As I recall, some readers of The New Yorker felt a bit of a shock in 1998 when they read Toni Morrison’s characterization of Bill Clinton as “our first Black president.” I was only a bit less electrified when it occurred to me, shortly after noon on Martin Luther King Day, that Donald Trump may be our first post-modern president.

By “postmodernism” in this context I mean Foucault, the most influential of the post-modernists who have focused on politics and the state.  If you want the political Foucault in a nutshell, it’s this: systems of power don’t just depend on structures of discourse, they somehow are those structures. How people talk about prisons or clinics or sex in the city guides and creates those institutions, and whoever controls the discourse controls the levers of power.

Which takes us to the Gulf of America, as viewed from a long way off—say, from the peak of Mt. McKinley. Trump has always been focused on names, describing, and discourse, and he has made his fortune on turning words into money, and money into power. If he says the Trump brand is worth millions, it is. (Unless, of course, a court rules otherwise, as it did in two civil and criminal cases in New York. As my sister in Texas says, by way of capping a list of Trump’s flaws, “He may not even be rich.”) In attempting to bend reality by bending words and applying his Sharpie to the Gulf of Mexico and Denali, our 47th president aligns himself with the crowd of renamers who have changed the labels of buildings and holidays in the belief that doing so somehow made injustice go away.  It won’t. Reality—not things, as the Wittgenstein of the Tractacus said, but the alignment between things and how we describe them—has a way of refuting the postmodernist belief that discourse is all there is—that you can turn a brand into dollars, or a meme into money.  In the next few years, let’s cling tight to facts and call things what they are, as best we can tell.

~Lee T. Pearcy
January 22, 2025

I don’t go into Bryn Mawr’s Canaday Library as often as I used to, before the building of the wonderful Carpenter Library for archaeology, art history, and classics, but last week I found myself in Canaday’s main lobby to pick up a book from Interlibrary Loan. The lobby was busy. Students sat at long tables cluttered with books, scissors, glue, ribbons, and other craft supplies, and heaps of other books waited on carts. A librarian I know called me over to look and explained. The students were learning to fold books into hedgehogs, flowers, hearts, and other decorative shapes.

The books, my librarian friend assured me, had all been deaccessioned and were available online.

Here, I thought, is one of those moments when one can see the old order of knowledge giving way to the new. The books on the carts appeared to be mostly back issues of periodicals, which are indeed available online and therefore surplus to requirements at a modern academic library. To the undergraduate students especially, academic books—the physical, bound volumes—are beginning to be alien objects. In 2019, when last I taught a course at Bryn Mawr, I assigned a few books and had the bookstore stock them. Most of the students preferred to read them online, and the bound volumes languished in the bookstore. I had imagined a lost world, where “reading” for a course meant underlining, writing in margins (always in pencil, as one of my teachers advised me, so that you can eradicate stupidities when you’re older), taking notes in a notebook, and for much of my reading thumbing a dictionary and commentary beside the text that I was reading.

As I walked down Canaday’s front stairs, I thought of mummy cartonnage. In Egypt’s Ptolemaic and Roman periods, old papyrus documents, surplus to whatever the requirements of the day happened to be, were often torn into strips and converted to a kind of gesso-covered papier-mâché used to make mummy casings. When these cartonnages are dismantled, sometimes an important text, or parts of one, can be read—thus we know, for example, about Sappho’s poem on old age, and some new bits of Simonides. Classicists have to envision time in centuries and millennia, not the three or four decades that it’s taken to push into oblivion those books on the carts, waiting to be turned into hedgehogs. Will someone, someday a millennium or so from now, sit patiently unfolding a hedgehog hoping to recover what remains of volume 27 of Rheinisches Museum für Philologie?

~Lee T. Pearcy
March 29, 2024

This fall I’ll be offering some informal workshops at Bryn Mawr on Latin verse composition—the translation of English into Latin verse.  In view of all that has been written about the impact of artificial intelligence on academic work, I thought I’d better take another look at ChatGPT and see if it was any good at producing Latin elegiac couplets.

The short answer is, it’s not very good at all; or at least, I haven’t been able to get it to produce a line that scans, and it doesn’t seem to have much understanding of prosody or meter. I won’t bore you with the details, gentle reader, but I’ll be glad to hear from anyone who can produce a better result than I have.

At first thought, this seems puzzling. Making Latin verses, after all, seems like a mechanical process—so mechanical, in fact, that in the 1830s an English inventor, John Clark, produced a machine to do just that. (John’s cousin Cyrus founded Clark’s Shoes, which still owns the device.) If gears and wires and cogs can produce a Latin hexameter, and play “God Save the Queen” at the same time, why can’t ChatGPT?

There may be several reasons. The simplest possibility is that I’m asking the wrong questions, and that I haven’t yet hit on the combination of prompts, definitions, and rules that would lead ChatGPT to scan a verse correctly and produce metrical lines in Latin. Another possibility is that ChatGPT doesn’t (yet) have enough metadata about Latin words. It doesn’t, for example, know which syllables are long by nature. To ChatGPT, that is, venit = “he came,” with its long first syllable, and venit = “he comes,” with its short e, are indistinguishable, at least as far as their metrical shape is concerned. But the possibility that interests me most has to do with the nature of poetry and a possible limit of large language models like ChatGPT.

In poetry, words have heft. They are not mere signifiers pointing to something in the world, in the way that “cat” points to the animal curled up by the fireplace. They have a material presence, a shape and weight—even an agency–that come from the essential nature of language as sound in the air and give poetry its special powers. In Latin verse, that materiality manifests itself in the shape of words—dactyls, spondees, trochees, iambs, and the like—and in their positioning into feet and lines. In Vergil and Ovid and Horace, words don’t simply mean different things. They are different things.

For a large language model like ChatGPT, though, words are nothing except abstract symbols to be manipulated according to a set of probabilities. A word points to nothing except the word that is most likely to follow it in a particular context. There is no connection between between word and world, between signifier and signified, not even so arbitrary and tenuous a link as that between “cat” and the furry animal by the fire. Asking ChatGPT to recognize some other quality in words—their shape or scansion—and to produce Latin verse by arranging them into patterns amounts to asking it to treat words as objects and to do things with them. My attempt, at least, was no more successful that if I’d asked ChatGPT to fix my dishwasher.

When I contemplate the grim future laid out in some projections of the effect of artificial intelligence, it is no small consolation to know that words are not all there is, and that making things work in the world—fixing dishwashers, writing Latin verses, painting a picture—is still a human thing.

~Lee T. Pearcy
September 10, 2023