Beneath the Language

Beneath the Language

by Ken Liu

Originally published in On the Premises, July 2007.

 

 

After two weeks of last-minute cancellations and bad cell phone reception, I finally caught up with Cal Solulu in Shanghai. Just under six feet, Solulu was a barrel-chested man in his early thirties, with short legs and long arms that gave him the air of an orangutan.

He was speaking at a government-sponsored conference for the China Writers Association. CWA, a curious organization with no real Western equivalent, is best described as a government ministry cum writer’s guild dedicated to the quixotic task of preserving Marxism-Leninism orthodoxy while cultivating the freedom of a new Chinese literary voice. It seemed a strange audience for a man whose whole message was that ideology and literary taste alike were mere illusions.

Solulu isn’t bothered by the irony of his presence at the CWA.

“The young Chinese writers are more receptive to my message than anybody else. They way they explain it to me, the Chinese feel that they don’t have much of a native contemporary culture to speak of. In the last hundred years, they haven’t been able to come up with anything, a picture, a movie, a musical movement, that isn’t derivative of something the West — or even worse for them, Japan — invented. They are humiliated, but they are also less invested in the idea of creativity than we are in the West. This makes them more willing to listen to radical ideas.”

We were sitting in the lounge of the Manhattan Palace, one of Shanghai’s newest and most expensive Chinese-owned and operated hotels, an eighty-story concrete-and-glass tower erected in only 181 days in the perpetual frenzy of construction that defines modern Shanghai. As if illustrating Solulu’s point, there was nothing original or Chinese about the place: the muzak was a bad copy of jazz, the color scheme a bad copy of the Shanghai JC Mandarin, the furniture gaudy imitations of five-year old Japanese designs, and the drinks menu full of translation errors. The staff stood around aimlessly, green actors on a new set with no script. The most coherently Chinese thing about the place was that it felt like a copy of a copy of a copy.

But if Solulu is right, the Chinese don’t have to copy. Creativity, according to Solulu, is an algorithm that can be mastered and applied relentlessly, much like the way the Chinese have already diligently mastered the art of cheap and efficient manufacturing.

#

Solulu isn’t ashamed to admit that he began as a failed poet. He’s not ashamed because he thinks poets have no understanding of what they are doing.

“Most successful poets are like idiot savants,” he said. “They can no more explain how they write poetry than a fish can explain how it swims. If they are successful, they think they’re geniuses. If they are not, they think the critics are philistines. They’re ignorant about the mechanism of their art.”

As a boy, Solulu had poured his heart into “notebook after notebook” of poetry. He wrote letters to Ashbery, Merwin, and other names he found in the Poetry section of the local bookstore, asking them to be his mentors. (Only Margaret Atwood ever wrote back, though Solulu wouldn’t tell me what she said, only that it was “what you’d expect from her poems.”) He did all the “right” things: he read, he wrote, he collected rejection slips. But he never got published anywhere, not even in his college literary journal.

“It was embarrassing. It’s very difficult to maintain the argument that you are simply not ‘understood’ when your classmates are reading all the same books as you and share all the same references as you and they still don’t like your poems. Very disheartening.”

Then, sometime in the summer of his Junior year, he had what he called his “crisis moment.” He received his one hundredth rejection slip.

“I showed the rejection slip to my girlfriend at the time. To make me feel better, she said she’d read my submissions and give me comments. But the next day I could tell, before she even said a word, that she couldn’t even finish them. It was devastating. I couldn’t bear to write another word until I understood what it was that made good poetry good.”

#

I should stress here that Solulu isn’t interested in the theoretical debates that have raged among intellectuals from Plato’s guests to today’s literary critics and academics over what separates good poetry from bad, and how these judgments could and should be made. To Solulu these debates are nonsense. He knows what “good” poetry is — it’s simply what has proved popular over time — or, as he puts it, good poetry is poetry that “resonates with the greatest number of souls over time.” The much more difficult and interesting question for him is what makes one poem resonate more than another. He’s convinced that no one really understands that secret, especially not the successful poets.

“The things poets say about their craft and their process have almost nothing to do with what separates good poems from bad ones. In fact, poets are like the money managers on Wall Street” — and here I saw flashes of Wallace Stevens’ most unpoetic mien — “who go around with their stock charts and their P/E ratios, convinced that they know how to pick great stocks. And year after year, the vast majority of them fall short of the index, and neither the lucky ones who happen to win that year nor the unlucky ones who lose all their clients’ money understand what happened. But that doesn’t prevent them from babbling to the financial reporters that of course it was because it was the first Monday in March that the Dow moved up 20 points that morning. The successful poet and stock picker alike are good at only one thing: manufacturing ad hoc justifications for their good fortune.”

“That sounds a bit bitter, coming from a failed poet,” I said.

“No, no, no,” he said, laughing but emphasizing each “no” with two shakes of his head, that wild mane of unkempt white hair flying about. “I was bitter, but not any more. Now I have the answer.”

#

Solulu found his answer, as so many do in our age, in the heart of a machine.

The machine in this case is a spaghetti mess of code (he’s too embarrassed to show it to me) that Solulu wrote over five years. When this program is fed the sound of a poem written in modern English — a limitation due to the corpus used to train the program rather than an inherent feature of the algorithm — being read aloud, it will predict, with much greater than 99% accuracy, whether the poem will be sufficiently popular to be included in more than three poetry anthologies today.

“So the first thing you have to do is to define the problem: how will I objectively know when I’ve found a way to tell the good poems apart from the bad poems? The only legitimate measure is meme survival. In the short term there may be all kinds of noise that cause a poem to be popular: the author is a celebrity; the topic is politically relevant; Venus and Mars are in opposition; whatever. But in the long term, fads fade out, and only good poems survive.”

Obsessed with the problem, he went to flea markets and yard sales and bought up hundreds of poetry magazines published in the last century. First, he read through them, trying to see if he could predict which poems would survive thirty, forty years down the road. Then he diligently applied theories of aesthetics, from Aristotle to Addison, from Dostoyevsky to Vendler. It was often a challenge just to figure out what the critic even meant. In any event, these theories turned out to be no better at predicting success than random chance.

About four months into the project, Solulu hit a dead end.

“All these theories of poetics and aesthetics and the history of criticism and ideas were swirling around in my head, and I was trying out anything, everything, that could help me figure out the pattern of which poems survived and which ones didn’t. But one day I sat up, read a bunch of poems, and realized I hated all of them -no, more like I couldn’t tell whether I liked any of them. It was as if a musician suddenly woke up and realized he was tone deaf, or if you woke up and realized that you couldn’t tell the pretty girls apart from the plain ones. I was beauty blind.”

This beauty blindness lasted for several weeks, during which time he seriously considered suicide.

He was finally saved by our least poetic modern medium, television. He was at home, mindlessly letting the glow of the tube wash over him. The program was a documentary on the vanishing oral traditions of “some nomadic camel herding people out there in Mongolia.”

“There was this scene, which lasted maybe five or six minutes. It was just this old man reciting something. The subtitles were in white text, and I couldn’t read it against the white dunes on the bottom of the screen. I had no idea what he was saying, but it was beautiful. I could feel my heart beating in time with the rhythm of his speech. I could hear the meter, the rhymes, the musicality of the tones, even though I couldn’t understand what he was saying. I was enraptured, transported, in a way that I had never been before by poetry in English. I literally could hear the blood rushing in my ears; I was so excited. There it was, just like that, I could hear beauty again.”

The experience gave Solulu a series of insights into the problem. First, it had to be about beauty, not theory. Second, language was a distraction. Finally, the “good poem detector” had to be a machine.

#

“The core problem is that we don’t trust first impressions.” He waved his chopsticks in the air. “You see a painting, and you know whether you like it or not within a tenth of a second. But you don’t trust that feeling. You’ve been taught that art must be ‘appreciated.’ You think you have to understand the painting before you can really say if you like it. You think you have to read that little pompous card written by the museum curator, which gives you the painting’s title in French and its English translation, a one-sentence biography of the painter, and a few pieces of useless trivia. And if you’ve been to a good college, well, then I really pity you. Then you think you have to pay ten dollars for the little electronic wand with the tinny speaker so that you can punch in the number under each painting and listen to the little speech from the curator about why the painting was good. And at the end of it all, you think you’ve learned something. You think you now know why you should like the painting.

“But that’s just garbage. All of it. Nothing mattered after that first tenth of a second. Everything afterwards was just noise, epiphenomena, froth on the sea. But as a society, we have beaten into ourselves the idea that the noise and the epiphenomena are the real deal, that the justification is more important than the snap judgment. We know almost immediately if we prefer this painting to that painting and if we like this poem better than that poem. But we talk ourselves into a muddled confusion when we try to explain that preference. The intellect gets in the way.”

That was insight number one. Step two?

“There’s this misunderstanding that because poetry is composed with words, language and meaning and reasoning had to be the most important part of the appeal of a poem. But poetry is a lot more like music.”

Solulu got his inspiration from a company called Platinum Blue, which was in the business of predicting which songs by unknown artists would be hits. Their technology ignored everything except the pure mathematical shape of the songs.

The idea struck a chord with Solulu.

“When I was listening to the old Mongolian herdsman I was experiencing poetry in a raw form that we seldom do any more. Mostly we read poetry silently, and we analyze poems as patterns of words on a page. But the written language is a distraction. Poetry is about the spoken word, and the spoken word is just syllable and sound and fury. Listen to this:

“Do you know it?”

I shook my head. It sounded German, or maybe Klingon.

“But I bet that you could hear that it was poetry. And I bet you had an opinion of whether you liked it.”

He was right. I knew it was poetry. I could hear the hard pounding beats, like sword against shield. And I wanted to hear more.

“That was from the Battle of Maldon, in the original Anglo-Saxon. It’s related to modern English, but I don’t think you understood a word of it. You know nothing of its imagery, metaphor, its history, its place in the canon of English poetry, but you knew you liked it. In fact, for most people who had to take Anglo-Saxon in school, these two lines are probably the only ones they will remember. The sound of good poetry is that powerful. It’s like the chorus of a catchy song. It hooks you and pulls you in, and you don’t care that the lyrics are nonsense.

“I could try the same experiment with you with a popular old poem in Chinese, in Japanese, in Arabic, and it would be the same every time. If it’s read by someone with even the smallest bit of musical talent, you’ll be able to tell that it’s poetry, and you’ll be able to tell me whether you like it or not, just like that.”

Solulu doesn’t mean that you don’t need to understand the language to understand the poem. Of course you have to know what the words mean to understand the story being told, the sentiment being expressed, or the revolutionary message that’s embedded. But he makes the radical claim that understanding is, in fact, a distraction for determining whether you like a poem. The sound of a poem is always there, like the bass line of a song, even if you are just reading the poem to yourself, silently. The musical shape of a pop song determines our emotional response, even though we may end up thinking that we like it because of its clever lyrics. In the same way, the sound of a poem is what really moves us, even if we think it’s about the clever words.

#

At first glance, Solulu’s program cannot be used, directly, to generate a good poem. This is a consequence of the way the program evolved as a neural network. Solulu refined it over time by reading it poetry from various period publications. Each poem’s mathematical signature was a bit of input. He then found out which of those poems survived in multiple modern anthologies used in undergraduate literature classes, a reasonable proxy for success. These survivors were then marked and fed to the program again, and the program was asked to discover patterns in their signatures that distinguished them from the poems that did not survive. It did so simply by brute force trial and error until it generated some complex mathematical function that mapped all the inputs to the right outputs, but the function is really a black box. You can ask it questions and get back an answer, but even Solulu has no idea how it really “works.”

To test the program, you simply feed it a new poem, and check the program’s prediction against real life.

For Solulu’s theory to be of use, it’s okay for the program to generate false positives (i.e., poems that the program predicted would succeed but in fact did not), but it’s imperative that it generate few, if any, false negatives (i.e., poems that the program predicted would not succeed but in fact did). This is because Solulu’s theory presupposes that no inherently bad poem can ever succeed in the long run, but it is possible for an inherently good poem to never be given a fair chance to spread (think of Emily Dickinson if she had simply locked all her poems away).

A consequence of this asymmetry is that if you run it over old poetry magazines you may well end up discovering overlooked good poets. It’s not hard to imagine that someone who’s very talented but also a bit of a recluse may publish only a few times in an obscure journal and never gain more than a few readers. If none of them happens to be an important editor or prominent critic, the poet would end up languishing in obscurity for eternity. Using his program, Solulu has already managed to discover two poets from the eighties who, he assures me, “are as good as anything on Harvard syllabuses today.” (He’s currently editing collections of their poems for Harvard University Press.)

If you believe past performance is a good predictor of future success, Solulu’s algorithm undeniably works. He has run it on English poems as far back as the Renaissance, and the false negative rate is lower than one hundredth of one percent.

#

Solulu has detractors.

I met with Len Keene, the Stuart B. Dunbar Professor of English and American Language and Literature at Harvard University, a few weeks after I came back from Shanghai.

“Cal was a student of mine,” Keene told me. “People find that surprising. Maybe he’s proof that I’m a terrible teacher.” He laughed.

Keene thinks Solulu is part of a general trend in today’s academic world in which reductionism — whether in the form of E.O. Wilson’s sociobiology, Noam Chomsky’s universal grammar, or Cal Solulu’s “good poetry detector” — is attacking the realm of culture, taste, society, the traditional enclave of the messy humanities.

Keene can live with the first two steps in Solulu’s logic (although he isn’t exactly happy with them). Of course, under the logic of his first step, if good poetry is to be evaluated by immediate impressions of beauty rather than theories of worth, everyone is a suitable judge of poetry except the academics. But Keene is used to this kind of anti-ivory tower populism, and besides, the academy has proved itself quite adaptable by inventing the field of cultural studies to co-opt its critics. Under the logic of Solulu’s second step, it is the sound of poetry, rather than its sense, that determines the bulk of our impressions of beauty in a poem before the “sense” and “meaning” come in to confuse us. Keene finds these ideas more quaint than troubling, since the analysis of poetry as predominantly a form of music hasn’t been taken seriously, by anyone, in a long time.

It’s the third step in Solulu’s reasoning that really irks Keene.

“Fundamentally, what we find beautiful must be explained at the level of neurobiology,” Solulu wrote in a public, online debate with Keene last June. “And I am convinced that our individual neurobiology isn’t so different from each other. We are all members of the same species, sharing pretty much the same genes, and at a deep level, our brains will likely react similarly to similar stimuli. If I find a particular pattern of colors pleasing, it is because the light from that pattern of colors causes my brain chemistry to change in a certain way. But you, as a fellow member of the human species, likely has a brain very similar to mine. That same pattern of light is thus likely to cause similar changes in your brain chemistry, and you are also likely to find that pattern of colors pleasing.

“We do not need to understand exactly which patterns of vowels and consonants will produce these pleasurable states or how or why. It is enough that we will feel this way at a visceral level, at a level that is beneath and above thought.

“But thoughts inevitably come in, and our brain chemistry is altered as we try to outdo each other in cleverness, in coming up with reasons and explanations for that fleeting first impression upon which so much depends. So much then becomes muddled and confused. The epiphenomena froth over and hide the deep currents and waves of real beauty.

“We need a way to access the non-linguistic, non-analytical core of our common heritage as homo sapiens, the human animal, without all the noise and distractions. But it’s too late to get back to that Zen-like state in ourselves. The best we can hope for is to model that core of ourselves, to recreate the beauty-detecting faculty in a machine.”

When I asked Keene about these comments, he was silent for a long time. He was one of these people who, instead of filling silence with useless chatter, are comfortable with it. Remaining in my chair across from his desk, I glanced over the spines of the books in his office. A handsome copy of the Douay Bible sat on his shelf, dwarfing the collection of C.S. Lewis paperbacks next to it. I looked at Keene, pointed at the Bible, and raised my eyebrow. He smiled and invited me to take a walk with him.

“I am not qualified to criticize Cal’s science, though I think he is misusing science. My disagreement with him is really one of philosophy. His arguments have an unhealthy attraction for undergraduate students, the way all elegant, beautiful, and astonishingly bad ideas do. The assault on humanism from this relentless drive to reductionism has been ongoing for more than two centuries, but there is something particularly cold about Cal’s assault. He has declared that we should turn over our aesthetic judgments to a machine. He is arguing, in essence, that not only can machines play chess better, they can also better judge what is beautiful and worth reading. In making an argument that is premised on reason he has discarded reason as an empty shell. It is a meaningless conclusion, a view of the soul as a void. And if I may misquote Charles Darwin: there is no grandeur in this view of life. It is ugly, and I despise it.”

Over lunch Keene, who was Solulu’s academic advisor in college, told me a story from Solulu’s time as a student. Solulu had broken up with his girlfriend after she somehow insulted him. That June, he ran against her for class marshal, even though he had never before expressed any interest in the position. He campaigned hard against her, but lost in the end. And every year since then, he would write a letter to the alumni magazine arguing that the class marshal system was outdated and should be abolished.

I didn’t understand why he was telling me that story.

After lunch, as I carefully counted out my half of the bill, Keene added, quietly but purposefully, “Cal holds grudges. He can hold a grudge against something that he loves. I don’t like to say this about him, because it verges on an ad hominem. But I think if he can’t be the best at something, he has to show that it isn’t worth doing.”

#

A few weeks earlier in Shanghai, Solulu had been adamant that his program was an affirmation of art. “If you look at what this program is saying, it’s an incredible vote of confidence in the value of art. The program tells you that from the 1500s to now, despite all the revolutions, all the social changes, all the ebb and flow of ideas and wealth and class and race and colonialism et cetera et cetera, aesthetic judgment is remarkably consistent. Despite all our fears that ‘art’ is nothing more than the representation of power and privilege in a particular moment of time and a particular place, it turns out that there really is something universal about it after all. A poem that was pleasing to a sixteenth-century nobleman is just as pleasing to a twentieth century middle-class co-ed — once we learn to keep all the intellectual masturbation out of it. There’s something timeless and beautiful about each of these poems that have survived: whether it’s ‘To His Coy Mistress’ or ‘The Munich Mannequins.’ I don’t know how you can see that as a bad thing.”

He has his fans. Efforts are underway by others to distill a similar algorithm for paintings, novels, and other forms of art. Others are less sanguine. A group of English professors have tentatively reached out to their engineering colleagues and asked for volunteers to show that Solulu’s methods are flawed. So far there have been no takers.

#

I had said earlier that Solulu’s program can’t be used to generate good poetry, but that isn’t quite right. A black box “good poetry detector” can be used to generate good poems in a process akin to natural selection. If you have a lot of budding poets who are producing a lot of experimental work, you can feed their productions through the detector and continue to cultivate those that show promise. Let these survivors produce more works and rinse and repeat until you are down to just a few of the best poets, and now you have a selection process immune from nepotism, discrimination, privilege, and class — a perfect way to develop truly great art.

This is what made him so interesting to the CWA. Solulu’s algorithm promised the hungry, proud young Chinese writers a solution to their cultural deficit. If the machine could be broadened beyond poetry, it would show the way to a scientific method for the mass production of good, successful popular culture. Because Solulu’s algorithm was premised on universal and language-agnostic assumptions, the Chinese could imagine that if the algorithm was fed with Chinese works, the successes would have universal appeal while still being undeniably Chinese.

But the method of production for this hypothetical success would not bear any resemblance to our understanding of “creativity” and “art.” Writing poetry would be like working in a factory. One can understand Keene’s rage, which is really more like despair.

The prospect of a relentless army of Chinese artists producing variations for his algorithm to sift through until a national champion is produced seemed to me a dystopia. Solulu does not see it that way. I am not sure which of us is experiencing epiphenomena.

#

At the end of my interview, I turned off the voice recorder and thanked Solulu for his time. He was packing up his laptop and preparing to go back to his hotel room for the night.

On a whim, I asked him if he was happy with his discovery. He was quiet for a while, but he didn’t like the silence.

“I began this, you know, because I wanted to understand what made good poetry good. I wanted to know the secret so that I could write good poetry.” He would not make eye contact with me.

“And now I do know what makes poetry good — or at least my program does. If I really wanted to, I could just feed drafts of my poem to it, and see if each new draft gets better or worse. Or I could just read it all my old notebooks. There’s bound to be a good poem somewhere in there.

“But I have no interest in doing any of that. I guess there’s a part of me that feels that would be cheating. I know it’s illogical, but that’s how I feel. To be honest, I have no interest in reading poetry these days, and even less interest in writing. It’s not that I’m beauty blind again. I can tell when a poem is good. But as soon as I get that impression that it’s good, I think about the sounds, the pattern underneath, the thing that pulled me in before I was even aware of it, and then I feel that I’m being manipulated, and I have to stop. I suppose it’s a bit like how fashion photographers can’t stand to pick up Vogue because they know what goes into the pictures. Or maybe it’s more like loving magic tricks as a kid and then being shown how it’s done. I don’t know. I can’t explain it really well.”

He continued to look away. He was fidgeting with his laptop bag, opening it, closing it, opening it again. It reminded me of a child who, peeking behind a curtain for his sitter and finding no one there, incredulously peeks behind it again and again, as though by sheer effort of will to conjure her out of thin air.

### End ###