfully: simply an intensifier current
Awfully kind of you. The adverb, which meant "so as to inspire reverence" in c. 1300, finishes the job the adjective started, exactly as very did four centuries earlier. The same road, walked twice.
Etymonline, s.v. awfully (adv.)
Where it stands nowToday: bad, or, as an adverb, merely intensifying.
красный beautiful to red
The one Russian entry, because the explorer above claims three languages and drift is not an English disease. Proto-Slavic *krasьnъ meant beautiful, from *krasa "beauty", and it still does in the other Slavic languages. Only Russian pulled it onto the color, probably along the same association that gives English a healthy color: the beautiful things (ripe berries, dyed cloth, a face with blood in it) were red.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
11c. beautiful, fine, splendid obsolete
Old East Slavic красьнъ
Old East Slavic and Old Church Slavonic красьнъ, from Proto-Slavic *krasьnъ, from *krasa "beauty". Compare краска "paint" and красить "to color", whose earlier senses look like "blush" and "to decorate".
Vasmer, s.v. красный
-
12c. to 15c. finest, of the highest quality restricted
Красный угол, the icon corner of a peasant house, is the "fine corner", not the red one. Красна девица in folk formula is a beautiful girl. Both survive with the old sense intact.
Vasmer, s.v. красный
-
1515 red sense used today
First attestation of the color sense, in Muscovite diplomatic correspondence, and unique to Russian among the Slavic descendants of *krasьnъ. The older color words were чермный and червлёный. This date comes from Wiktionary, the site's unreviewed tier, and is the one date on this rail I would most want an editor to check against a print dictionary.
Unreviewed source. Flagged rather than quietly used.
Wiktionary, s.v. красный (unreviewed tier)
-
17c. Red Square acquires its name restricted
The square in Moscow acquires the name in the seventeenth century, and what it meant at the time is genuinely disputed: "beautiful square" is the usual reading, but by the 1600s the color sense was already 150 years old. The uncertainty is the point. A rail of first attestations cannot tell you which sense a given speaker had in mind.
Contested. Included because it is contested.
Vasmer, s.v. красный; the dating of the name is standard in Moscow topographic histories
-
19c. red is now the ordinary sense current
Beauty has passed to красивый (from the same root) and прекрасный. Plotted as a period, not a date.
Undated. Placed by editorial judgement.
Vasmer, s.v. красный
-
1918 communist, Bolshevik current
The Workers' and Peasants' Red Army, Рабоче-крестьянская Красная армия, is founded on 28 January 1918, and the color sense acquires a political one on top. A thousand years from beauty to a flag.
Standard reference; the RKKA is founded 28 January 1918
Where it stands nowToday: red, and after 1917 also communist. Beauty went to красивый and прекрасный.
girl any child, then half of them
For two hundred years girl meant a child of either sex, and if you needed to be specific you said knave girl for a boy and gay girl for a girl. Then the word quietly gave up half its range. Nothing replaced the old general sense; English simply stopped having a plain word for a young person, which is why it now reaches for child, kid, or the borrowed youth.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
c. 1300 a child, of either sex obsolete
form gyrle
The first English sense is simply child. The origin before that is genuinely unknown: Harper reports the guesses (an unrecorded Old English diminutive, a Low German cognate) and endorses none of them, which is unusual for him and worth noticing.
Etymonline, s.v. girl (n.)
-
late 14c. young people of both sexes obsolete
The Summoner, in the General Prologue, had the yonge girles of the diocise in his power (line 664). The line is about the young people of the diocese, not about young women, and it is the standard demonstration that the word had not narrowed yet.
Dated to the composition of the Canterbury Tales rather than to a manuscript.
Chaucer, Canterbury Tales, General Prologue, line 664
-
late 14c. a female child sense used today
The narrowed sense appears in the same period as the broad one, and for a while the two live side by side. Harper notes that the specialisation to the female was complete by the end of the century in most texts.
Etymonline, s.v. girl (n.)
-
c. 1500 a maidservant, a female servant obsolete
A specialisation of the specialisation. This is the sense behind a girl who does for us, and it is the one that makes the word's later use for adult women feel, to many ears, like a demotion carried over from the servants' hall.
Etymonline, s.v. girl (n.)
-
1640s a sweetheart, a young woman being courted current
By the middle of the seventeenth century the word covers a grown woman as well as a child, which is where the twentieth century's arguments about it start.
Etymonline, s.v. girl (n.)
Where it stands nowToday: a female child, and, contentiously, an adult woman. The old sense is entirely gone, and modern readers regularly misread Chaucer's girles as women.
sad full, then steady, then sorry
Sad begins where satisfy and satiate begin, at a Proto-Indo-European root meaning enough. Old English sæd means having had one's fill. From there the word takes the long way down: a person who has had enough is weary of the thing, and a person who is weary is heavy, and a person who is heavy is grave, and a person who is grave is sorrowful. Each step is short. The distance is only visible at the end.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
Old English sated, full, having had one's fill obsolete
Old English sæd
Old English sæd, from Proto-Germanic *sadaz, from a Proto-Indo-European root *seh₂- "to satisfy", the same root behind Latin satis "enough" and so behind satisfy, satiate and asset.
Plotted at a conventional point inside the Old English period; no first-citation claim is made.
Etymonline, s.v. sad (adj.)
-
c. 1300 weary or tired of; also firm, steadfast obsolete
The hinge, and it splits two ways at once. Having had enough of something becomes being tired of it; and being full becomes being solid, so a sad cake is a heavy one and a sad man is a reliable one. The second branch is why sixteenth century prose praises people for being sad.
Etymonline, s.v. sad (adj.)
-
mid-14c. grave, serious, dignified obsolete
Steadiness read as gravity. A sad counsellor is a serious one, and the word is still a compliment.
Etymonline, s.v. sad (adj.)
-
late 14c. sorrowful, mournful sense used today
The modern sense, arriving while all the others are still in use. Harper marks it as the sense that eventually drove the rest out, which took about three hundred years.
Etymonline, s.v. sad (adj.)
-
1899 inferior, contemptible, pathetic current
The last step down, and the only one that is about the speaker rather than the subject: to call a thing sad in this sense is to decline to take it seriously. Harper dates it to 1899; the slang of the last thirty years has made it the most productive sense the word has.
Etymonline, s.v. sad (adj.)
Where it stands nowToday: sorrowful, and, since the twentieth century, pathetic. Latin satis is a cousin, so sad and satisfied are the same word arriving from opposite directions.
thing a parliament, then anything at all
The widest word in English started as one of the narrowest. Old English þing is a meeting, specifically a public assembly with judicial power; the Icelandic parliament has been the Alþingi, the all-thing, since 930, and Norway's Storting still carries the word. From assembly it moves to the matter brought before the assembly, then to any matter, then to any entity whatever. A word that named a court now names the universe.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
Old English a meeting, an assembly, a council with judicial power obsolete
Old English þing
Old English þing, and the same word across Germanic: Old Norse þing, Old High German ding. Harper reports the older sense as "appointed time", which is what an assembly is before it is a place.
Plotted at a conventional point inside the Old English period; no first-citation claim is made.
Etymonline, s.v. thing (n.)
-
930 the general assembly of Iceland current
Old Norse Alþingi
Founded at Þingvellir, the assembly plains, and still the name of the Icelandic parliament. It is the clearest surviving evidence of what the word used to mean, and it is not evidence from a dictionary but from a building.
The founding date is the conventional one and comes from the Icelandic historical record, not from a lexicographer.
Standard reference; the Alþingi is conventionally dated to 930
-
Old English the matter, the affair, the business before the assembly restricted
The first broadening, and the one that does all the work: from the meeting to what the meeting is about. Latin did the same thing in reverse with res, which means both a legal case and a thing.
Etymonline, s.v. thing (n.)
-
c. 1200 any entity, being or object sense used today
The sense that swallowed the rest, and by now it will take almost any referent at all: a physical object, an event, an abstraction, a person one is being coy about.
Etymonline, s.v. thing (n.)
-
1590s hustings: the house-assembly, then the platform, then the campaign restricted
From Old Norse hús-þing, a lord's household council, by way of the platform the London court sat on. When a candidate is on the hustings today, the word is doing its original job under three layers of paint.
Etymonline, s.v. hustings (n.)
Where it stands nowToday: anything at all, and the emptiest noun in the language. The old sense survives in hustings and in the Scandinavian parliaments.
literally to the letter, or not
The complaint about literally is that it now means its own opposite. The history is less dramatic and more interesting: the word bleached exactly the way very and really and truly bleached before it, and it did so under the pens of the writers the complainers most admire. A word whose job is to certify that no exaggeration is present is a natural target for exaggeration; that is not a modern decline, it is what intensifiers are for.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
1530s in a literal sense, word for word sense used today
From literal, from Latin litteralis "of letters", from littera "a letter of the alphabet". The original job is to distinguish the plain reading of a text from the figurative one, which is a question that mattered a great deal to sixteenth century readers of scripture.
Etymonline, s.v. literally (adv.)
-
1680s used for emphasis, of things that are not literal current
Harper puts the figurative use in the 1680s, within 150 years of the word's arrival. It is not an error creeping in at the edges; it is the standard fate of any word that asserts its own truthfulness.
Etymonline, s.v. literally (adv.)
-
1876 the intensive, in print, from a canonical author current
Tom was literally rolling in wealth. He was not. Twain is describing a boy with a pile of traded marbles and bits of glass, and the joke depends on the word already being usable as pure emphasis.
Mark Twain, The Adventures of Tom Sawyer (1876), chapter 2
-
1925 the intensive, again, from a stylist current
He literally glowed. Gatsby does not emit light. Fitzgerald was a careful writer, and the sentence is not a slip; the word is doing the work that positively or fairly would do.
F. Scott Fitzgerald, The Great Gatsby (1925), chapter 5
-
20c. the sense becomes a shibboleth current
The objection is younger than the usage it objects to. What makes it unusual among usage complaints is that the objectors are right about the mechanism: the word really is losing its content. They are only wrong that this is new, or that it is a fault.
Undated. Placed by editorial judgement; the complaint has no first citation, it has a slow accumulation.
Etymonline, s.v. literally (adv.), which records the contested status
Where it stands nowToday: both senses at once. The OED and Merriam-Webster both record the intensive use, note that it is contested, and decline to condemn it.
azafata basket to cabin crew
One of the two non-English entries here, and the odd one out, because its history is not a slide but a resurrection. Arabic as-safaṭ, a basket, gives Spanish azafate, a shallow tray. The woman who carried the queen's tray was the azafata, an office at the Spanish court. The court went, the office went, the word went. Then in the middle of the twentieth century Spain needed a word for the woman who carries things down the aisle of an aeroplane, and someone reached into the archive and pulled out a dead one.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
Andalusi Arabic a basket, a woven container obsolete
Arabic as-safaṭ
Arabic safaṭ, with the article al- fused on in the Spanish, the same way it fused onto azúcar and aceite. Plotted in the middle of the Andalusi period for convenience; no attestation claim is made.
Corominas, s.v. azafate; DRAE, s.v. azafate
-
15c. a flat woven basket or tray restricted
form azafate
The object, in Castilian, before anyone is named after it: a low basket for carrying small things, of the kind a servant hands to somebody who outranks them.
Corominas, s.v. azafate
-
17c. the lady in waiting who carried the queen's tray obsolete
A defined office of the Spanish royal household: the azafata handed the queen her clothes and jewels on the azafate. Covarrubias describes the post in his dictionary of 1611, which is the usual citation for the sense.
Covarrubias, Tesoro de la lengua castellana (1611), s.v. azafate; DRAE, s.v. azafata
-
19c. the office and the word both lapse obsolete
With the household reformed and the object out of use, the word survives only in dictionaries and in descriptions of the old court. Plotted as a period, not a date: dictionaries record arrivals far better than departures.
Undated. Placed by editorial judgement.
DRAE, s.v. azafata
-
1940s a stewardess, cabin crew sense used today
Revived for commercial aviation, in preference to borrowing the English air hostess. It is a rare thing to watch happen: a word taken out of retirement because the new job resembled the old one closely enough, someone carrying things to seated people who are not to be kept waiting.
Dated to the decade rather than to a citation. The DRAE gives the sense without a first attestation.
DRAE, s.v. azafata
Where it stands nowToday: a flight attendant. The Real Academia now gives the aviation sense first and marks the court office as historical, so the revival has fully displaced the original.
пошлый customary to vulgar
The second Russian entry, and the sharpest pejoration in this set. Пошлый is built on пойти, to go, and originally means what has gone on for a long time: ancient, customary, established. The same root gives пошлина, the customs duty, which is still in the tax code and still means exactly that, what has customarily been paid. Then Russia acquired a class of people for whom nothing was worse than being ordinary, and within a century the word for traditional had become the word for trite.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
16c. of old standing, customary, established obsolete
form пошлый
From пойти "to go", by way of a sense "that which has gone on": the old way, the way it has always been done. In chancery Russian it is a term of approval, applied to rights and customs that are respectable because they are old.
Plotted inside the Muscovite chancery period; the sense is well attested but is not dated to a citation here.
Vasmer, s.v. пошлый
-
16c. a customs duty current
form пошлина
The same root, and the sense preserved intact: пошлина is what is customarily paid, and it is still the ordinary Russian word for a duty or a levy. A word can pejorate while its own sibling, standing in a tax office, does not.
Vasmer, s.v. пошлина
-
18c. ordinary, low, common obsolete
The turn. Old and customary becomes merely usual, merely common, and in a westernising court common is not a compliment. Vasmer notes the shift as a development within Russian rather than a foreign influence.
Plotted mid-century; the shift is gradual and not tied to a first citation.
Vasmer, s.v. пошлый
-
19c. banal, trite, vulgar, in poor taste sense used today
The modern sense, and by the 1840s it is a weapon. Gogol's characters are made of it, and the word becomes one of the standard tools of Russian literary criticism, which is unusual for a word that started as praise.
Vasmer, s.v. пошлый
-
1944 the untranslatable noun, exported current
In his book on Gogol, Nabokov transliterates пошлость as poshlust and defines it as cheap, sham, falsely important, falsely beautiful: not merely bad taste, but bad taste that is pleased with itself. The pun on lust is his own and he says so.
Vladimir Nabokov, Nikolai Gogol (1944), chapter 3
Where it stands nowToday: vulgar, banal, in bad taste, with a strong charge of contempt. The noun пошлость is famous in English mostly through Nabokov, who spelled it poshlust and spent several pages insisting it has no English equivalent.
atlas a Titan, then a book
The word this site is named for, and the one kind of broadening you can usually date to a single volume: a proper name generalises into a common noun. Greek Ἄτλας is one Titan holding up the sky. By 1636 an atlas is any bound book of maps, and English takes the sense from a title Mercator's heirs printed in 1595. Nothing on this rail is obsolete, which is what makes it broadening in the plain sense: the word only ever added.
One century per division from 1000; everything earlier is compressed into the zone left of the break. Nodes mark first attestations, not the years senses began.
-
c. 700 BCE the Titan who holds up the sky current
Ancient Greek Ἄτλας
Hesiod puts him at the western edge of the world: "Atlas through hard constraint upholds the wide heaven with unwearying head and arms, standing at the borders of the earth before the clear-voiced Hesperides". Note what he holds. It is the sky, not the earth and not a globe, and the globe on his shoulders is a later picture that the map trade helped spread. The name is traditionally read as a- plus the root of τλῆναι "to bear, to endure", so "the bearer"; Beekes takes it as Pre-Greek instead and compares Berber adrar "mountain".
Hesiod is roughly 700 BCE and the plotted year says so. The axis floor was 100 BCE until 2026-08-17, so this node and the Herodotus one below it were both parked at the wall of the compressed zone with a note saying the position was not a date; the floor is 800 BCE now and both sit at their real years.
Hesiod, Theogony 517-520 (Evelyn-White translation); Etymonline, s.v. Atlas, for the two etymologies
-
5c. BCE the mountain range in northwest Africa current
Ancient Greek Ἄτλας
Herodotus already has the name off the person and onto a place: a mountain "narrow and circular on all sides", so lofty its summits are never out of cloud, which the people living under it call the pillar of heaven and after which they are called Atlantes. The ocean past it takes the adjective Ἀτλαντικός, "of Atlas", which is where the Atlantic gets its name. So the first thing this proper noun generalised to was a landscape, two thousand years before it was a book.
Herodotus is conventionally placed in the mid 5th century BCE; 430 is the plotted year. Until 2026-08-17 this node and the Hesiod one were drawn at the same place despite being nearly three centuries apart, because the axis floor was younger than both.
Herodotus, Histories 4.184; Etymonline, s.v. Atlantic
-
1580s the Titan; a man of great endurance current
The English word arrives as a name and nothing else. A person carrying an impossible load is an Atlas well before any book is one.
Etymonline, s.v. Atlas (1580s)
-
1595 the title of one book of maps restricted
Latin Atlas sive Cosmographicae Meditationes
Gerard Mercator died in December 1594 and his heirs brought the work out the next year as Atlas Sive Cosmographicae Meditationes de Fabrica Mundi et Fabricati Figura, the first book of maps to carry the word. Which Atlas he meant is where the popular telling goes wrong. Mercator's dedication takes not the punished Titan but the Atlas of Diodorus Siculus: a philosopher-king of Mauretania, an astronomer, the first to publish the doctrine of the sphere, whose sky-bearing Diodorus explains away as a garbled memory of that globe. "I have set this man Atlas, so notable for his erudition, humaneness, and wisdom, as a model for my imitation." The 1595 title page follows the dedication: Atlas is seated, measuring a globe, at work rather than under a punishment. The crouching figure with the world on his shoulders belongs to Antonio Lafreri, who put it on his made-to-order map collections in the 1570s and never used the word, so the picture and the word arrived separately and were about two different Atlases. Etymonline, which supplies most of the dates on this rail, repeats the usual error and puts the Titan on Mercator's frontispiece.
A naming event, not yet a sense of the English word: at this date Atlas is the title of one book. Marked restricted because that is what it stayed.
Mercator's dedication in the 1595 Atlas, in the translation standardly quoted; Diodorus Siculus, Library of History 3.60, for Atlas the astronomer king; Etymonline, s.v. atlas, for the date and for the frontispiece claim this note disputes
-
1636 any bound collection of maps sense used today
The common noun, first in English in the title of Henry Hexham's translation of the Mercator-Hondius atlas, printed at Amsterdam. Forty years for one book's title to become the name of its genre, and the noun has outlasted every map inside it.
Etymonline, s.v. atlas (1636, first in the English translation of Mercator)
-
1734 the topmost vertebra of the neck current
The bone that carries the head, named for the Titan and not for the book: a second metaphor out of the same proper noun, running in a different direction. The date is the earliest quotation Wiktionary gives, in William Stukeley's Of the Gout, and like everything else on this rail it is a lower bound rather than a birthday.
Unreviewed source, and a late lower bound. Flagged rather than quietly used.
Wiktionary, s.v. atlas (unreviewed tier)
Where it stands nowToday: a book of maps, and by extension any bound collection of plates (an anatomical atlas, a linguistic atlas), the topmost vertebra of the neck, and in mathematics a family of coordinate charts that covers a manifold, which is the map-book sense taken literally and is undated here because I could not source a first use.
Reading the rail
- First attestation of a sense
- Ancestor form, not an English attestation
- First attestation of the sense used today
- Still ordinary, so a rule runs to the present
- Survives only in fixed phrases
- No rule: obsolete, and no departure date is claimed
How to read the dates
- The dates
- Every date here is a first attestation: the earliest surviving written use of a word in that sense. A sense does not begin the year somebody happens to write it down, and for most of these words the real shift is decades or centuries earlier than the citation and gradual besides. Read the rail as a lower bound on each sense, never as a birthday.
- Still in use
- Dictionaries date arrivals far better than they date departures, so this file records when a sense is first attested and whether it is still in use, and never guesses when a sense died. A sense is marked current if it is ordinary today, restricted if it survives only in fixed phrases and compounds, obsolete if it is gone from living use.
- Placement
- The rail needs a number, so each date string is given an approximate year for plotting only: 'late 13c.' plots at 1285, 'c. 1300' at 1300, 'Old English' at a conventional point inside the attested Old English period. The date string beside every node is the source's own wording; the plotted year is mine. The axis has a floor at 800 BCE, lowered from 100 BCE on 2026-08-17 so that the two Greek nodes on the atlas rail could sit at their real years instead of at the wall of the compressed zone. A sense dated below the floor now fails the build rather than being drawn off the edge of the figure.
Sources
- Douglas Harper, Online Etymology Dictionary (etymonline.com), consulted 2026-07 and 2026-08. Harper compiles his dates chiefly from the OED, Barnhart, Klein and Watkins; where a date below is quoted as 'late 13c.' or '1570s' that is Harper's wording, and behind it usually sits an OED first citation.
- Max Vasmer, Etymological Dictionary of the Russian Language (Russian edition, Moscow 1964–1973), for the Slavic material.
- Joan Corominas, Diccionario crítico etimológico, and the Real Academia's Diccionario de la lengua española, for the Spanish material.
- Wiktionary, s.v. красный, for the 1515 first attestation of the color sense, and s.v. atlas, for the earliest quotation of the vertebra sense. Wiktionary is the site's unreviewed tier and is labelled as such wherever it is used.
- Direct citations (Chaucer, Shakespeare, the King James Bible, the Nicene Creed, Twain, Fitzgerald, Nabokov, and for atlas: Hesiod, Herodotus, Diodorus Siculus and Mercator's own dedication) are given with book, chapter or line so a reader can check them without a dictionary.
So the rail is a report and not a model, and its nodes have to be read narrowly. Each one is a first attestation: the earliest surviving written use of the word in that sense, which is a fact about the survival of documents at least as much as a fact about the language. A sense does not begin the year somebody happens to write it down, and for most of these words the real shift is decades earlier and gradual besides, so every node is a lower bound. The figure also cannot tell you when a sense died, and it does not pretend to: obsolete senses get a node and no rule. It cannot show frequency either, so a sense with one recorded citation and a sense every speaker used look identical. And the discreteness is itself an editorial artifact. A lexicographer decides where to cut a continuum of use into numbered senses, and a different lexicographer cuts differently; the number of nodes on a rail is a decision about presentation, not a census. Harper’s own note on nice says the quiet part out loud: in many sixteenth and seventeenth century examples it is impossible to say which sense the writer meant.
What the rail can show is shape. It shows direction, and that direction is often monotone: silly goes downhill at every step from “blessed” to “foolish” without once turning back, while knight goes uphill over the same centuries and villain, the same social arrangement seen from below, goes down. It shows coexistence, which the prose of a dictionary entry hides: the narrow sense of deer begins inside Old English while the wide sense is at its widest in early Middle English, so drift is not a queue and a word does not finish one sense before starting the next. It shows where dead senses are kept, which is almost always compounds and scripture (sweetmeat, and the herbs given “for meat” in Genesis 1:29). Most of all it shows the controls, and the controls are the argument. German kept selig “blessed” and Knecht “servant” exactly where English found them, and the other Slavic descendants of *krasьnъ still mean “beautiful” while Russian красный means “red”. Nothing inside those words obliged them to move. That is what it means to say semantic change is historical rather than mechanical, and it is why the right figure here is a row of dated citations with their sources attached rather than a curve.
Three risks
Three risks shaped the design. The first is epistemic: etymology is settled at the center and contested at the margins, and a graph’s crisp edges can launder uncertainty into false confidence. My rule for v1 is to curate only textbook-standard families and to make every edge carry its source; boring etymologies, well cited, beat spectacular ones. The second risk was the extraction fantasy: everyone who looks at Wiktionary’s etymology templates thinks the pipeline is a weekend project, and everyone who has parsed MediaWiki markup knows better. The resolution is instructive: the pipeline did ship fast, but only because wiktextract had already spent years doing the MediaWiki parsing; my extractor consumes its structured output. The hard part turned out to be exactly where predicted, just relocated: not parsing, but trust; hence the two-tier design, where extraction feeds a review queue instead of silently feeding the site. The third risk is visual: etymology graphs degenerate into hairballs fast, and the answer built into the design is that the explorer never draws the whole graph. It draws one family at a time, laid out in generations, which is a bet that the interesting unit of etymology is the family and not the network. The extractor has since tested that bet with families it assembled rather than families I chose, and they went into the same layout without it changing, which is the visual half of the same sentence the pipeline section makes about the schema. If a family ever does defeat the layout, that finding goes here, because having somewhere to report the answer is what writing the risk down was for.