Augmented Criticism and Rhetorical Figures

My subject is “Augmented Criticism and Rhetorical Figures.” If that sounds highly technical, let me assure you that Adam and I are literary critics first and digital humanists second. That is, we use computers only to augment traditional research inquities, that are rooted in philology.

Here, for instance, our inquiry is into rhetorical figures, or the patterns of repetition and variation that make poetic language memorable, compelling, and beautiful.

Let’s start with two beautiful examples:

The first is taken from Shakespeare’s Richard III: the king’s self-recrimination in Act 5.

There are at least two overlapping structural figures at work in this passage: including anadiplosis (or repeated words at the end of one clause, and the beginning of the next: … A, A …), and gradatio (or a sequence of anadiploses: … A, A … B, B … C, C … and so on).

Gradatio is the figure we’ll focus on in this paper. Its definition as anadiplosis in sequence is simple enough, and so is this example. It’s Richard’s narrative progression from tongues to tale.

Look at an example like this, and you can see how each repetition lodges a word(s) more firmly in your working memory At each step, Richard considers and reconsiders the meaning and implications of each word. Repeating a word reinforces it in the listener’s mind, varying its meaning, reapplying it to new contexts.

Now consider this more complex form of gradatio, from Troilus and Cressida; this is Pandarus’s address to Helen. This entire project began when I read Troilus, and noticed this compelling pattern. This instance is actually in prose; I just artfully rearranged it.

It’s also not just any gradatio, but an incrementum, whose movement isn’t just sequential or narrative, as in Richard’s mind; it’s also semantic.

That is, there’s a rise or fall of meaning. Consider the geographical semantics of Lenin’s statement: “whoever controls Berlin controls Germany, and whoever controls Germany controls Europe.” Its rise from city to state to continent makes it an instance of incrementum, which is (as I say) a subset of gradatio.

In Pandarus’s formulation, blood begets thoughts beget deeds; food causes the physiological response that causes a mental state that provokes action. This is semantic growth, not just sequential movement.

Now, let’s consider three early modern defintions of gradatio, which is sometimes also known as climax.

First, Thomas Wilson in 1553 simply defines it structurally, as anadiplosis in series, which he memorably describes as a ‘knitting’ of clauses.

Here is Abraham Fraunce’s 1588 definition, which gives us a little more semantic complexity. He adds the “degrees and steps” of meaning, suggesting there’s a distinction between the anadiploses.

Fraunce’s choice of the word ‘steps’ could imply a striding or a climbing movement (or both). But the word ‘degree’ is unambiguously hierarchical, as in Ulysses’ memorable speech on degree in Troilus (1.3):

O, when degree is shaked,
Which is the ladder to all high designs,
Then enterprise is sick! How could communities,
Degrees in schools and brotherhoods in cities,
Peaceful commerce from dividable shores,
The primogenitive and due of birth,
Prerogative of age, crowns, sceptres, laurels,
But by degree, stand in authentic place?

Those are just three of the seven times Ulysses repeats the word ‘degree’ in this speech. It’s all about ladders and steps, hierarchies and prerogratives.

And it contains this beautiful gradatio:

Then every thing includes itself in power,
Power into will, will into appetite;
And appetite, an universal wolf,
So doubly seconded with will and power,
Must make perforce an universal prey,
And last eat up himself.

This one is also an incrementum, by the way: a decline from power to will to appetite, the outcome that debases power. All you need for incrementum are, well, increments. For naming purposes, whether they incline or decline is immaterial.

The definitions culminate in George Puttenham, who characteristically ‘Englishes’ foreign/classical terms: “the marching figure.” His definition is a bit turgid – but in essence, to “proceed by double … to the first [word] that was spoken” is a fine summation of gradatio’s structure. Marching forward is a structural definition of the term.

Here, by contrast, you have a semantic definition: a rising movement, of climbing up a ladder (another of Ulysses’s terms). The epitaph is a good example because it describes a social ascent with a figure of ascent, from wisdom to wealth to friends. (It alos contains a tidy chiasmus in the third line, an inverted repetition of synonyms: weal … woe || sickness … health).

I could do this all day, quoting definitions and showing you examples, but you get the point: gradatio is a structural repetition, sometimes with semantic increments.

If you look again at Puttenham’s example, there’s a further complication. Sometimes, what’s repeated is just part of a word, or its dictionary head-word (what the linguists call its lemma). So ‘wisdom’ repeats ‘wise,’ here.

Here’s another example of that complication, Sidney’s gradatio in the opening sonnet of Astrophil & Stella. ‘Read’ morphs into ‘reading’; ‘know’ into ‘knowledge’.

As I said earlier, gradatio is a figure that’s memorable, compelling, and beautiful. Like all rhetorical figures. Like them, it’s also not too complicated.

Or is it?

My core argument in this paper is that doing digital text-analysis is a self-confounding, paradoxical action, like going west to go east. But it earns similarly rich results.

In sum, it reduces the complex features of a text in order to expand their complexity, by gathering multiple instances with common features.Each instance complicates your grasp of that commonality. Each one expands the set, making it richer and harder to encompass.

Take rhetorical figures: these linguistic structures are ready-made for digital text-analysis, because they are variations on a relatively simple theme: a linguistic structure of repetition and variation, simple enough to be memorable, simple enough that even a computer can understand them.

That’s a fallacy, of course: computers can no more understand texts than they can read them. But computers have the dogged patience (as it were) to take every word of every clause of a given text; and then to look up its dictionary head-word (say, for ‘reading’ > see ‘v. to read’); and then to compare each head-word with every other head-word from the preceding and following clauses.

What else is gradatio? We can reduce it to this kind of formula, a set of linguistic instructions, because that’s what it is: “A reduplication … of the same word or sound,” as Fraunce called it; wherein “one word proceeds double to the first that was spoken,” in Puttenham’s description.

Anything you can reduce to a formula, like that, you can also turn into an algorithm – or a process in which inputs are turned into outputs. In our case, we turn long, complex texts into short snippets of gradatio.

That’s what we’ve done, and in a moment I’ll show you some of our outputs. We began with Shakespeare (specifically, Troilus) and noticed this figure; we reduced it to this formula; and we wrote an algorithm to find it in a larger corpus.

That corpus is 400 plays printed between 1576 and 1642 that Martin Mueller gave us.

We reviewed the outputs & expanded our understanding of gradatio; or rather, we replaced our initial confident, simple formula with a more hesitant, complex understanding of how this figure works.

I’ll show you what I mean, but first let me put our process in more familiar terms.

Research happens something like this: something provokes a question; then we gather evidence to address it; and then that evidence provokes new questions.

In our case, the ‘something’ is gradatio: when we read Puttenham, and Shakespeare, and we see this figure. We then want to understand how it works, so we remember the formula (A > A, B > B, …) so that we can gather more examples (textual evidence).

Or if you’re us, you let the machine gather it for you, and it finds 114 instances in your 400 plays. That is 114 out of 8000+ anadiploses that occur in sequence.

We defined ‘anadiplosis’ as words recurring within 4 words of any clause-division; and ‘sequence’ as, well, that recurrence happening in 3 or more successive clauses.

I should say that this definition leaves us highly dependent on editorial decisions: we use punctuation to mark clause-breaks, and punctuation varies across editions.

Then we review this evidence, and make arguments that pose new questions.

Some of those 114 instances made us rethink the definition of gradatio. That brings me to our key question: are our definitions of figures adequate to their instantiations?

It turns out that Shakespeare and Puttenham and Sidney are just the head of the proverbial Leviathan. We got the whole Leviathan, and if you consider Thomas Hobbes’s frontispiece, the figures who make up that large, ungainly body made us rethink our definition of gradatio.

Okay, enough with the figural metaphors. Let’s move on to the figures themselves.

First, the mundane examples. This first one challenges all our conventional thinking about figures: it’s not beautiful, it’s hardly memorably, but it is (in an unsettling way) compelling. Who is this hoary speaker? What is he (I presume) proposing to lie close to?

The second example, to me, reads like Dr. Seuss. But I’m somehow less interested in knowing who these killers and their victims are.

We can answer these questions easily enough; I find a search in Google Books is pretty reliable. But we ought to stay attentive to more important questions: like, are these typical of how gradatio works?

The plain fact is that of the 114 gradatios our algorithm found in these 400 plays, most are like this. They meet the definition of the term, but they’re hardly beautiful. Some feel like they’re almost accidental.

Let’s consider some more extraordinary gradatios.

Here you have anadiploses_ in sequence (i.e. gradatio) that includes a small _incrementum: smile > laugh. I love this example, if only for its list of audience reactions to the range of performances the speakers (players, perhaps?) can put on. And for the jumble of those reactions following the list of genres (here in purple); I can’t find any correspondence.

Like I said a moment ago, the circumstances of this figure – its speaker; the immediate audience; the play’s genre, author, or year – are less important than how it’s working. We can explore that metadata if it helps us later; right now, we have a lot more data to get through, to categorize.

This one is a beautiful example of double-gradatio, thanks to a simile (set off in bold). A stone sits in a > ring that adorns the > hand; just as a king sits in a > court that is set apart from the > land. And it rhymes!

Our algorithm doesn’t classify the first repetition as an anadiplosis (of stones > stone) because it’s not within four words of the clause-break, but it did find the two anadiploses in the latter 2 lines. That suggests we may need to expand our matching-range beyond four words.

Here, the effect is even better. Structurally it’s a series of isocolons, or a series of short clauses of identical structure: (“From the A to the B”) – as indicated in bold.

Inset in that sequence of twelve terms, there’s gradatio of sons > thief > host. The algorithm also missed the first one, here, because of the intervening clause.

Now let’s look at some of the aberrations, to decide what implications they have for our evidence-gathering process. If any.

The question when we face an aberration is, an aberration from what? Does the problem lie with us, or with it? Is our definition of gradatio adequate to it? Or could it simply never qualify as this figure?

In what follows I use the term ‘unit’ to refer to a word, or a cluster of words, that is/are repeated across a clause-barrier (i.e. punctuation) to make anadiplosis.

Here in this slide, you have renowned > books of fame > brass. What I like about this aberration is it reveals how variable the units are, in two senses: part of speech (adjective > compound noun > noun); and number (a word > a 3-word subclause > a word).

It’s tempting to ‘tweak’ the script so we never get part-of-speech disagreements or number disagreements between units. But that would be wrong; it would commit the logical fallacy of excluding instances that challenge our premises (i.e. the standard definition of gradatio), rather than enlarging that definition.

As I said, is our definition of gradatio adequate to this example? We need, rather, to expand that definition to encompass it.

Here’s another problematic subset of the outputs to raise: cases like this where the units are identical, or nearly identical.

The first is not an ordinary list, but a series of anaphoras (repeated words at the beginnings of clauses); each clause is also an isocolon, each of equal length and similar structure.

Is it ‘really’ gradatio? Only if we’re exceedingly generous in our definition of the figure, I think. To borrow Puttenham’s metaphor, this feels more like marching in place than marching forward.

The second quotation is two three-part isocolons in series. Each two-word units that work like epizeuxis (which Richard Lanham defines as an emphatic “repetition of a word with no other words in between”) or palilogia (“repetition for vehemence or fullness”).

But is it gradatio? There’s no sense of successive movement, let alone growth.

To return to our core research question: are our definitions of figures adequate to their instantiations? Or do those instantiations, inversely, change the definitions?

We’ve concluded that gradatio may need as much splitting and subdividing as we find aberrations from the formula.

Some will be excluded entirely – as I said, I’d argue to exclude “a thousand widows, and a thousand wives,” (and so on) from the definition of gradatio. But others will expand our definition, and that’s the purpose of this augmented evidence-gathering.

Just as incrementum is a subset of gradatio (i.e. marching with a semantic climb, or decline), so too “upon ’em, upon ’em, they fly, they fly” might be an epizeuxal gradatio, or a new figure altogether.

So that’s where we’re headed: not toward a refined and distilled definition of gradatio, but toward a more complex, multi-variable one.

This paper’s main purpose has been to show you our outputs, particularly the more interesting examples that the algorithm found in these texts – and to ask you if these results change our settled definitions of gradatio.

So, some concluding questions: where might we go next?

• to other editions of our dramatic corpus?
• beyond drama? (imagine John Lyly’s Euphues!)
• beyond the early modern period?
• beyond this figure?
• and/or beyond the idea of searching for a given figure altogether, to searching for every discernible pattern of repetition, which we then classify?

I welcome your suggestions, and as well, I welcome any examples of gradatio that you encounter.