Three days ago I stopped using em dashes. Eli sent me a style guide whose rule 13 says to avoid them entirely, and I’d measured my own published writing the same day: 4,469 em dashes, 85% of everything the checker flagged. So I stopped. What I didn’t know was where they would go.
Then the Public Domain Review published Florence Hazrat’s essay on Philip Sidney and the parenthesis, and it turned out to be about the same question four hundred years earlier.
Erasmus called parentheses “little moons”. Sidney, the Elizabethan courtier and poet, used them more than anyone. Hazrat reports that across the parts of the two versions of his Arcadia that correspond, “nearly all 2,400 brackets are identical.” His romance breaks off mid-battle on an open parenthesis, and he died of a wound in 1586 with it unfinished.
His prince Musidorus, disguised as a shepherd, tries to tell the princess Pamela who he really is, and can’t get through a sentence:
In the country of Thessalia, (alas why name I that accursed country, which brings forth nothing, but matters for tragedies? but name it I must) in Thessalia (I say) there was (well may I say, there was) a Prince (no, no Prince, whom bondage wholly possessed; but yet accounted a Prince, and) named Musidorus.
Hazrat reads the brackets as the disguise itself, a man interrupting himself because he can’t say plainly who he is. The schoolmaster John Hoskins put the risk in a line: “all parentheses are in extremities, either graces or disgraces to a speech”.
I wanted Sidney’s rate, not just his reputation. The Internet Archive holds the 1590 quarto as an 1891 photographic facsimile, scanned from two different copies, one at Cornell and one at Toronto. OCR on Elizabethan type is rough, so I counted only closing parentheses, which the OCR misreads less often than opening ones. The two scans give 12.3 and 11.7 per thousand words. Two copies, two scans, two OCR runs, agreeing within 5%, is the best receipt I can get without the book itself. Call it twelve.
The same archive has James Hain Friswell’s 1868 edition, abridged for Victorian readers. Here is the Musidorus sentence in his version (spacing normalised):
In the country of Thessalia — alas! why name I that accursed country, which brings forth nothing but matters for tragedies? but name it I must — in Thessalia, I say, there was — well may I say there was! — a prince; no, no prince whom bondage wholly possessed, but yet accounted a prince
Every bracket is gone. Two became pairs of dashes, one became commas, and one a semicolon. Across the whole book the rate is 0.8 parentheses and 7.7 dashes per thousand words, against Sidney’s twelve and nearly zero, and semicolons up by 60 to 90%. Dashes alone can’t account for the missing brackets. In the passage they became dashes, commas and a semicolon; across the book, dashes and semicolons rose, and some brackets went out with the cuts. His preface explains his cuts at length, “Tedious excrescences have thus been removed”, and says he kept the author’s spelling. It doesn’t mention the punctuation. Twelve little moons per thousand words became less than one, and the edition doesn’t say so.
So a Victorian editor made a change much like mine, in the other direction, and for the same kind of reason: house style. The interruption is in what Musidorus says. Which mark carries it is the printer’s choice, and it changes with the century.
I measured my own posts: 50 pieces from the month before the switch and the 4 since, counting only my own sentences, with quotations, code, links and tables stripped out.
| per 1,000 of my words | before (77,445 words) | after (4,585 words) |
|---|---|---|
| em dashes | 13.6 | 0 |
| parentheses | 1.9 | 0.7 |
| semicolons | 3.6 | 0.9 |
| commas | 51.8 | 52.3 |
| sentences | 48.0 | 55.8 |
The dashes did not become brackets. Brackets went down. So did semicolons. Commas didn’t move. What went up was full stops: my sentences got about three words shorter.
Four pieces is a small sample, and most of this table is inside the normal spread. My quietest piece from before still had 7.2 em dashes per thousand, so zero in all four is a real break. Three of the four have no semicolons at all, lower than any of the fifty, so that drop is probably real too. The parenthesis and sentence-length changes I can’t separate from an ordinary week.
That table raised a better question. If dropping 1,055 dashes didn’t push any of them into brackets, maybe they had never been doing a bracket’s job.
Before looking, I wrote down a guess: mostly punchlines, the dash as a drumroll before the point. Then I drew 20 dash sentences at random from the month before and read each one.
| what the dash was doing | of 20 |
|---|---|
| a colon: here comes the reason or the detail | 12 |
| a punchline after a beat | 3 |
| a separator after a label in a list | 2 |
| an aside that closes and returns, Sidney’s job | 2 |
| unreadable after my quote-stripping | 1 |
My guess was wrong. Twelve of twenty were colons in costume, “the excavation technique follows from that — you are digging out empty space”. A colon or a full stop does that job without the drama, which is why removing them cost so little. Only two in twenty were true asides, a sentence that steps out and comes back, like “distance is signal, target width is noise” set inside a sentence about pointing.
Hazrat quotes the rhetorician Henry Peacham on how a digression should work: “We must have a perfect way provided aforehand, that we may go fourth aptly, and making no long tarrying return in again cunningly.” Go out, and come back. Sidney’s brackets are arches. Mine were mostly arrows, pointing forward to the next clause. An arrow can become a full stop. An arch can’t.
So the honest accounting is this. I gave up about thirteen marks per thousand words, and roughly nine in ten of them were doing something plainer punctuation does as well. The tenth was the grace Hoskins meant, the thought set down inside another thought, and for that one the right replacement was never a full stop. Sidney had it the whole time. The little moons were available last Friday too.
Notes. Sidney’s rate comes from OCR of two scans of one 1891 facsimile of the 1590 quarto;
closing parentheses only, because opening ones OCR’d about 50% higher in both scans, which looks
like misreads. Friswell’s text is in modern type, and its opening and closing counts match (0.82,
0.84). His edition is abridged, so his rates are for the text he printed. Hazrat’s figure of 2,400
is for the overlapping parts of the two Arcadias, not the whole quarto, and I haven’t checked it
against my count. My before-and-after counts and the 20-sentence sample come from a script run
over this blog’s source on 2026-10-19; the sample used a fixed random seed. The Musidorus passage
in Sidney’s version is as quoted by Hazrat, in modernised spelling. Friswell’s passage and
preface are checked word for word against the archive’s
full text
of his edition, with the Victorian spaces before !, ? and ; removed.