A dispute about a word · counted

What the First Sleep Was First Of

For twenty-five years the best-known claim in the history of sleep has been that pre-industrial Europeans slept twice, waking for an hour in the dark between a first sleep and a second. In 2023 a historian answered that the phrase does not by itself establish the pattern, and listed readings it is equally compatible with. Both sides argue from quotations, and no count on either side has ever been set against a denominator. This page does that, in hand-keyed books printed between and , and hands you every occurrence.

Sixty-three fragments

The claim is A. Roger Ekirch's, made in the American Historical Review in 2001 and expanded in At Day's Close in 2005. Until the modern era, he argued, an interval of quiet wakefulness split the night in two, and English had ordinary words for the halves. People rose to poke the fire, prayed, talked, made love, and above all remembered their dreams, because they were waking straight out of them.

The evidence was a collection of quotations. The 2001 article says how many, in a footnote on page 364:

discovered sixty-three references within a total of fifty-eight different sources from the period 1300–1800 Ekirch (2001), 364, as quoted in Boyce (2023), n. 48

By 2024 the collection had grown. Ekirch writes of "upwards of two thousand" allusions found to date, and more than two hundred excerpts posted on his university site. The method is unchanged: gather the shards, and read the pattern in them.

In 2023 Niall Boyce, in Medical History, made the objection that a collection of shards invites:

the various quotations that Ekirch presents in his study might support the idea of segmented sleep if read with that concept in mind, but may also suggest other possibilities Boyce (2023), "Have we lost sleep?", Medical History 67(2)

And then the sentence the whole argument turns on:

In brief, the phrase ‘first sleep’ is not of itself indicative of a habitual pattern of well-defined, separate periods of sleep, but might have other meanings depending on the context. Boyce (2023)

He lists the other meanings. It might be the opening phase of one continuous sleep. It might be the first stretch of a night that was simply broken. Or it might name a quality rather than an order, because early sleep is the heaviest. For that last reading he has a seventeenth-century witness. Joseph Caryl, expounding Job in 1656, glosses the verse about deep sleep falling on men like this:

in the former part or beginning of the night for the first sleepe is the deepe sleepe; and we use to say that a man, especially a weary hard-wrought man, is in a dead sleepe, when he is in his first sleepe. Joseph Caryl, An Exposition upon the Book of Job (London, 1656), sig. M4r, as quoted in Boyce (2023), n. 61

Read Caryl closely and he is more careful than the argument built on him. He locates the first sleep "in the former part or beginning of the night", which is a stretch, and he says a man is in a dead sleep when he is in his first sleep, which is co-occurrence. He is not quite saying the three phrases are interchangeable. What Boyce takes from him is the weaker and still substantial reading that first sleep may be doing the work of a quality word rather than an ordinal one.

That is testable, and this is the shape of the test. If first sleep is a quality word it should be distributed through English the way the other quality words are. If it names a bounded stretch of the night it should not. English also offers a stretch of night nobody disputes was numbered and bounded, the first watch, so that goes in as a positive control: an instrument that cannot find the watch has not earned the right to pronounce on the sleep.

The counting neither side did

Boyce did search a corpus. He reports it in one line, and it is the only published count on either side of this argument:

Searching EEBO for the terms ‘second sleep/sleepe/slepe’ in full text with the same parameters used above retrieves fifteen hits. Of these, only five contain explicit references to both a ‘first’ and a ‘second’ sleep. Boyce (2023)

For first sleep he gives no number at all, only "yields many such examples". And his footnote 64 says exactly what a count of this kind is worth, and what it needs:

Nevertheless, assuming that errors and omissions occur more or less at random (in that they are no more likely to affect the frequency of detection of one term than another), they can give a very rough indication of how common particular terms may have been in early modern print. Boyce (2023), n. 64

That is the design of this page, taken further than "very rough". Errors that fall equally on every term cannot fake a difference between terms, so the whole argument below is built out of comparisons: first sleep against dead, sound and deep sleep, and against the first watch, all scored by one mechanical rule that has never been told which is which.

What was counted

The Text Creation Partnership transcribed early printed books by hand, page by page, from the microfilm: Early English Books Online for 1473 to 1700, Eighteenth Century Collections Online after it, and Evans for early America. Every file carries the year the book was printed. All of it is public domain and all of it is on GitHub, which is where these numbers come from: texts, words, fetched, converted to running text, counted, and discarded.

Hand-keyed matters here. This is not optical character recognition guessing at black letter. Somebody read the page and typed what was on it, and where they could not read it they marked a gap rather than inventing a word. So where the phrase is missing, it is missing from what a person read on the page. That is not the same as missing from the printed record: this is a transcribed subset of what survives, gaps are marked rather than guessed at, and a word the printer broke across a page turn can still be missed.

The number nobody had

Across the whole corpus, first sleep and its spellings occur times, in of the books. That is per million words. Sleep in any form is mentioned times, so first sleep accounts for of every thousand mentions of sleep.

It is not, however, a common thing to say. Of the distinct words this corpus puts immediately in front of a form of sleep, first ranks . Dead, deep, sound and sweet are all several times more frequent. So if the claim is that the vocabulary was ordinary in the sense of being everywhere, this corpus does not support it. If the claim is that it was ordinary in the sense of being used without explanation when it was used at all, that is a different claim, and the passages below are where to test it.

It is worth being careful about what that adds. Ekirch's published figure in 2001 was sixty-three references; by 2024 he speaks of upwards of two thousand allusions gathered by hand across six centuries, several languages and every kind of source. This corpus is not a larger pile of quotations than his. What it supplies is the thing a pile cannot have: a denominator. Not how many times the phrase was found by somebody looking for it, but how often it occurs against how much was printed, and against everything else people said about sleep.

Boyce's fifteen, reproduced

Boyce searched EEBO with no date restriction, "thus covering the years 1475–1700", and got fifteen hits for the second sleep. Restricting this corpus to the same window gives occurrences in texts, and texts that name both a first and a second sleep, against his five. The window holds texts and words here.

These two searches are not the same search and are not required to agree. Boyce queried EEBO's index of a larger set of books; this queries the hand-keyed subset of them, matching a fixed list of spellings. What the comparison establishes is the order of magnitude, and that both find the same striking thing: whatever the first sleep was, the printed record of the second one is tiny. Ekirch concedes the asymmetry himself, and names the alternatives, which are counted here too:

The advice to ‘lye to sleep again’ was one of various expressions occasionally employed in place of the term ‘second sleep’. Others included ‘morning’, ‘latter’, or ‘last’ sleep. Ekirch (2024), "Reflections", Medical History 68(3)

Of that list, morning, latter and last sleep are searched for here. "Lye to sleep again" is not, and cannot be: no list of spellings catches a paraphrase, so every second-sleep count on this page is a floor for the practice even where it is a fair count of the phrase.

The ordinal test

If first in first sleep is doing ordinal work, pointing forward to a second, then the printed record should carry seconds in something like the proportion it carries them for a thing that really is numbered. The night watches give the yardstick. For the watches, second occurrences run at of first ones. For sleep the same ratio is , and even counting morning and latter sleep as seconds it only reaches .

Ekirch's answer to that is not that the second sleep was rarely there but that it was rarely worth writing down:

Unfortunately, these instructions do not provide detailed descriptions of segmented sleep, but why would they? Neither is it possible to find detailed discussions in early modern texts of other routine bodily functions, such as urinating, defecating, consuming food, drinking water, and breathing. Ekirch (2024), "Reflections", Medical History 68(3)

That is an argument from silence, and arguments from silence are cheap. It is also, on its own terms, hard to fault: the ratio above cannot distinguish a thing nobody did from a thing nobody bothered to mention. So the ordinal test is reported and then set aside.

The frame test

Here is the measurement the dispute actually turns on, and it is a question about grammar, not about atmosphere.

A phrase that names a completed stretch of time takes completed-stretch frames. You say after his first sleep. You say he had slept his first sleep. You say his first sleep being ended. A phrase that names a quality of sleep takes state frames instead: he lay in a dead sleep. No individual passage is judged. The word lists that do the judging are, and they are printed in the working; every class of hit is scored by the identical rule, in a window of four words before and six after.

The same test inside one book

There is an obvious objection to all of that, and it is a good one. Books that say first sleep might simply be a different kind of book from books that say dead sleep: narrative prose, where "after this, he did that" is how you move a scene along, against verse and sermon where it is not. If that were the whole story, the frame difference would be a fact about genre with nothing to do with sleep.

So here is the same test run again, restricted to the books that use both kinds of phrase. Same author, same page, same register, same century, same printer. The only thing that changes is which phrase is being used.

The bench

The mechanical rule is one reader with one crude eye. Here is the same discrimination, handed to you, and made blind: the adjective is covered up. Read the passage and say whether the hidden word was first, or one of the words meaning heavy. Chance is one in two.

Everything found

This is the part a pile of quotations cannot offer. Below is the whole set, not a selection: every occurrence of the disputed vocabulary in the corpus, oldest first, with the book it came out of. The quality adjectives and the watches are sampled at one in six, which is said again under the panel. The spelling is the printer's. If the reading offered above is wrong, the passage that shows it is wrong is in here.

The books that name both

One class of passage is harder to read away than any other: a book that names a first sleep and a second one. Boyce reports that five of his fifteen hits name both, which is a count of hits and not necessarily of separate books, so the comparison here is loose in his favour. Here is the whole list this corpus yields, assembled in your browser by intersecting the concordance above, so you can see exactly which books they are and read what they say.

The manuals of health

Boyce quotes a further objection, from Janine Rivière, and it is a sharp one because it points at the genre where an ordinary bodily habit ought to leave a trace:

although A. Roger Ekirch suggests premodern people typically experienced a pattern of segmented sleep, this is not discussed in manuals of health or discussions of the proper regimens of sleep. Janine Rivière, Dreams in Early Modern England, as quoted in Boyce (2023), n. 14

Ekirch answers that one in 2024 with Francis Bacon, who prescribes an elixir to be taken "between sleeps", so the Bacon case is already in the argument and is not offered here as news. What a mechanical sweep adds is that the search can be run over the whole genre at once instead of remembered. It finds Bacon twice, in Sylua Sylvarum (1627) and in the Historie Naturall and Experimentall, of Life and Death (1638), and it finds an English regimen from a century and a half earlier. This is the Gouernayle of Helthe, printed about 1490, instructing the reader which side to lie on:

slepe fyrst on thy syght side for that is kyndely for thy dygestio rhall be better / for then lieth thy lyuer vnder thi stomak / as fyre vnder a caudren: And after thi fyrst slepe turne on thy lifte syde that thy ryght side maye be r • sted of thy longe lygyng theron / And whan thou hast layen theron a good while and slept turne ayen on thi ryght side and ther slepe all nyght forth Gouernayle of Helthe, London, ca. 1490 · TCP A01993, as transcribed (• marks where the transcriber could not read the page)

Be careful about what that shows. It is a manual of health, and it does discuss the proper regimen of sleep, and it structures the night around a completed first sleep after which the sleeper does something. That is a real counterexample to the sentence quoted above, found by a search rather than by memory. But it prescribes turning over, not getting up, and it says nothing about an hour of wakefulness. It supports the reading that first sleep named a stretch that ended. It does not, on its own, put anybody out of bed.

After the corpus ends

The transcribed books stop in 1800, and that is a problem for the most interesting half of the claim, because Ekirch has moved the date. In 2024 he writes that the transition away from segmented sleep

occurred later than I had first speculated in ‘Sleep We Have Lost’. In light of numerous nineteenth-century sources, segmented sleep remained prevalent well into the 1800s. Ekirch (2024), "Reflections", Medical History 68(3)

Which puts the disappearance entirely outside the window Boyce's EEBO search can reach. Two other corpora cross it, and neither is as good as the first: Google's scan of printed books, which gives a frequency and never a line of context, and the Library of Congress's Chronicling America, machine-read pages of American newspapers. That collection is usually described in the tens of millions of pages. The figure used below is instead the one its own search returns, pages across the decades queried, because that is the denominator these counts were actually taken against.

What this settles, and what it does not

What it does not settle is what people did in their beds, and no corpus can. A word can be the name of a real practice, a literary convention, an archaism kept alive in print after it died in life, or all three at once in different books. A phrase is not a polysomnogram, and the furthest a count can go is to say which reading the language is friendlier to, and hand over the whole set so the reading can be argued with.

It is worth being exact about which of Boyce's alternatives the frame test touches, because it does not touch them all. He offered three. That first sleep names the opening phase of one continuous sleep, and that it names the belief that early sleep is the heaviest: those are the two the grammar argues against, and the second is the one Boyce reads out of Caryl, though Caryl's own sentence is narrower than the reading. His third reading is untouched. If first sleep meant the first stretch of a night that happened to be broken, in a person or a situation where sleep was broken, every number on this page would look exactly the same. A completed stretch of sleep with more night after it is what the language shows; that this was the ordinary shape of an ordinary night, rather than what you say when your night went badly, is a further claim the words cannot carry. Boyce's other point, that several of Ekirch's set pieces happen in prisons, on watch and in shared chambers, is likewise untouched: this page counts phrases, and never asked what room anybody was in.

Two other kinds of evidence bear on it, and they disagree with each other. In a laboratory in 1992, Thomas Wehr moved volunteers from a 16-hour photoperiod to a 10-hour one, and their sleep, in his abstract's words, "usually divided into two symmetrical bouts, several hours in duration, with a 1–3 h waking interval between them". In 2015 Gerald Yetish and colleagues put wrist actigraphs on Hadza, San and Tsimane people near the equator and found the opposite: sleep that "was not interrupted by extended periods of waking", and no second sleep their algorithm could score. Ekirch answered that the equatorial null does not settle Europe, and conceded the counterpoint was "welcome, albeit singular". So the long night in a lab produces the pattern, and the short night in the field does not, and the historical record is the third witness. This page is an attempt to make that third witness say something a stranger can check.