I am a human who did reasonably well academically and reads a lot (although I didn't study English in formal education past about age 16). If I'm honest, even with the breakdown, I'm not really seeing strong parallels between all pairs of sections X and X'. The first attempt that said "this is mostly just a linear progression from problem to resolution" is probably also what you'd get from my first attempt.
I would however be more than capable of emitting text saying "1 mirrors 1', and 2 parallels 2', and 3 corresponds to 3'" if I had the idea that this was what you wanted because you'd rejected my first answer and provided the sections. So long as you didn't push too hard on asking me to explain exactly how and why those particular parallels.
Try this:
While ChatGPT has more or less captured the parallelism in in the St. George story, I thought I’d say a bit more. Here’s slightly different version of I came up with in my original analysis:
1. Impurity: Dragon poisons the town.
2. Appease the dragon, first with sheep, then with children.
3. The king pleads to the villagers to excuse his daughter from beiing sacrified. The villagers refuse.
4. Daughter sent to the lake, dressed as bride.
5. Saint George arrives. Princess tells him to go.
Ω. Saint George wounds the dragon, under Christian protection.
5’. St. George captures the dragon with the Princess’s girdle.
4’. The princess and St. George lead the dragon back to Silene.
3’. King & people convert to Christianity so that St. George will slay the dragon.
2’. George kills the dragon.
1’. Purity: Church and spring with curative waters.
The impure state at the beginning (1) is mirrored by the pure state at the end (1’). The town was sick now it is a source of curative waters. The townspeople attempt to appease the dragon, to no avail (2). But then, just before the end, St. George slays the dragon (2’), eliminating the need for appeasement.
The king and the villagers are at odds over his daughter in 3. But they are united in 3’ and convert to Christianity. In 4 the daughter is sent to the lake, and thus presented to the dragon, as a bride. In 4 the daughter and St. George lead the dragon back to the village. In 4 the dragon had been driving the action, but in 4’ it is passively led.
In 5 St. George just happens upon the lake where the princess tries to send him away. In 5’ St. George calls on the princess (“give me your girdle”) to help him lead the dragon away. Finally, in the center (Ω) St. George takes command of the dragon by wounding him and thus bringing it under his power.
Cross-posted from New Savanna.
Tentatively, very tentatively, yes. ChatGPT has done it, once. I believe a more powerful engine could do more.
But: What is ring-composition?
Quickly and briefly, ring-composition or ring-form is a text with a form like this: A, B, C...X...C’, B’, A’. It’s a text with a central section and the other sections are arrayed symmetrically around. ChatGPT will say a bit more about that later.
Why am I interested in ring-composition? In principle I’m interested in any and all literary form. In practice it is easier to look for a specific kind of formal structure. While some interesting and important texts exhibit ring-composition (e.g. “Kubla Khan,” Heart of Darkness, Hamlet) I have no idea how prevalent it is, nor does anyone else. I suspect it’s a minority form, perhaps even a small minority.
I have come to think of literary study as something like biology: it all starts with description. Biologists have spent centuries seeking out and describing life forms. Some have been described in great detail, others only enough to place them in the taxonomy, and we have all degrees of description in between. Well, literature is like biology in that we have a myriad of objects for study, each with unique features. But literary scholars haven’t undertaken the work of describing our texts. Oh, there’s some work, but little as deep and rich as what biologists have done in their domain – I give links below to material that justifies that claim.
I am particularly interested in form. There’s an inerradicably subjective element to meaning, but form, I believe, can be objectively described. I’ve done enough descriptive work over the years to know that it is painstaking, difficult, and tedious, but not rocket science. It would be useful if we could present our texts to an LLM and have it undertake the difficult tedium of descibing texts. Maybe even have two or three different LLMs describe particular texts independently. Then, just as biologists check one another’s descriptive work, so human critics can vet the work of LLMs. Descriptions of important texts would be examined in great detail, while muc of the work that LLMs have done on minor texts, those texts that no literary critic ever works on at all (like all those 19th century novels lurking in basements in library shelves) may never be checked at all. But it’s there if anyone’s interested.
But could LLMs do the work? We don’t know. In the rest of this post, however, I present some small indication that, yes, they might well be able to do it.
First time out: What’s ring-composition? [December 1, 2022]
The first time I had access to ChatGPT I asked it about ring-composition. It’s response was underwhelming.
My prompts are in bold while ChatGPT’s replies are in plain face.
The definition is OK, but not at all specific. It would not be a very useful guide. The list of texts seems OK, though Moby Dick seems iffy to me. Critics working on 19th century novels don’t go looking for ring-composition. The answers to the last two questions are too vague to be at all promising. On the whole I had no reason to believe that ChatGPT had a useful idea about ring-composition.
Then, without any specific expectation – though I wasn’t expecting much, I decided to give ChatGPT another try. Not the new and improved ChatCPT Plus, which is hooked up to GPT-4, but just the plain old Chatster. It surprised me.
Much better: What’s ring-composition? [August 31, 2023]
It’s running the August 3 version.
That’s far more detailed that it gave me back in December. The question is, can it analyze and actual text and identify a ring-form? I decided to have it work on a specific example, the story of St. George and the Dragon, which I had already analyzed. I’m using a text from Wikipedia.
Since I know that it does exhibit ring-composition I immediately checked the thumbs-down button. As soon as I had done that ChatGPT offered another answer, which bewildered me as I hadn’t seen that behavior before. Here’s that second answer:
That's better. I decided to push it a bit further:
Chatster’s explanation of the symmetry is correct. Given that I had presented it with the breakdown – I have no idea whether or not it could have done that itself – producing that explanation would not have been difficult, certainly not for a human. But I find it hard to deny that what it did required reasoning, not deep reasoning, but reasoning nonetheless.
What's next?
Surely a more powerful engine, one fine-tuned in the analysis of literary form, could do better. What about film? Films exhibit ring-composition as well. Some I’ve analyzed: Pulp-Fiction, the 1954 Gojira, the 1933 King Kong, Miyazaki’s The Wind Rises, and episodes of Disney’s Fantasia.
I would like, however, to end on a cautionary note. The descriptive and analytic task I gave ChatGPT was a relatively easy one. I gave it a particular text, a short and uncomplicated one, and asked it simply to identify whether or not it exhibited a particular formal property. Given that it had already given a reasonable account of that property, it had some of idea of what to look for.
Still, it got things wrong. Until I told it that it was wrong. Then it gave me a more or less correct answer. Whey didn’t it give me that answer first?
But there is a deeper problem. When I am working on texts, I do not know whether or not a given text is a ring composition. There are no identifying traits that I have been able to see. You begin the analysis and see where it goes. Maybe ring-composition will emerge and maybe it won’t. What I want from an LLM that’s working on formal analysis is that it be able to work on any text and find the formal features that are there. That’s an open-ended task. While ChatGPT’s performance with St. George and the Dragon is significant, there is a long way from that to the kind of assistance I am looking for in a LLM.