In Milan, in the mid-380s, a young African rhetorician went to call on the bishop and found him reading. That is the entire incident. Augustine wrote it down anyway, because he could not account for it: Ambrose sat with the book open, his eyes travelled the page, and his voice and tongue stayed still. Augustine and his friends stood in the room and watched him do it. They left without asking what it was. He was still turning it over years later, in the Confessions.

The passage gets read as a story about Ambrose — a great mind, a guarded man, an early glimpse of the interior life. It is really a story about the page, and about who was equipped to read one.

A Latin book of the fourth century was an unbroken ribbon of capitals. No spaces between the words, no lower case, punctuation sparse and unreliable. THETEXTRANLIKETHIS, line after line, and the reader's first task was not comprehension but division: finding where one word ended and the next began. That work was done by the voice. You read aloud, or under your breath, and you heard the joins as you cut them, and the sound told you whether you had cut in the right place. Reading was a performance before it was an understanding. The majority view among classicists is still that reading aloud was the rule in antiquity and silence the oddity worth remarking on — which is precisely what Augustine did with it.

So Ambrose's trick was not spiritual, and it was not genius. He could read without his voice because Latin was his own language. A native ear does the dividing for nothing.

Nine hundred years then went by without anyone putting the spaces in, and the obvious explanation is the wrong one. A fourth-century page was not an object to be looked at. It was a score to be sounded, and the divisions were the reader's own work, done with the mouth. Marking them in advance would have been like printing the fingering above a tune a man had played since childhood: no help to him, and faintly insulting. A trained reader did not meet the unbroken line as an obstacle. He met it as the page. That the same arrangement kept out everyone without the training was true, and no one in a position to change it was troubled by it.

The spaces came from the other end of the world, and from people who were failing at Latin.

In the seventh and eighth centuries, in Irish monasteries and then in the Anglo-Saxon houses the Irish founded, scribes began cutting the ribbon apart. Paul Saenger's account is the standard one: word separation, and with it silent reading, begins in manuscripts copied by Irish scribes, and does not reach the Continent in any general way until the late tenth century.

The reason is the part to hold on to. A sixth-century Italian still spoke a Latin close enough to Jerome's that his native intuitions carried him through an unbroken line; he did not need the joins marked because he could hear them coming. The Irish and the Anglo-Saxons had no such inheritance. Their own languages were nothing like Latin, they had learned it out of books as a wholly foreign tongue, and their manuscripts fill up accordingly with aids for a reader in difficulty. They were not exporting a native habit, either. Irish grammarians analysed their own language in groups rather than in separate words, which is exactly the instinct least useful to an Irish student parsing Latin. Which is why early Insular separation is full of characteristic errors: prepositions still fused to their nouns, joins made in the wrong place by men working the rule out as they went.

They put in the space because they could not hear it.

Then it took three hundred years to travel, because the people it had to reach could still hear well enough to manage without it. The advantage was the obstacle. Every line on this page was cut apart for you in advance by someone who found the language hard.