Feeding America · 74 cookbooks · 1803–1922
A Piece of Butter the Size of an Egg
Every recipe is a program with holes in it, and the holes are where the writer assumed you already knew. This page reads 48,644 recipes out of 74 American cookbooks, hand transcribed from the originals, and scores each of them on a single question: when it tells you how much, how much does it actually tell you?
There are four ways to say an amount, and they form a ladder. At the top, a unit whose size is fixed somewhere outside your kitchen: a pound, a pint, a teaspoon. Below that, a unit that is whatever object you happen to own: a teacupful, a wineglassful, a bare spoonful. Below that, a comparison to a thing: butter the size of an egg, a lump as large as a walnut. And at the bottom, the shrug: a little, a few, enough, to taste.
Read the ladder backwards and it is a scale of assumption. The top rung assumes nothing about you. The bottom rung assumes you have cooked this before.
Read one
Below is a real recipe, printed as its book printed it. The amounts are marked live in your browser by the same code that scored all 48,644, so you can watch the classifier work and watch it miss. Drag the year. Nothing here is a picture of an analysis; it is the analysis.
The classifier is wrong sometimes and you can catch it doing it. On the eel recipe it opens with, from 1803, the words three pieces of are marked as an amount and are not one: they are the three pieces the eel was cut into four sentences earlier. That is the shape of its errors, and the section at the foot of this page measures how many there are.
The century, one book at a time
Each dot is one book: its year across, and across the four classes, the share of that book's amount expressions falling in each. Seventy-four books, 169,993 expressions. The line is the decade mean taken over books, not over recipes, so a single encyclopedic volume cannot stand in for its decade.
The named share climbs and the other three fall, but not steeply and not smoothly: Spearman's rank correlation with year over the 74 books is for the named share and for the vessel share. What moves harder is the particular vocabulary. Count only the three units of the modern kitchen set, cup and teaspoon and tablespoon, against the household vessels they replaced, and the crossing is sharp.
What Fannie Farmer actually did
The Boston Cooking-School Cook Book of 1896 is the book everybody names, and its author is the one figure in this history with a nickname. Two versions of the claim are in circulation and they are not the same claim. Britannica gives the careful one:
A state humanities council gives the strong one:
The strong version is false, and this corpus is one place you can watch it be false. The cup and the spoon were already there. of runs at cup-and-spoon mentions per thousand words, ahead of most books published after Farmer. Her own predecessor at the same cooking school, Mary Lincoln, was printing itemised ingredient lists and a full table of cup-and-spoon equivalents in 1884, and telling readers where to buy the equipment:
Twelve years later Farmer describes the same object in almost the same words, down to the quarters and the thirds and the half pint, and adds that it may be bought at any store where kitchen furnishings are sold. She did not introduce the instrument. Laura Shapiro, whose chapter on Farmer is titled with the nickname, dates the arrival of the measuring cup to the mid 1880s.
What she did is visible in a different number. Of the 74 books, the one with the smallest share of shrugs is hers, and her second book is third.
Share of amount expressions that are indefinite
| Rank | Year | Book | Indefinite | Named |
|---|
The lowest six of 74. Rank 1 and rank 3 are the same author, eight years apart.
And here is the thing the word counts give away. Search Farmer's 1,424 recipes for the word that made her famous and you will find it time. She says it at the front instead, three times in three consecutive sentences, and never has to say it again:
Lincoln in 1884 had the opposite default. For her the unmarked spoonful is a heaped one, and the level spoonful is the marked case that needs a word of its own:
She says it again, flatly, a page later:
The behaviour of the two books matches their rules exactly. Counting the words heaping, heaped, rounded and rounding inside the recipes themselves: Lincoln uses one of them times in 809 recipes, Farmer times in 1,424, a rate fourteen times lower. Each uses the word level exactly . Neither is being sloppy. They are two conventions, twelve years and one city apart, and between a heaped teaspoon and a levelled one lies most of the soda in every cake in the book.
Not everybody followed her into the front matter. Mrs. Rorer, writing six years later, says level times inside her own recipes, once per spoon, over and over, as though the word were still doing work. It was. The convention was not yet a convention.
That is the real invention, and it is not the cup. It is the idea that a unit is not a unit until the way you fill it is part of the definition. Before that, a teaspoon was an object. After it, a teaspoon was a quantity.
The tables that disagree
Most of these books print a short table telling you what their own words mean. It is the only place in the corpus where a book states a unit instead of using one, which makes it the direct test of whether the standard was standard. Pulling every such statement out of the non-recipe text gives equivalences from books. Here is what they say.
| Year | Book | Value | As printed |
|---|
Three teaspoons make a tablespoon. Every American cook knows this; it is stamped on the back of the measuring spoons. In this corpus books say three and say four, and one book is on both lists. Mary Lincoln, 1884: four teaspoonfuls of liquid make a tablespoonful, and three teaspoonfuls of dry material make a tablespoonful. Neither line is a mistake. The ratio depends on what is in the spoon, because for her a dry spoon is heaped and a spoon of liquid cannot be.
Which is the whole point. The conversion table is not possible until the filling rule exists. Farmer's front matter does not just tidy the kitchen; it is the precondition for the arithmetic. After levelling, three is three, for everything, always.
The disagreements do not stop at spoons. Cups of sugar to the pound run . Saltspoons to the teaspoon are four in Lincoln and in the 1919 International Jewish Cook Book, and eight in the 1901 Settlement Cook Book. Two of the books print density tables that agree with each other to the last ounce across six commodities, eight years apart, which is not independent measurement; it is copying.
One table in the corpus is simply wrong. Sarah Hale's 1839 list of liquid measures runs down correctly through half pints and gills and then says that two quarts make a gallon. It is not a transcription slip. The page itself says it.
What survived all this is Lincoln's arithmetic, not Farmer's fame. Three teaspoons to the tablespoon and sixteen tablespoons to the cup, the ratios she printed in 1884, are still exactly the American standard: NIST Handbook 44 gives the tablespoon as three teaspoons exactly and the cup as eight fluid ounces exactly, half a liquid pint. The numbers did not change. The instruction to level it off is what arrived.
The bestiary
Before the cup won, an amount could be any object in the room. Every comparison in the corpus, of them across distinct objects, extracted by one mechanical rule and sorted by how often American cooks reached for each. The egg is the ruler of this world; the walnut is its second. Further down are the coins, and the coins are a clock: a cook measuring butter against a shilling is being read in a country that stopped minting them.
| Uses | Object | Years | First appearance, verbatim |
|---|
The oven that was never a number
One measurement in these books never gets standardised at all. Of the recipes that mention an oven, describe its heat with an adjective, quick or moderate or slow or brisk or slack. give a temperature in degrees, and of those come from just two books. The corpus runs out in 1922 with the oven still un-numbered.
It is not that nobody had thought of it. In 1868, twenty-eight years before Farmer, a French chef writing for American kitchens tells his readers exactly what to do, and what he tells them to do is take a drill to the stove:
Nobody did. The measured kitchen arrived unit by unit, and the last unit to arrive was heat.
The check
What the corpus is. Feeding America: The Historic American Cookbook Dataset, Michigan State University Libraries: 76 catalogued cookbooks, transcribed by hand from the originals rather than by OCR, and encoded with explicit recipe boundaries.
What the classifier is. A unit word alone is not an amount, so an expression is counted only if a real numeral governs it, or it carries the -ful suffix, or the word of follows it, or the unit is one you cannot put anything into (a pound, a gill, a teaspoon). A cup, a glass and a bowl are objects as well as measures and stay under the stricter rule, which is why put it in a cup is not marked above. The whole rule is research/recipe-measurement/measures.mjs, and this page imports that same file rather than a copy of its logic.
What it costs. Four independent readers annotated recipes drawn at random, seeing only the text and the class definitions, never the code or its output.
Where the dates come from.
What is refused. Elliptical amounts are not counted: in two spoonfuls of wine, two of water, and two of vinegar only the first carries a unit word, so the count is one, and the corpus is therefore an undercount of how often amounts are given at all. Bare counts (three eggs) are not amounts in any class. Two of the 76 files contain no marked recipes and are dropped; they are named above. Not all 74 of the rest are general household cookbooks: MSU's own type attribute marks 51 general, 10 famous, 6 charity, 4 frugal, 2 encyclopedic and 1 restaurant, and six books carry fewer than 60 marked recipes. A grocer's encyclopedia, a market guide, a bartender's manual and an ethnography of Zuni bread are all in here, and they sit in the scatter as visible outliers rather than being quietly removed.
What is not settled. One row of Sarah Hale's 1839 table of measures reads, in the transcription, Two quarts are one gallon, which is wrong by a factor of two. The somatic class is the least reliable of the four; its agreement with the readers is the lowest, and its early-century level should be read as approximate. The equivalence extractor reads the leftmost unit on each side of an equals sign, so a row offering two equivalents at once (1 wineglass, or 1/2 gill, or 1/4 cup) contributes one of them, not both.
Reproduce it. node research/recipe-measurement/fetch.mjs downloads the dataset, then build-books.mjs, analyse.mjs, build-equivalences.mjs and build-page-data.mjs rebuild every number on this page. The offline verifier is verify-the-size-of-an-egg.mjs at the repository root; it re-derives each figure from the committed artifacts and compares it against the string this page prints, so a number cannot drift out of the prose without the check going red.