Poemage: Visualizing the Sonic Topology of a Poem

Poemage is an interactive tool designed for the close reading of poetry by visualizing its sonic topology. It identifies complex rhyme sets and maps them as fluid paths across the text. This "flow" metaphor helps scholars explore linguistic patterns and ambiguity.

Transcript

Unknown: 0:00

Okay, let's unpack this. I think we often walk around with this, this mental map of the world, and on that map, we divide, I don't know, pretty much everything, into two very distinct, very separate buckets, right? These two big silos, exactly, silos. And in one of those, you have the cold, hard, rigorous logic of computer science. It's the world of the engineer. It binary zeros and ones. Totally it's true or false. And if your code is ambiguous, if it has any wiggle room, it crashes. It has to be rigid for it to work at all. And then in the completely opposite bucket, you know, usually on the other side of the university campus, you have poetry, right? The world of the artist, the world of the artist and poetry is it's fluid, it's emotional. It relies almost entirely on ambiguity, on the things that aren't explicitly said, yes, on the white space between the lines, on the invisible weight between words. It's all about feelings you can't quite pin down with a simple definition. It's a classic gap, isn't it, between the quantitative and the qualitative. The engineer wants to solve the problem. The poet wants to experience the problem exactly. And usually those two worlds, I mean, they don't really talk to each other, or if they do, it's like trying to mix oil and water, they just separate. Immediately. We just separate. But today, we are looking at what happens when you force those two worlds to collide. We're asking a really fundamental and honestly, a kind of wild question, which is, Can a machine help us understand the soul of a poem? And we have to be really careful here with our terms, right? When we say understand, we don't mean it in some kind of sci fi sense. No, no. We aren't talking about an AI that reads a sonnet and suddenly starts weeping, right? We're not talking about a robot gaining sentience or, you know, a soul of its own. Not at all. We're talking about visualization. Can a computer help us see the music inside the text? Can it show us the invisible structure? Yeah, the sonic topology. That's the term they use. The thing that makes a poem make us cry or laugh or just stop and think, because when you read a great poem, you feel something physical, right? A rhythm, a pulse. Can we actually map that? It is a question that really, really challenges our assumptions about what creativity is, but also what data analysis is for. So for our deep dive today, we're jumping into a fascinating research paper. It's titled poemage, visualizing the sonic topology of a poem, and it's by Nina McCurdy and her colleagues from the University of Utah. It's this incredible collaboration, Sonic topology. I want to just stick a pin in that phrase immediately, because it is so evocative. It doesn't sound like computer science. It sounds like we are mapping a physical terrain, you know, mountains and valleys and rivers, but the terrain is made entirely of sound waves. That is exactly what it is. And I think the key insight we're going to get to, and this is really the mission of this deep dive, is that this isn't about computers coming in to solve poetry, right? The researchers made that very, very clear. This isn't about an algorithm telling you this poem is 42% sad, or this poem is definitely about death. We've calculated it exactly, which is often what people fear, right? When they hear a term like digital humanities, they think it's about reducing art to a spreadsheet. Yeah, we analyzed Hamlet and determined he was just tired, precisely. That is the reductionist fear. But this project, this polish project, it's about creating a new lens, a new way of doing what literary scholars have always done, which is something called close reading. So it's a tool for humans. It's a tool for humans. It's about giving the human reader a kind of superpower, the ability to see patterns that the naked eye or even the naked ear might miss, because our brains just aren't wired to hold that much data at once exactly. You can't remember every single S sound in a 50 line palm and see how they all connect a computer can. So to set the stage for you listening, our goal today is to understand how this specific tool homage actually works. We want to get under the hood. We want to see how it helps poets and scholars spot things that they've been missing, maybe for centuries. And to do that, we have to navigate a few heavy concepts. We need to look at this clash they talk about, between distant reading and close reading, right? And we need to really dig into this central metaphor of a poem as a flow or a topology. And then we're going to look at some really surprising discoveries, like what happens when you visualize every single rhyme in a poem all at the same time, the beautiful mess, as they call it, the beautiful mess. I can't wait to get to that part. It sounds like my desk on a Monday morning. And finally, we'll look at how tools like this aren't just for analyzing old poems, but how they are actually helping poets write new ones through things like erasure and sentos. Okay, so let's start with the landscape. You mentioned this tension between distant reading and close reading. I feel like distant reading is a term I hear a lot, especially in digital humanities circles, but I'm not sure everyone listening knows exactly what. It implies it sounds a bit like reading a book from across the room. What are we actually talking about there? So distant reading is a term that was popularized by the literary scholar Franco Moretti, and it's basically exactly what it sounds like. You are zooming way, way out. You're not reading one book, you're not reading a book. You are reading a library, right? Or maybe even a whole database of libraries. Okay, so you're using computers to analyze these massive amounts of text, huge corpora of books or articles, to find patterns that a single human couldn't possibly live long enough to find exactly. Think about the Google Enneagram viewer you ever played with that? Oh, yeah, that tool where you can type in a word like freedom or science or love, and it shows you a line graph of how often that word was used in all the books they've stanned from like the year 1800 to today. That's a perfect example. Or Wordle before it was a game, yeah, the original Wordle was a visualization tool that made word clouds tag clouds right? Where the most common words of the biggest all of those are tools for spotting macro trends. You can see these big, slow shifts in language over decades, over centuries. You can make claims like, oh, look, the use of religious terminology dropped by 20% in the late 19th century. So it's sociological. It treats literature as data points in history. It's Big Data applied to books. Yes, it's great for historians. It's great for sociologists of literature, but, and this is the huge but, it is pretty terrible for understanding one specific poem, because a poem isn't just a bag of words, right? You can't just count the words in a poem and say you understand it. No, if I count the words in the Raven by Edgar Allan Poe. I might find that the word Nevermore appears a lot, but that number doesn't tell me why it's so chilling. It doesn't tell me why it's scary. Precisely. If you just throw a poem into a tag cloud, you lose the syntax, you lose the rhythm, you lose the order of the words, you lose the content, you lose the very thing that makes it a poem, the paper puts it really well. A poem isn't a pile of bricks. It's a house, and the arrangement, the architecture that matters more than the raw materials and distant reading just counts the bricks. It ignores the architecture completely. And that brings us to the other side of the coin, close reading, this is the traditional stuff. This is the traditional bread and butter of the humanities. If you were an English major, or really even just in a high school English class, this is what you did. You sit there with a single text, a sonnet, a short story, and you go over it with a fine toothed comb. You read it slowly. You read it out loud. You look at the rhyme scheme, you look at the meter, you look at the metaphors. You're asking, these deep questions you're looking for the affect how it makes you feel and why it makes you feel that way. Why did the author break the line here instead of there? Why did they choose this specific word Azure instead of just saying blue? Yes. And the problem from a computer's perspective is that close reading is incredibly complex. It's subjective. The paper describes it as looking at how syntax, rhyme and metaphor all interact across the temporal and spatial field of the text, temporal and spatial. That's interesting phrasing, I guess a poem takes time to read. That's the temporal part, right? It unfolds over time, but it also sits on the page in a specific shape. That's the spatial part, the visual Exactly. And trying to keep all those interactions in your head at once is it's really hard. Our working memory is limited, and for a long, long time, people just thought this kind of nuance was completely beyond the capability of computers. Computers are good at counting beans. They aren't good at understanding the flavor of the beans. They aren't good at vibes. That's a good way to put it, they're not good if I the paper actually calls the task of trying to support close reading a wicked problem. I love that term. It sounds like something from a fantasy novel or maybe something you'd hear in Boston. That's a wicked problem. I know, right? It's an actual design term, isn't it? It is, yeah, it comes from two design theorists, Horst Whittle and Melvin Webber back in the 1970s and they made a distinction between a wicked problem and a tame problem. A tame problem Sounds easy, not necessarily easy, but solvable. A tame problem might be incredibly complicated, like fixing a flat tire or even playing a game of chess. The rules are clear. The goal is clear. You know when you solved it, checkmate. Tire is inflated. Done exactly. There's an end point. You win or you lose. But a wicked problem, a wicked problem is different. A wicked problem is a problem where you don't really know what the solution is until you find it. Or even deeper than that, you don't even really know what the problem is until you start trying to solve it, like fixing the education system or solving world poverty. There's no game over. You win screen for that or interpreting a poem in close reading, a scholar doesn't usually start with a clear hypothesis, like I bet there are five rhymes in the first stanza. They start with an open mind. They explore. They wander around inside the text. Wander around the text until some kind of meaning or insight starts to emerge. It is process of discovery, not a process of verification. So how on earth do you build software for that if the user doesn't know what they're looking for, how do you code a button for it click here to find something interesting? That is the million dollar question, and that's why a lot of previous attempts largely failed or fell short. The paper mentions a few like poem viewer or myopia. And what were the issues with those? Well, they argue that some of them were too rule based. They tried to force the poem into a rigid structure that the computer understood, rather than one that the poet intended. They simplified it too much, or they made the human do all the work up front. Right? Some of them required the user to manually code all the data. First, you had to go to the poem yourself and tag every single rhyme, tell the computer this word rhymes with that word, and only then would the computer visualize it for you. But if you have to find all the rhymes yourself, the computer isn't really helping you discover anything new. Exactly, it's just drawing a picture of what you already know. It's a fancy highlighter. It's treating the poem as this static, dead object, just a thing sitting on a page. And that's not how poets see their work at all. No, they see poems as dynamic, as living things, as energetic systems. So the whole goal of poemage was to try and bridge this gap, to automate the boring part, the detection of all these Sonic patterns, so the user doesn't have to do the grunt work, but to do it in a way that still allows for and even celebrates the ambiguity that poets love. That's the key. Don't solve the ambiguity, visualize it. Let's talk about the people involved here, because this wasn't just a bunch of computer scientists sitting in a basement trying guess what poets guess what poets want? No, absolutely not. That would have been a disaster. This is a two year design study, a deep collaboration. So you had the computer scientists on one side, led by Nina McCurdy, and on the other you had poetry scholars and practitioners. So academics who study poetry but who are also poets themselves. They live and breathe this stuff and the dynamic there was, well, the paper describes the poet's attitude as skeptical enthusiasm, skeptical enthusiasm. I feel like that just captures the humanity of it perfectly, doesn't it? It does. Because if I'm a poet, my first thought is probably, you're gonna turn my beautiful, delicate art into a spreadsheet. And there was definitely resistance. The paper talks about this fear that the computer would kill the experience of the poem, that it would sanitize it, or oversimplify it, that the ghost in the machine would somehow drive out the ghost in the poem. Well put, but there was enthusiasm too. The other side of the coin, yes, there was this deep curiosity. They wondered if this technology could help them answer some really fundamental questions, questions they've been wrestling with their whole careers, like, what, like, what makes a poem? A poem? Is there an actual visible structure to the sound that I'm feeling but that I can't see? Can you show me the math behind the magic? So how did they actually build it? Did they just write the code for two years and then hand it over and say, Here's version 1.0 good luck. No, that would have failed. It was very iterative, very conversational. These are really interesting methods called technology probes. Technology probes, okay, what is a technology probe? Is that different from just a regular prototype? It is in a really subtle but important way, a prototype is usually a draft of the final product. You give it to a user and say, here's the app, please try to use it to do your job. The goal is to test usability. A technology probe is more like a toy or a specific, focused experiment. It's designed not to test usability, but to see how a user reacts to a new capability they've never had before. So they weren't trying to give them the finished product. They were just giving them these small experimental tools to play with exactly they would build a little feature, send it to the poets and just watch what happened. It's like tossing a new, weirdly shaped toy into a sandbox and seeing how the kids play with it. Do they build a castle with it? Do they throw it at each other? Do they ignore it completely? Right? It's about understanding the user's desires and mindset, not just their workflow. And what did they find from these probes? What was the big insight? They found something absolutely crucial. The poets did not want the computer to solve the poem. They were not interested in a button that said, analyze meaning or identify themes. They didn't want answers. They wanted to explore, they wanted to play. They wanted to see the connections for themselves, but they wanted to be the ones to draw the conclusions. That distinction is so important. The tool isn't the critic. The tool is the flashlight, exactly. And that insight, that single idea, drove the entire design of the system. They realized they needed to build a tool that didn't give answers, but that visualized the flow. Okay, so let's get into the sonic topology part. This is the core metaphor, because the way they describe the poem, the language they use, is really beautiful. They talk about it as a flow. Yes, this is the conceptual metaphor. It anchors the whole project, and it came directly from the poets. When you talk to poets about their work, they don't describe a poem as a grid or a database. They describe it as a living, fluid thing. So the team adopted that metaphor. Imagine the poem is a river. Okay, I'm picturing a river. The text is the water. The text is the water and the linguistic devices, the words, the sounds, the rhymes, the alliterations, they are moving through that space. They are the currents in the water. And the way those currents interact with each other creates the topology of the river. And if the poem is a river, then the rhymes are, well, eddies, whirlpools, Eddie's, whirlpools, currents. The paper uses this fantastic term, Sonic turbulence. Sonic turbulence. I love that. It sounds like a band name, it does, but it's a great description of where the energy is in a poem. If you have a place in the text where multiple patterns intersect, maybe you have a perfect rhyme and an alliteration and a slant rhyme all hitting on the same word are in the same line that creates high energy, that's turbulence, that's Sonic turbulence, and that's where the magic happens. It's not just about two words sounding like at the end of a line. It's about how those networks of sounds drag the reader's ear through the text. It creates a pull, a forward momentum, a pull. Exactly think about when you read a really satisfying rhyme. It snaps you forward. It connects the beginning of a thought to the end of a thought. It almost folds time in a way, it folds time. How so well you remember the sound from three seconds ago, from a previous line, and it links directly to the sound you're hearing right now. That linkage creates a structure that isn't physically on the page. It's in your ear and in your memory. Okay. So how do you turn that metaphor, turbulence, flow, folding time into data. Because the computer can't feel turbulence, it only knows code. You have to abstract it. You have to break it down into components. They built this topology out of three main concepts, okay, what's the first one? First you have what they call poem space. Poem space sounds like a sci fi dimension, Captain, we're entering poem space. It's simpler than it sounds, but it's absolutely vital. Poem space is just the 2d layout of the poem on the page. But here's the important part, the white space matters the empty parts of the page. Yes, the line breaks matter. Where a word physically sits on the page is part of its data, because in poetry, a line break is a breath, it's a pause, it's a musical rest. If you remove all the line breaks from a palm and just write it out as a normal paragraph, you completely destroy it. You destroy it exactly so palm space preserves that original geometry. That's the foundation. Okay, what's number two? Number two is rhyme sets. These are simply groups of words that are linked by some kind of sound. So the words cat, hat and bat are all in the poem. They form a set. Got it simple enough. And finally, number three, you have the path. This is the line, the visual connection that is drawn between the words in a set. So if cat is on line one, and that is on line four, the software draws a line between them, yes. And the way those lines, those paths, interact with each other is what gives you the topology they looked for specific types of interaction, like, what are the paths intersecting? Do they cross each other like swords? Are they overlapping, sharing the same route for a little while before splitting off. Do they merge together or diverge from single point? It's like looking at a subway map of the sounds in your head. That is a perfect analogy. You can see where all the lines converge at a busy station, like Times Square, and where a single line goes off into the suburbs all by itself, exactly. And poemage is the tool that draws that custom subway map for you automatically for any poem you give it. So let's visualize the machine itself. We're sitting in front of the computer. We've loaded up a poem into poem edge. What are we looking at on the screen? So the interface is divided into three main views, and they're arranged from left to right. It's designed to follow the natural workflow of a user exploring the poem. Okay, walk me through it. Left side first. On the far left, you have what they call the set view. This is basically your menu of ingredients. It lists all the different rhyme sets that the software has detected in the poem. It might say, rhyme Group A, cat, hat, bat, rhyme Group B, C, me. Tree. So it's a list of all the different sounds and the words that make them precisely, and you can browse through them. You can click on one to highlight it and see it in the other views. Okay, so that's the raw material. Then in the middle, in the middle, you have the poem view. This is the actual text of the poem formatted exactly how it would look on the printed page. The poem space. And it's interactive, very interactive. When you hover your mouse over a word here, it lights up, and so do all the other words. It's connected to the paths appear, and on the far right, the far right, is the path view. Now this is the most abstract part. It removes the text of the poem entirely, which shows you the nodes, which are the words represented as little dots, and the curves connecting them. It's the pure Sonic topology the subway map. Without the street names exactly. It's the pure structure. So you have the raw data on the left, the actual art in the middle, and the abstract structure on the right. It lets you toggle between reading the poem and seeing the poem's internal architecture. I want to go back to the set view for a second. The menu of ingredients, you said it lists the rhymes, but earlier you said it's not just counting cat and hat. How sophisticated is this rhyming engine? Because I know my phone's auto correct barely understands me half the time. It is incredibly sophisticated. I mean, there's really the core of the whole system. Pomage supports 24 different rhyme types. 24 Wow. Okay, I can probably name three perfect rhyme, slant rhyme, maybe internal rhyme, if I'm feeling fancy, right, most of us stop there, but pomage goes deep into the linguistics. So they have the classics, like perfect rhyme, which is what we all think of. But then they have slant rhyme, alliteration, which is the repetition of initial sounds, assonance, repetition of vowel sounds, consonants, repetition of consonant sounds. Okay, so those are the standard literary terms you learn in English class. Yes, but then it gets really technical. They have what they call phonetic matches. This is based on the actual physical way you speak the words. They use something called the CMU pronouncing dictionary to break every word down into its phones. And phonemes are the smallest individual units of sound in a language, right? Like the k sound in cat, exactly. But it goes even further. They analyze things like mouth placement. Mouth placement. What does that even mean? It's fascinating linguistics. It's about where in your mouth the sound is made. So for example, is your tongue in the same position when you say the P in pen and the B in boy, I guess so they both feel like they're at the front of my mouth with my lips. They are. They're both what linguists call bi labial plosives. You make the sound with your two lips. Pomage knows that so it can detect rhymes based on how your mouth physically moves, even if the sound isn't 100% identical. That is wild. So it's finding connections that are almost muscular. It's a kinesthetic rhyme. Your mouth is doing the rhyming. It is. It connects words that feel the same in the mouth when you speak them. But here is where it gets really, really interesting for me. They also track visual rhymes. Visual rhymes, you mean words that look like they should rhyme, but don't exactly. The technical term is I rhymes. Think of the words cough and bow, C, o, u g h and B, o, u g h, they both end in o, u g h. To your eye, they match perfectly, but to your ear, cough and bow are totally different sounds, so the software captures that mismatch. Yes, because for a human reader, that creates a specific kind of tension. Your eye expects a rhyme, and then your ear gets a surprise. That is a poetic device, and pomage visualizes that conflict between the eye and the ear and character clusters. What's that? That's even simpler. It's just words that share substrings of letters, like light and flight, or there and here, or even finding hidden words inside other words. The tool allows you to toggle all of these 24 types on or off. You can look at just the perfect rhymes or just the visual ones, where you can mix and match. Which brings us to the feature that seems to be the fan favorite among the poets, even though it seems to break all the rules of good design, the beautiful mess, the beautiful mess. This is literally a button that a user can click that turns on all the rhyme sets at the same time, every single connection, every perfect rhyme, every slant rhyme, every alliteration, every eye rhyme, all visualize at once. Now, from a data visualization standpoint, that sounds like an absolute nightmare. The first rule of data viz is clarity. You want to reduce clutter. You want minimal ink. As Edward tuft, who's like The Godfather of data visualization, would say, exactly in standard visualization design, this is a total failure. It's too much noise. It looks like a ball of yarn that a cat got into. It's an illegible scribble. But the poets loved it. They adored it. It was their favorite feature. Why? Why would anyone want a messy, unreadable graph? Because to them, the chaos is the point. The beautiful mess shows the sheer mind boggling complexity of the poem's music. It validates their feeling that a good poem isn't simple. It's a dense, deeply interconnected web of sound. It's like seeing the code of matrix for a second, you see the underlying structure of reality. That's it, and practically, it helps them spot outliers. If the entire screen is a chaotic mess of lines, but there's one single word sitting there with no lines connected to it at all, that stands out. It shows you the silence in the middle of all the noise, the one quiet spot in the orchestra, and that quiet spot is often profoundly meaningful. We'll get to an example of that in a minute, I promise. But first I want to talk about how they draw these lines, because you mentioned the flow metaphor. If you just draw straight, rigid lines between the words, it looks like a geometry problem. It doesn't look like a river, right? And this was a major design challenge. They had to. Solve. If you draw a straight line from a word on line one to a word on line five, you're going to slice right through the text on lines two, three and four, and that would be super confusing. You'd think the words in the middle were somehow part of that rhyme Exactly. It would create false connections. So the software does something really clever. It reroutes the line. It uses an algorithm to find a path for the line that curves through the white space between the words. So it avoids the text. It avoids the text. They use a specific mathematical tool called cubic Bezier curves. I don't know the math behind those, but I know what they look like from using design software. They're not straight. They look like like vines or strands of hair or river currents. Yes, they're organic curves. They meander. And the effect is that the connection lines themselves look fluid. They mimic the flow metaphor perfectly. It makes the data visualization look like it belongs in a poem. It respects the esthetic of the art form it's analyzing. And they can also change the shape of the poem itself, right? They call it deforming poem space. This is another really cool feature for playing around. You can view the poem in its original layout exactly as the poet intended. Or you can click a button and compress all the white space so every word and space is just one character wide, like it's been squished together, right? Or you can go the other way and view it as evenly spaced nodes, ignoring the original layout completely. Why would you want to do that? Why would you want to distort the poem to ask a specific question, does the poet's choice of line breaks actually change the sonic shape of the poem? Does the music look different if you ignore where the poet decided to hit the enter key so it lets you test the visual structure of the palm against the auditory structure of the sounds. It separates the visual rhythm from the auditory rhythm to see how they interact. It's another way of making the invisible visible. Okay, let's get concrete, because all this theory is fantastic, but I want to hear about the aha moments, the case studies where this tool actually showed a human being something they had never seen before. Okay, there are two really great examples in the paper that show the power of this. The first one involves a very, very famous poem. This is just to say by William Carlos Williams, oh, I know this one. This is the apology note poem. It goes something like, I have eaten the plums that were in the ice box and which you were probably saving for breakfast, forgive me, they were delicious, so sweet and so cold, that's the one. And it's a deceptively simple poem. There are no big, fancy words. It reads like a note that was literally left on a fridge. And in fact, it effectively was a note left on a fridge for his wife. It feels very casual, very plow spoken Exactly. But when the researchers loaded it into pomage and turned on the beautiful mess, when they visualized every single Sonic connection they could find, they discovered something startling. What happened? What did the screen look like? The screen lit up with lines. It was a dense network. Almost every single word was connected to some other word by some kind of sound, plums connected to probable through the plosive P sounds sweet, connected to eat, Ice Box, connected to breakfast. It was this incredibly rich web of sound hiding in plain sight. So it shows that even though it sounds like casual everyday speech, it's actually highly structured. It's dense with music. It's incredibly dense, except for one word, which word was it? The word you, you as in which you were probably saving for breakfast, the person he's apologizing to, yes, the recipient of the note, the victim of the plum theft. In the visualization, the word you was an island. It was sonically isolated. There were no lines touching it. It was cut off from the rest of the ball. Wow. So the speaker, the guy eating the plums, he's wrapped up in this sensual, delicious, sweet, cold experience. Everything is connected for him, the plums, the ice box, the taste, it's all one beautiful mess. And the you, the you, the person who lost their breakfast, is completely left out of that Sonic Web. They're sonically other. Exactly the tool revealed this deep emotional truth about the poem just by showing what didn't connect, it visually emphasizes the separation, the distance between the speaker and the listener, the selfish pleasure of the eater versus the cold isolation of the person they wronged. That is profound, and it's something you might feel when you read it, consciously or subconsciously, that little bit of a cold shoulder in the apology. But seeing it on the screen like that that makes it undeniable. It creates empirical evidence for a subjective feeling. That's the real power of the tool. It took a feeling, I feel like the you is isolated, and turned it into a visual fact. Look the data shows you is literally isolated on the graph. It's incredible. What was the second case study? This one was with the poem night by Louise Bogan and one of the collaborators on the project, a poetry scholar, had been studying this specific poem for years. She knew it inside and out. She'd written about it, she'd taught it so she was an absolute ex. On this one text, she didn't expect to find anything new. Probably not, yeah, you know, she probably thought, I know this poem better than I know my own family. But she was looking at the path view, that abstract view with just the nodes and curves, without the words, the pure subway map view, right? And she just noticed a visual pattern anomaly, a node anomaly, as they call it a node anomaly. Okay, basically, she just looked at the rhythm of the patterns. Most lines in the poem had say about four nodes, four Sonic beats, four points of connection. It was a regular, rhythmic visual pattern. Line after line, you see four dots. But then there was one line that had only two nodes, and they were squished very close together, so there was a visual gap, a break in the established rhythm, a huge visual gap in the pattern. And when she looked back at the poem view to see what that line of text actually was, she had this aha moment. She realized that this visual gap corresponded exactly to the turn of the poem, the turn. What's that in poetry, especially in sonnets, but in other forms too, the turn, or the Italian term, is the Volta, it's the crisis point. It's where the mood shifts dramatically, where the argument of the poem changes direction. It's the pivot. So the visual structure, the sonic topology, was perfectly mirroring the semantic tension of the poem. The break in the visual pattern was the break in the story, precisely. And she said, this is the amazing part. She said she had never seen that structural pivot so clearly, until the visualization pointed it out from the corner of her eye. From the corner of her eye, yes, it wasn't that the computer understood the meaning of the poem. The computer has no idea what a turn is. It was that the computer showed a pattern anomaly, and the human expert was the one who realized, oh, my gosh, that anomaly is meaningful. That's such a really important distinction. The computer isn't the critic. The computer is the flashlight. It just points at things. The human is still the one who has to see what's in the dark. Yes, the computer says, Hey, look at this weird spot over here. And the human expert says, Oh, that's not a weird spot. That's the Volta. That's the emotional shift of the entire poem, and that leads perfectly into the creative side of this, because what started as a tool for analysis quickly turned into a tool for creation. This is the part of the paper where they start talking about screw menutics. I have to say, I just love that word screw menutics. It sounds slightly naughty, but it's a legitimate academic term, right? It is. It was coined by a digital humanist named Steven Ramsey, and it's a brilliant play on the word hermeneutics, which is the fancy philosophical word for the theory and study of interpretation. Right? So, if hermeneutics is interpretation, then screwmeneutics is the idea of screwing around with a text in order to interpret it exactly. It's about being playful, using technology for what they call playful deformation. Instead of treating the text like this holy, sacred relic that you can't touch, you break it, you remix it, you mess it up to see how it works. So instead of reverently analyzing the text from a distance, you get your hands dirty, you hack it, you hack it. And poem turned out to be an incredible tool for screwing around, specifically for a form of poetry called erasure poetry. Okay, erasure poetry is where you take an existing page of text, like a page from a novel or a newspaper article, and you take a Sharpie and you black out most of the words, leaving just a few words visible, and those leftover words form a new poem, right? It's a subtractive art form. It's like sculpting. You're carving away the words you don't want to reveal the art that was hidden inside. And pomage made this easy, incredibly easy and in a totally new way, because pomage can isolate sounds, a user could select a rhyme, say all the words in a poem that have a sound and then tell the interface to erase everything else, so you're left with just the sonic skeleton of the original poem, just the words like Shush, Shadow, Rush. Yes, you're creating new art based on the sonic bones of the old art. And one of the collaborators, one of the poets, got so into this that she actually exhibited these poem erasures in an art gallery. The tool for studying poetry became a tool for creating poetry that is so cool it just completely blurs the line between the critic and the artist. And then there's the cento. Yeah, this story for me, this was the one that really blew my mind. The cento define that. For us, a sento is a patchwork poem. It's a poem made entirely of lines that are stolen from other sources. So it's like a musical remix, or like a ransom note made of words cut out of magazines, a literary remix. That's a perfect description. You might take a line from Shakespeare, a line from the Bible, a line from a cereal box, and you stitch them together to make something new. Okay, a patchwork poem. So one of the poets in the study, she made a cento using lines she took from a New York Times article about Pi Day. Okay, so she's taking journalistic prose, boring, factual sentences about mathematics and trying to turn them into a poem by rearranging them exactly. And then she used pomage to analyze her creation. She put her sento, this poem made of other people's words side by side with her own original poems that she had written from scratch. And what did she find? When she compared them, she found that the sento, the stolen poem, looked sonically identical to her original work. The Sonic topology was the same. Wait a minute. How was that even possible? She didn't write the words, she stole them from the New York Times. She didn't write the words, but she chose them. And what pomage revealed was that even when she was stealing lines, she was unconsciously selecting the ones that fit her own personal Sonic fingerprint, her Sonic fingerprint, yeah, without even realizing it, she was gravitating towards words and phrases that had the same sounds and rhythms as her own poetry. The paper mentioned she had a specific reliance on words like now source clouds. She was basically recreating her own sound using someone else's text. She couldn't help but sound like herself, even when she was literally using someone else's voice. She was a cover band that sounded exactly like the original. That is incredible. It implies that we all have a sonic style that is totally subconscious. We hear a certain rhythm in our own heads, and we just force the world to fit it. It raises this really provocative question, doesn't it, how much of our artistic style is a conscious, deliberate choice, and how much is just an unconscious attraction to certain vibrations? Are we just remixing the same few sounds over and over again because they feel good in our mouths and in her ears? And without the visualization, she never would have known that. She would have just thought, hey, this is a cool poem I made. She wouldn't have seen her own reflection in it. Exactly the visualization made the unconscious conscious. There's one more feature I want to touch on, which connects back to this idea of ambiguity. You know, computers hate ambiguity. They want a word to mean one thing and one thing only. But poets, poets live for ambiguity. It's their stock and trade. Take a simple word, like wind, W, I, N, D, is it wind like the air blowing through the trees, or is it wind, like you need to wind the clock? A computer usually has to guess one. It looks at the context and makes a decision right or wrong, right? But pomage embraces the confusion. It allows users to manually toggle between different pronunciations. But it also has this wild shuffle feature, a shuffle feature like on a music playlist, sort of it randomly shuffles through all the possible pronunciations of all the ambiguous words in the poem to show you an ensemble of possible flows. So it's not showing you one map, it's showing you all the possible maps exactly, maybe in one version of the poem, the word rhymes with mind. In another version, it rhymes with pinned. It lets you see the multiverse of the poem all the different ways it could be read. And that is something you can't really do in your head. You can't hold 10 different rhyme schemes in your mind at the same time, but the screen can. It allows you to see the poem not as a fixed, static object, but as this shimmering field of possibilities. It treats ambiguity as a feature, not a bug. It seems like the ultimate value of this project wasn't just the software itself. I mean, the tool is amazing, but it was the process. Absolutely. The paper talks about poemage as a disruptive technology, but not in the usual Silicon Valley sense of disrupting an industry or putting taxi drivers out of work. No, it disrupted the poet's brains. How so, what do they mean by that? Well, by having to explain poetry to computer scientists, the poets were forced to be incredibly precise about their own craft. You can't just tell a coder make the visualization look like flow, right? The coder is going to say, define flow. Give me the coordinates, give me the vector for the curve exactly. You have to define what flow is in computational terms, you have to define what a rhyme is. Is cough and Bower rhyme, yes or no. The computer forces you to make a binary choice about something that is fundamentally ambiguous. And that process forced them to move from this vague idea of, I find everything interesting to a very specific goal. I need to visualize the topology. It sharpened their own critical thinking. So the machine, in a way, taught the humans how to be better humans, or at least better more precise scholars, I think so, it validated that poetry has a geometry, a real, mapable structure. It proved that the soul of the poem isn't just some ineffable magic. It has a structure. It has a topology that can be mapped and seen. And that brings us right back to our mission. For this deep dive. We wanted to see if a machine could help us understand the soul of a poem. And I think the answer is yes, but not by replacing the reader, no, by giving the reader a telescope or an x ray machine. An X ray machine, that's perfect. It shows you the skeleton under the skin, exactly. The tool we build to analyze the art ends up becoming a tool for creating new art. The cycle continues. So here's my final thought to leave you with. And this is something the paper doesn't explicitly say, but it's been nagging at me throughout this whole discussion. Go for it. Okay, so we've seen that pomage can reveal these hidden Sonic. Structures that subconsciously manipulate our emotions, like the isolation of the word you in that William Carlos Williams poem. We might not notice it consciously, but our brain feels it. It creates that Sonic turbulence that drags us through the text and makes us feel a certain way, right? It triggers an emotional response through pure geometry. So if we can visualize that topology, if we can map the perfect emotional structure of a sad poem or a joyful one, could we teach a machine to write a perfect poem, not by understanding the meaning of the words, but just by following the geometry, by creating the perfect Sonic turbulence. That is the ultimate question, isn't it, if you can successfully map the soul, can you then manufacture it? If we know the precise shape of sadness, can we just 3d print a sad poem that is guaranteed to work on a human brain? It's a little scary, but it's also absolutely fascinating. It's something to think about next time you're reading a poem or even scrolling through your news feed, is that emotional reaction you're having to the reaction you're having to the words? Is it real, or is it just good topology? I will definitely be thinking about that for a while. Thanks for diving in with us on this one. Always a pleasure. Bye.