The Sentence the Machine Missed
By Noam Shemla · Speeches That Lasted
TL;DR We measured the speech Winston Churchill gave the House of Commons on 4 June 1940 with the same instrument our members record into. It returns 94 words a minute, which our pace band scores at 50 out of 100. That is the same floor Selma landed on, for the same reason: the tool was built for a three-minute clip, and this speech runs forty. The more interesting failure is one the pace score can’t see. The phrase this speech is remembered for reads like eight defiant sentences, but grammatically it is a single one, the longest in the address, and our repetition figure only looks for phrases that open a NEW sentence.
The Selma article published a score that was wrong and said why. The Gettysburg article published a clean sweep and said what it couldn’t see. This one publishes something in between: a correct score, on a speech whose most famous device turns out to be built in a way our own instrument cannot register at all.
What copyright lets us show
Unlike Obama’s Selma address or Lincoln’s Gettysburg text, this one is not free to quote. Churchill’s wartime speeches remain in copyright (in the UK, for the length of his life plus seventy years, so until 2035) and are administered by the Chartwell Trust through the literary agency Curtis Brown. The full text we measured is reprinted, by permission, on the International Churchill Society’s own site. That permission belongs to them and gives us no licence to reproduce it here. So this article does what every Tier B entry in this series does: structure, timings and short attributed fragments, without the speech in full.
There is a second reason this entry looks different from Selma and Gettysburg, and it has nothing to do with rights. No audio recording of this speech, as delivered, exists. Broadcasting a sitting Prime Minister’s Commons statement live was not the practice in June 1940. Churchill did later read several of his wartime speeches into a microphone for a 1949 gramophone record. It is a fine recording, and the source of the popular belief that this is what the Commons heard, but it is a different performance, given nine years later, in a recording studio, to no audience at all. Treating it as this speech’s audio would measure the wrong thing and call it the right one. So, like Gettysburg, this entry is text only, and says so instead of quietly reaching for the 1949 tape.
The number that comes out familiar
Twenty-nine filler words in 3,777, or 0.8 percent, for a clarity-from-fillers score of 94.6. That is a clean text, close to Obama’s 93.7 and nowhere near what a live, unscripted address usually produces. It reads like what it was: prepared, revised, delivered from a text Churchill had spent hours drafting.
A speech that earns its ending
The shape argues for itself. The speech does not open with defiance and coast. It opens on a long, dense sentence cataloguing exactly how the French front collapsed, and the sentences carrying the worst news (the fall of Boulogne and Calais, the Belgian surrender, the 30,000 casualties) run just as long and just as unflinching as anything at the close. The famous ending comes after 140 sentences spent earning it, and it is, by a wide margin, the longest sentence in the whole address.
The litany that isn’t there
This entry exists to publish one blind spot. Ask anyone what they remember of this speech and they will give you a list (beaches, landing grounds, fields, streets, hills) delivered as if in a drumbeat of short, separate declarations. Structurally, on the page, it is not eight sentences. It is one 141-word sentence, its clauses joined by commas, building without a full stop from “we shall go on to the end” all the way through “we shall never surrender” to the appeal to the New World at the very close. Our repetition figure is built to catch anaphora at the seams between sentences. That is exactly where King’s “I have a dream” sits, and exactly where this passage does not. A crude n-gram counter, applied faithfully, cannot find a device that the speaker built inside a single breath.
Gaps the text cannot fill
A second recording, a different claim: Churchill reading himself, nine years later
Everything above measures the address as delivered to the Commons on 4 June 1940, and that measurement will always be text-only, since no recording of it exists. A second, different artifact does exist, though. In 1949, Churchill sat in front of a microphone and read a portion of this same speech aloud for a Decca gramophone record. That recording survives, and an edited 12-minute excerpt of it is measurable. It is neither the full speech nor this delivery. Treating it as more than that would be exactly the mistake this article exists to avoid, so what follows describes it and measures it on its own terms.
The excerpt covers roughly 1,491 of the address’s 3,777 words (about two-fifths), read continuously in several passages (the "week ago today" military assessment, the Dunkirk evacuation account, the Air Force tribute, and the closing "nevertheless" through "we shall never surrender") with the passages about the Belgian king's surrender, the casualty count, and home-defence measures cut out. It is a highlights reading of the speech, and the 1940 audience never heard it, because it didn't exist yet.
- 121 wpm pace. Faster than the 94 wpm the full 1940 text implies. This is a shorter, edited passage read continuously, where 1940 was a 40-minute address to a nervous House.
- 158 Hz mean pitch. ±66 Hz. Notably lower and calmer than King’s 274 Hz on the same instrument.
- 0.22 avg volume. Range 0 to 1. A studio reading, with no room full of MPs.
- ~40% of the full text. 1,491 of 3,777 words, in several non-contiguous passages.
Churchill’s own 1949 studio reading of part of the speech: a different performance, measured as one.
Two things in this data deserve scepticism, and we’d rather say so than let the numbers stand unqualified. First, the pause count this entry’s corpus file records (122 gaps over 0.5 seconds) cannot be trusted as a count of Churchill’s own breath pauses. Most of those gaps cluster at durations just over the threshold, consistent with an edited recording’s cut points between the passages it kept, and not with a speaker pausing; only the longest, a 9.3-second silence near the very end, reads as plausibly deliberate. Second, the caption source for this recording carries almost no sentence-ending punctuation, which makes the sentence-length structure this corpus would otherwise compute (as it did for the 1940 text) unreliable here, so this entry doesn’t carry a rhythm or repetition figure. What the audio does support is pace, pitch and volume, and that’s what’s shown above.
What the numbers are good for: a real, if partial, sense of Churchill’s own voice. It is measurably calmer and lower-pitched than the recording most people know this series’ other subjects from, read at a pace that outruns what the full 1940 text implies, in a studio with no one to convince but a microphone.
Four recordings, and a slope that is ours
We later measured three more Churchill recordings (two 1940 broadcasts and the 1946 Fulton address), so the 1949 reading is no longer a sample of one. Set side by side they line up by something that has nothing to do with oratory.
We will be blunt about this, because it is the least flattering thing this series has found about its own instrument. Read naively the chart says Churchill declined over six years. What it shows is that our analysis was built to score a three-minute practice clip and scores long recordings lower. That bias runs through all twenty recordings in the corpus, and shows up most clearly in the speaker whose recordings vary most in length.
Where to look in your own talk
Repetition is worth noticing even when it does not announce itself at the start of a sentence. The most quoted device in one of the most quoted speeches of the twentieth century lives entirely inside a single sentence’s internal commas, which is precisely the kind of repetition a member building their own closing line can use and a crude automated check will never flag either way. The instrument here is looking in the wrong place, and knowing where it can’t see is as useful as knowing where it can.
The part that does transfer directly: the ratio. Two-thirds of this speech is unflinching bad news before it earns one sentence of resolve. If the resolve at the end of your own talk isn’t landing, a better closing line rarely fixes it. More often, too little of what came before it was frank enough to make the ending credible.
The same text analysis our members’ practice recordings go through. No audio analysis was possible, because no recording of this delivery survives.
Measured
- Word count against an estimated elapsed time
- Filler and hedging counts and rates, from the text
- Vocabulary diversity: 34.4 percent of words distinct
- Sentence count and the length of every sentence
- Positions of every sentence-opening phrase that repeats three times or more
Inferred, not measured
- Duration is estimated from the length historians attribute to comparable Churchill addresses of the period, not measured from a recording
- Sentence boundaries follow the punctuation of the reprinted text
- Repetition inside a single sentence, including the speech’s best-known passage, cannot be detected by a method that only looks at sentence openings
Sources
- Text. We Shall Fight on the Beaches, House of Commons, 4 June 1940. International Churchill Society. © Chartwell Trust, reprinted by permission of Curtis Brown, London. Quoted here only in short, attributed fragments.