Another month, another roundup of great book-related things I’ve read but didn’t manage to blog about. Let’s start with these guys:
Punctuation Personified
The image above is taken from Punctuation Personified: or, Pointing Made Easy, an 1824 primer for English boys and girls. They were part of a larger pedagogical series published by John Harris to meet the growing needs for children’s literacy in Britain. Grammar books were already unlikely best-sellers of the time, but children needed something a little more entertaining, so in leapt Mr. Stops, the punctuated person pictured above. In rhyming couplets, Mr. Stops guides little Robert through the wondrous world of punctuation, meeting with figures like Counselor Comma,
Who knows neither guile nor repentance;
A straightforward path he resolves to pursue
by dividing short parts of a sentence.
Here he is:
I found these, as I find most good things, in The Public Domain Review, introduced by Hunter Dukes. The illustrations are, alternatively, delightful and deeply, surrealistically weird, as children’s art from past eras inevitably is. I’ll be sprinkling them throughout this month’s link roundup for visual zest.
Punctuation Mummified
Punctuation Personified is an old book, but in the five millennia of the written word, most punctuation is actually quite new, most of it no more than a thousand years old. Whether in China, Mexico, or Iraq, early writing was always big blobs of uninterrupted, unpunctuated text. It wasn’t really until the 8th century Carolingian Renaissance that writing started to look more modern, with spaces, capital versus lowercase, and consistent paragraphing beginning to show up. This is the standard history you get from books like Keith Houston’s Shady Characters or Dennis Duncan’s Index, History of the. You should definitely read those books, because they’re great, but one thing they don’t get into very much is what it must have been like for ancient scribes to deal with these blocks of dense text all the time. We don’t have a lot to go on here, but Florence Hazrat rounds up the evidence that does exist for Antigone. Reading between the lines of plays, poems, and oratory, we know that sight-reading was so hard, it was considered a serious skill; that teachers of children often punctuated texts for their readers, much as Hebrew and Arabic writing still does today; and that poets actually exploited the lack of spacing for cool poetic effects. Vergil, for one, drops Latin bombs in The Aeneid like COLLECTAMEXILIOPUBEM. Depending on the spacing, that either scans as “a people gathered from Troy” (ex Ilio) or “a people gathered from exile,” (exilio), but both readings are equally true of Vergil’s Trojans. We don’t have this in English anymore except, as Hazrat points out, in website URLs. So at least we’ll always have penisland.net.1
The Most Expensive Books Ever Sold (Update)
This month, a nearly-complete Hebrew Bible from the 9th century AD sold at Sotheby’s for $38 million, making it one of the top five most expensive book sales in recorded history. At least, that’s according to this Wikipedia list. What kinds of books sell for the highest rates, anyway? Looking at the list, it’s pretty much what you’d expect: a copy of the United States Constitution, a printer’s manuscript of the Book of Mormon, one of Da Vinci’s notebooks, and old Biblical manuscripts. Three of the ten are Chinese. More intriguingly, two of them aren’t books at all, but letters from eminent medieval scholars. They’re not big tranches of letters, either: somebody paid $31,730,000 for a single letter written by the philosopher-poet Zeng Gong (1019 - 1083). Compared to that, the two letters of Zhao Mengfu that sold for $38.2 million were a relative bargain, at only $19.1 million per letter. This might have something to do with China’s greater reverence for calligraphy, but I’d be out of my depth speculating. More importantly, in the midst of all these heavy tomes about God and government, two of the most expensive documents in the world are simply as letters to friends. Isn’t that nice?
The First AI Novel (That Doesn’t Completely Suck)
The age of the algorithmic novel is upon us. Amazon’s Kindle Store is flooded with AI trash, and literary magazines like Clarke’s World have been so inundated with machine-written stories that they’ve glued their submissions mailbox shut and only publish commissioned stories now. LLMs still can’t write good fiction—and as Lincoln Michel points out, it doesn’t really matter if they can—but among all the shitty attempts, this month’s The Death of an Author might credibly claim to be the first mediocre AI novel. Stephen Marche, who has been on the AI writing beat since 2017, estimates that by feeding various LLMs descriptions, requests, and queries, he was able to get software to generate about 95% of his 100-page murder-mystery. You can read an excerpt for yourself at Wired, and Marche’s afterword explaining his methods and ideas is available at The Atlantic. Having read the excerpt: it’s not terrible. It’s not good, either, but it’s better than most attempts at LLM fiction that I’ve seen. Considering how much work it is just to get LLMs to stick to a consistent voice and remember what the weather is supposed to be like in a scene, I can recognize the enormous amount of prompting it must have taken Marche to do this. Still, I agree with the Michel piece linked above: this is a party trick, and a cool party trick, but using LLMs to imitate the novel is a dead-end. AI fiction will prosper if it can find ways to write generative, customizable, interactive narratives—that is, something human writers can’t do.
The Labor of Audiobooks
A few decades ago, audiobooks used to account for, at best, about 10% of a book’s total sales. These days, that’s more like a baseline, with the upper range topping out around half of all sales. That’s according to this jointly-authored article at The Los Angeles Review of Books. For Big Five publishers, making audiobooks has gone from a niche service reserved for top-sellers to a necessary part of the production package for almost every book. This requires a whole new workforce of professional narrators, with responsibilities that are still being defined: besides voice acting, audiobook teams are increasingly becoming a kind of extra set of editors, offering suggestions and interpretations to improve how a book reads.
If a narrator is a member of the Screen Actors Guild (and many are—the auditioning-actor-to-Audible-narrator pipeline is vast), she might make upwards of $200–$250 per finished hour. In other words, for portraying a junior lawyer in some legal thriller, she can bill almost as much as a real junior lawyer does. When you factor in the time it takes to prepare for a book, however, and the fact that a single hour of an audiobook usually takes about three hours to record, the real hourly wage for that narrator might be closer to $70. That is far more than you can earn by delivering portobello burgers or boba tea, but the work is just as varied, taxing, and unreliable.
Like most acting gigs, however, the work tends to be feast or famine. One voice actor tells the authors that beginning narrators need to read around 70 books a year—about six per month—just to pay their rent and establish a name for themselves. More importantly, narrators almost never make any royalties whatsoever from their work, even as their labor is becoming increasingly important. But since there’s a constant stream of young, hungry voice actors who can work from anywhere at any time, and with AI voice narration getting better quite literally by the day, narrators have no bargaining room, and have to accept what they can get.
No Russians Allowed
There was a big stir earlier this month in international letters when Russian-American journalist Masha Gessen suddenly resigned from the board of PEN America. Gessen didn’t give an immediate reason, though it clearly had to do with a PEN World Voices conference featuring panelists from both Russia and Ukraine. The Ukrainians, obviously, were uncomfortable with Russians—even “good” Russians opposed to the vulgar, bloody oprichniki squatting in the Kremlin and bellowing outright lies to the world—sitting next to them onstage, or even appearing at the event at all. The Russians were disinvited, and Gessen resigned.
Last week, Gessen finally opened up in an interview with Isaac Chotiner at The New Yorker. The story is pretty much how it looked on the surface: Gessen is a free speech advocate, they had personally invited the Russian writers, and done everything they could to separate them from the Ukrainians. When PEN America went over Gessen’s head and disinvited the Russians, resignation seemed seemed better than acquiescence. But Gessen doesn’t come across as angry or bitter—in fact, they’re very empathetic towards the Ukrainians, who felt, as Ukrainians often do, that Russians tend to upstage and overshadow them at every turn, from coverage of the war to prominence at, say, PEN conferences. Highly recommended reading if you’re into Eastern European literature, as I am.
Context is Everything (AI Edition)
Between the world-straddling ChatGPT, Google’s continuing attempts to make Bard happen, and Facebook continuing to insist that LLaMA is totally going great and we’ll all get to try it really soon, tech startup Anthropic’s Claude LLM has been a bit of an also-ran in the last few months. That might change with this month’s announcement that Claude’s context window has been expanded tenfold to 100,000 tokens. In civilian terms, that means Claude can read and remember around 75,000 words of English prose (for comparison, ChatGPT can only handle about 3,000 words at a time). Anthropic tested this by feeding the entire text of The Great Gatsby. After a few minutes, the program had read the whole text, and was ready to provide accurate summaries and answer quiz questions. Obviously this is just a snappy demo for the program’s real uses, which will inevitably be boring corporate analytics, but as a bookworm, I want to see Claude reading more literature and history. You could feed a program like this hundreds of documents—the birth records of some medieval Welsh village, business contracts from 19th-century France, shards of Greek text from archaeological digs—and come up with all kinds of fascinating leads. AI is probably dumb and overrated as far as it comes to igniting a new industrial revolution or destroying life as we know it, but if all this only turns out to be a trillion-dollar investment in digital humanities scholarship, I’ll be satisfied.
If you’re reading this at work or in public, I promise you it’s not porn. It’s Pen Island. Get your mind out of the gutter.