Methodical banality

Like today's great linguistic models, 16th-century humanists used techniques to automate writing – to the detriment of novelty and originality.

By: Hannah Katznelsons / AEON
Translated by: Agron Shala

Renaissance scholar and educator Erasmus of Rotterdam opens his polemical treatise Ciceronian [Ciceronianus, sive de optimo dicendi genere, 1528] describing the completely messy writing process of a character called Nosopon. Ciceronian is structured as a dialogue in which two accomplished writers, Boulephorus and Hippologus, attempt to dissuade Nosopon from his paralyzing obsession with stylistic perfection. Nosopon explains that it takes him weeks of futile writing and rewriting to produce a simple letter, in which he asks a friend to return some borrowed books. He says that writing requires such intense concentration that he can only do it at night, when no one else is awake to distract him, and even then his perfectionism is so pronounced that a single sentence takes him an entire night. Nosopon constantly revises what he has written, but remains so dissatisfied with the quality of the language that he finally gives up.

The Nosopon problem may sound familiar. Who hasn’t spent a lot of time mulling over the wording of a simple email, at least once? Today there’s an easy solution: we have large language models (LLMs) to write our letters for us, offering suggestions for what we might say and how we might phrase it. When I fed Nosopon’s intended query into GPT-4, it immediately generated this:

Hey [Friend's Name],

I hope you're doing well! I just realized that I haven't received those books that I lent you a while ago. Don't rush, but when you have a chance, I'd like to get them back. Let me know what you think! Thanks!

Nosopon

But there was a solution even in the 16th century. A humanist education along the lines of Erasmus could train students to produce papers of any length, on any topic – quickly, easily, and eloquently. The French humanist François Rabelais, a contemporary of Erasmus, seems to have understood this writing technique as a way to automate text creation in a way that, from today’s perspective, looks a lot like how LLMs work. If we want to understand the great linguistic models – what they are and what they are not capable of doing – we can look to earlier versions of the same technology: like Erasmus’s humanism. We can also read authors like Rabelais – who thought about automatic text generation in this regard – as someone who appreciates the effectiveness of Erasmus’s generative technology, but at the same time sees it as detrimental to the social force of language and, ultimately, as a destruction of language as a means for moral and political life.

Rabelais was a monk, physician, personal secretary, and spy, but today he is remembered primarily for his five-part literary prose, Gargantua and Pantagruel [Les Cinq livres des faits et dits de Gargantua et Pantagruel], a family of aristocratic giants (in the literary sense) who cope with life in 16th-century France while being much larger than everyone else. His second book in the series, Gargantua (1534), deals essentially with the advantages and costs of a humanist education. Those of his characters who receive humanist training produce distinctive and sophisticated language, but when they speak or write, they always communicate first and foremost the fact that they have had an extraordinary education – not so much what they actually have to say.

When we read a sentence such as: “No more just cause for pain can arise among men than when, from those from whom courtesy and kindness are justly expected, they receive torments and injuries” (the author’s translation of Rabelais), or we take the modern chatbot equivalent: “It is deeply depressing when those who are charged with the care or respect of us, instead subject us to ill-treatment” (GPT-4) – we are more likely to be impressed by the linguistic skills of the author or the linguistic model he has generated, than to reflect on the idea that it is bad when someone who should treat you well treats you badly.

Erasmus and Nosopos both wrote in Latin, which language, like every humanist, they had learned through careful imitation of the great Latin writers of antiquity (despite the substantial difficulties of achieving native fluency and eloquence in a second language). Renaissance intellectuals argued vehemently that Latin whose was to be imitated, although in fact they had very little disagreement. Cicero was the prime choice; his position as the best prose stylist in antiquity was undisputed. The main debate, then, was between the Ciceronians, who believed in imitating their hero to the exclusion of all other models, and the eclectics, who believed in imitating Cicero along with other models. Erasmus was eclectic. In the course of Ciceronian, his representative, Bulephorus, gradually persuades Nosopon, the leading Ciceronian, to side with the eclectics. To understand the arguments he presents and the issues at stake in Nosopon's conversion, you need to understand how humanist stylistic imitation worked.

Humanists read and reread the classical authors they wanted to emulate in their writing style—not unlike the way a large language model (LLM) is trained on a corpus of data. They absorbed, as much as possible, the prose qualities of their models and also kept a sort of vocabulary for bad days in something called a commonplace book, which means that as they read, they would transcribe a host of prominent words, metaphors, idioms, and clichés—usually organized by theme—which they would then use in the writing process. An eclectic writer would gather the strengths of many different authors and genres, while a Ciceronian would copy only from Cicero. Writing with a notebook would therefore allow an eclectic to benefit from a wider range of reading to make his prose more versatile, while a Ciceronian's style would become increasingly homogeneous in Ciceronian terms. Boulefor notes that Cicero himself read widely among his contemporaries: if Cicero were alive today, he argues, he would produce texts adapted to the modern world and, consequently, quite different from anything in the surviving Ciceronian corpus. In other words, Cicero would be eclectic.

But, Ciceronian makes it clear that Ciceronians and eclectics differ not only in the content of their notebooks, but also in the way they used them. For eclectics, they are a kind of safety net: if Bulephorus cannot find the right word, he can simply refer to his notebook, just as we can refer to a thesaurus, saving himself all the time he would otherwise have spent looking it up himself. For a strict Ciceronian, however, a notebook is more like a safety checkpoint. The nosopon generates language without help, but feels the need to meticulously verify the Ciceronian origin of every word or linguistic structure he wants to use in his notebook, which, on the other hand, has become a tedious list of Cicero's corpus. This insistence on conscious and enforced control of the writing process makes it impossible for him to write fluently, to the point where he refuses to choose verbs without first making sure that Cicero has used the particular form of choice he needs. It is no wonder that a single sentence takes him an entire night of work.

Erasmus’s aim as a teacher was to train his students in precisely the kind of fluency that Nosopon lacks. He achieved this by giving his students fictional or historical scenarios that required sensitive rhetorical treatment – ​​in a way very similar to how we might instigate a major linguistic model today – and by having them write letters or speeches “in role”. For example, they might write a love letter from Paris to Helen, or they might write like Agamemnon, calling on his allies to make war on Troy, or else begging Menelaus to forget Helen and avoid the terrible human cost of such a war. They might write like Menelaus, cursing Paris for stealing his wife, or perhaps forgiving Helen for abandoning him. (Erasmus gives a host of examples, including many of these, in his pedagogical manual On writing letters.)

When Rabelais’s humanist characters finally get a chance to put their training into practice – trying to avoid a war with their neighbor, Picrocol – the texts they produce read like exercises from an Erasmian schoolroom. The giant Grandguzie sends his son, Gargantua, a letter urging him to return to defend the family’s land from his war-mongering neighbor, Picrocol. At the same time, Grandguzie’s ambassador, Ulrik Galeu, delivers a speech against the war, urging Picrocol to desist from his aggression. Rabelais gives the full text of the letter and the speech. They sound good, but they are quite long and quite boring: there is something generic about them. They could just as easily have been produced by Renaissance students who were stimulated by these scenarios, or by LLMs who simulate them.

Moreover, no function. Grandguzieu and Galeu may decry the horrors of war, insisting on the moral obligation to avoid it, but both speech and letter, in fact, hasten violence – rather than prevent it. They fail to make peace because they are not specific.

Language automation, whether by Erasmians or by large language models (LLMs), relies on a rejection of novelty: both function identically by breaking down seemingly new situations and topics into familiar elements, so that those situations can be addressed with language that already associates with these elements in the training corpus. What this means for Grandguzieu and Gale is that the humanist mentality that enables them to speak so well also leads them to enter into conflict with a certain arrogance – with the assumption that they can anticipate everything the other side might have to say, based on what they have already read.

Indeed, the irony of their situation lies in the fact that Rabelais' humanist characters constantly emphasize the necessity of establishing a dialogue with Picrocol, but prevent it from happening, precisely with the language with which they express the importance of listening to his perspective. This language is not peaceful, but defensive.

Grandguzie writes to Gargantua saying: “I have kindly sent to [Picrocolus] several times to understand in what, by whom, and in what manner he thought he had been wronged, but I have received no answer from him, except a deliberate challenge.” But this is not true. Grandguzie sent his ambassador to Picrocolus only once, not several times—and finally so that Gale could not reach him, let alone return with an answer.

Perhaps Grandgozieu is lying because he has no real interest in avoiding war, and the letter serves to create a footprint that establishes geopolitical moral superiority. But what if we consider this false claim not as a lie, but as something closer to the so-called “hallucinations” of grand linguistic models (LLMs)?

An LLM responds to a request with text that it “considers” possible or appropriate in the context set by that request. Because it is so context-sensitive, it will tell me about the unicorn if I ask it for a fairy tale, but not if I ask about the wild animals of North America—but that does not mean that it has an internal representation of reality, or of the meaning of what it means for something to be real (although, of course, when asked, it can generate text describing such a representation). It does not “understand” that Reason The reason why words like “unicorn” and “raccoon” appear in different kinds of contexts is that raccoons really do exist. This is acceptable as long as we ask LLMs for information that is common and consistent—like how many teaspoons are in a tablespoon. But when we ask them for something unusual or highly nuanced, they may simply fabricate something that seems like a plausible answer—like a bogus legal precedent (where LLMs’ performance is no better than a random guess), or an NBA lineup with players from the wrong teams.

Something similar may have happened with Grandguzieu, who can embody the role of a good Christian-humanist prince, in the same way that an LLM embodies the persona that its wearer assigns to him – as a set of discursive and rhetorical traits. After all, he is navigating a scenario that seems to have been taken directly from Erasmus’s moral teachings: we know that he “must” do everything to avoid war, because we have read Erasmus’s many anti-war writings. And, indeed, his letter says everything that we, or he himself, might generally expect from a virtuous prince in such a situation (or from a student of Erasmus who imagines himself to be one). And, so, while Grandguzieu engages in the highly trained and semi-automatic process of writing, he may write untruths as a result of a kind of discursive reflex – acting not against the truth, but without reference to it at all – again, like an LLM, who only approximates reality by saying the kinds of things that, based on his training, he expects users to expect him to say.

Recall Erasmus’s argument against Ciceronianism, which is that no single author can provide the general linguistic resources for every possible discursive situation. But the difference between Erasmus’s eclecticism and the strict Ciceronianism he ridicules is a matter of degree, not kind. In both models, the humanist can afford novelty only to the extent that the tools he draws from a given corpus allow him. But since the eclectic corpus is larger and more diverse than the Ciceronian one, its limitations are more subtle – and LLMs are even more subtle, because they are trained on corpora beyond comparison in size to those two approaches. And yet Rabelais’ humanists display the same solipsism as LLM users today, who, thinking “with” LLMs, simply refract their own thought through a body of words from others.

This is not wrong, but it is not communication.

Both – the diplomatic letter and the speech – are genres of communication, by definition: they exist to produce an immediate effect on the other person for whom they are written. The letters written by Erasmus’s students, by contrast, are pedagogical exercises that only the teacher reads. Gale’s diplomatic mission fails because he treats it as an exercise. Although he is sent to retrieve Picrocolus’ version, he merely simulates attempting to do so through a series of rhetorical questions: “What folly drives you now …? Where is faith? Where is law? Where is reason? Where is humanity? Where is the fear of God”? Somewhat ironically, he then reproaches Picrocolus for not easing the conflict through dialogue, even though he himself is failing to do so: “If perhaps any evil has been done by us … you should first have investigated its truth, and then reproached us for it.”

This brings us to another consequence of the solipsism of autonomous language, which is the degeneration of what the British philosopher JL Austin, in the 50s, called illocutionary force – that is, what it means to someone who has said something, as opposed to the (locutionary) meaning of what has been said. The illocutionary function of the language of an LLM is compromised, mainly because there is no unique social agent speaking: an LLM cannot marry, place a bet, or christen a ship (to use Austin's examples). But even when a person uses an LLM to generate language to send to another person, there is a loss of social ownership over the language in question – which, like an ambassador, is sent to address someone on someone else's behalf.

Gale’s diplomatic mission fails because he is blind to the pragmatic dimension of his language. His speech is extremely aggressive and offensive. He describes Grandgusia’s previous alliance with Picrocol as a “sacred friendship”; he tells Picrocol that his military aggression proves “that nothing is sacred or untouchable to those who are freed by God and reason to follow their perverse inclinations”; he says that his behavior has been “beyond all bounds of reason, so repugnant to ordinary feeling, that it can hardly be understood by the human mind.” And yet, if we consider Gale’s mission as a social act – it is clear how the deeply offended Picrocol perceives it – it clearly announces a desire for war, even though its narrative content is pacifist. We can read this tonal discrepancy in two different ways: it could be a cunning attempt on Gale's part to escalate the war while maintaining the victim's stance. Or, it could simply be that Gale simply fails to consider the illocutionary force of his speech, which he treats as a general rhetorical exercise on the evils of war, rather than as a diplomatic communicative act.

There is a simple ethical lesson to be drawn from all this. If we want to confront – and make peace with – the “other” – we must avoid the example of Gale, who sets his discourse in motion without considering the pragmatic effect it has on the listener. We must, instead, imitate Picrocolus himself, who greets Gale not with a speech but with a question: “What have you to say?”

By structuring the outbreak of war around a series of humanist failures to ask and listen, Rabelais characterizes those failures as moral, and locates them in the cognitive structures that constrain Erasmian rhetorical training. The charge is this: that Erasmus gives his students a technology for producing language as an end in itself, but does not teach them how to communicate. Rabelais shows us that, when the production of discourse is automated, it becomes strictly monological and loses its socially illocutionary power. This kind of autonomous language is like an ambassador: it speaks on behalf of OUR, but can't speak si us. /Telegraph/