Resisting Perfection

I was born in a small city in Guatemala, whose cobblestone streets and colorful colonial architecture have long housed a healthy community of American ex-pats. Growing up there, Spanish and English came to me like twins: chronologically, one was handed to me before the other—Spanish. But English followed so quickly that both languages feel identically mine. My education was completely bilingual; conversations with friends moved between the two languages as easily as our backyard frogs leapt in and out of the water.

Perhaps this explains why I come up empty when people ask me why I write in English. To me, it never felt like a choice. I love the English language. I love its expansive vocabulary, its rhythms, its flexibility. I love its emphasis on clarity and efficiency. And yet, every once in a while, I feel acutely that English is not—not really—my mother tongue. That my command of it is only almost perfect.

I notice it when I stumble on an idiom, or when I see how effortlessly my native-English-speaking friends compose clever puns. And most recently, I’ve noticed it in the meandering, clause-laden sentences that populate my literary prose—sentences that have the cadence and inefficiency of my first-born Spanish, and that my computer’s new AI-powered proofreading software insists on cutting into chunks.

In those moments, the English language becomes more like a house: specifically, my grandparents’ house. A place where I’m always welcome and nourished, where I am very comfortable, but one that is ultimately not my own.

I recently used the AI-powered tool to proofread my 300-page novel manuscript. When I noticed the nifty “readability” feature, which scored my document according to how (in)accessible it was, my insecurity (or: perfectionism) got the best of me—I accepted suggestion after suggestion until my score improved. Aside from correcting typos and punctuation errors, the program often shortened my long sentences or split them into two, removing words it deemed unnecessary.

The program made so many suggestions that at some point I stopped and asked myself: was I a terrible writer? The more my readability score improved—which is to say, the more my sentences became “objectively” better—the more I wondered if this was still my prose.

And yet, I hesitated to reject the suggestions. The temptation to hand over choices to algorithms and large language models was stronger than ever. Could I impose my syntactical taste on a language that was not “mine”?

~

I’ve been thinking about all this since reading Cathy Park Hong’s essay “Bad English” in which she explains that because she grew up around “bad English,” she was “bad at English.” For years—until a professor encouraged her to use her own version of the language in her poetry—Hong’s manner of speaking was a source of shame.

She writes that her “bad English is a dying art because the Internet demands we write clear, succinct poems that stop us mid-scroll. If you want to truly understand someone’s accented English, you have to slow down and listen with your body. You have to train your ears and offer them your full attention. The Internet doesn’t have time for that.”

The Internet rewards legibility. It rewards accessibility, which in a globalized world requires a universal aesthetic. This is precisely the aesthetic the AI writing tool was imposing on my novel when it rewarded me with a better score every time I “simplified” a sentence.

After reading her essay, I attended a lecture by Hong in which she referenced French Caribbean theorist Édouard Glissant’s oft-quoted claim that colonized people “must reclaim the right to opacity.” According to his theory, developed in his Poetics of Relation and elsewhere, those who are colonized should not have to make themselves fully transparent (i.e. understandable) to their colonizers by adopting their language, (e.g. standardized English or French) in order to be respected by them. Rather, those who are colonized have a right to resist linguistic assimilation and to express themselves in their irreducible “mélange of language” to preserve their culture and identity amidst the flattening forces of colonial power.

Adopting this theory and its vocabulary into the age of internet surveillance and data extraction, Hong suggested that we might all cling to a degree of opacity to make it less easy for machines to exploit our work (and for governments to censor it). She granted us, her audience, permission to go as far as misspelling words or using cryptic, phonetic spellings to make our writing less accessible to Artificial Intelligence. If it also became less accessible to the reader, then so be it: let the reader work. Let the reader slow down. In other words: let her think.

Confronted by Hong’s call to resist the flattening forces of algorithms, it struck me how eager I’ve been to attain perfection. Though my childhood experience with English was vastly different from hers, I was still aware of being perceived as an outsider, and I always assumed that I did, in fact, have an obligation to make myself understood. To assimilate into the norms of the English language to earn recognition and status. And while that eagerness—combined with a writer’s fondness for control—has served me well as an academic, it had never occurred to me that it could be detrimental to me as a creative person. Using AI might make my sentences more legible, but it also makes them more like everybody else’s.

One day, for fun, I decided to test out what the AI writing tool would do to Nabokov’s prose. The Russian-born author had an enviable mastery of the English language and is regarded as one of the twentieth century’s greatest prose stylists. His prose is poetic and lush, and often also demanding (at least for me).

I typed two sentences from Lolita into my word processor:

I have often noticed that we are inclined to endow our friends with the stability of type that literary characters acquire in the reader’s mind. [...] Whatever evolution this or that popular character has gone through between the book covers, his fate is fixed in our minds, and, similarly, we expect our friends to follow this or that logical and conventional pattern we have fixed for them.

The proofreading software (ProWriting Aid) gave the paragraph a “readability grade” of 15. Its methodology is based on the Flesh-Kincaid readability scores developed in the 1970s for the U.S. Navy to ensure that its documents and manuals were understandable to its personnel. The scores were later adopted as a tool in general education to assess the accessibility of texts. They range from 0 to 18, with 0 being the easiest to read and 18 being readable at the post-graduate level.

A score of 15 for Nabokov’s paragraph suggests the text is understandable to a U.S. college student. Next to the score was a standard warning icon (an orange triangle with an exclamation mark in the center). A slider offered an additional visual cue: a score of 15 pushed the slider into the yellow zone, meaning that it was outside the software’s “target” range of readability.

When I accepted the software’s suggestions to rephrase the text, the readability grade “improved” from 15 to 10 (accessible for a 10th grade student). The slider moved into the green zone of readability and the yellow warning sign transformed into a green checkmark.

The revised lines read as follows:

I’ve often seen how we give our friends the same fixed traits that fictional characters get in our minds. [...] No matter how a beloved character evolves, their ending is set, and we expect friends to act predictably.

The sentences are undeniably clearer and more accessible. But much less poetic.

~

When I’m writing, I’m striving for clarity, but I’m also striving for sentences that delight me with their rhythm.

According to the spellchecker, my novel (which it analyzed in sections) had an alarmingly high proportion of long sentences. It underlined them in red. Sometimes they connected two ideas with conjunctions; sometimes they contained too many subordinate clauses, as though modeling my patterns of thought.

I cannot say for certain that this style is the result of my Spanish grammar pressing into my English syntax, bending it with its weight, but I wonder.

While English has incredible lexical flexibility—words can be morphed and combined quite freely—it has strict, simple rules for where nouns and verbs go in a sentence. On the other hand, Spanish has greater verb flexibility: conjugations express person, tense, and mood in a single word so that speakers have more freedom to rearrange word order in the sentence without losing grammatical clarity.

Spanish also relies extensively on the subjunctive mood to express doubt, desire, and hypotheticals. The subjunctive is almost always expressed in a subordinate clause, keeping the sentence’s meaning open until the end. This naturally produces longer, complex sentences. For example: Es posible que mañana llueva. “It is possible that tomorrow it will rain.”

Since English uses modal words such as “might,” “could,” or “would” to express the same idea (“It might rain tomorrow”), the subordinate clause can be collapsed into a single modal word within a shorter sentence.

Each language, therefore, has its own range of rhythms. When I started accepting the AI suggestions to make my sentences “simpler,” I noticed they lost some of the musicality I’d first heard in them. Same as with Nabokov’s paragraph. And I wonder if the reason is that my inner ear is always listening for both languages at once.

In the end, this exercise did not so much give me permission to be intentionally opaque as much as it gave me permission to accept a degree of inherent opacity in my English prose style. To accept some “imperfection” in my composition. And it’s not only because innovation results from variety and experimentation. It’s because soon, this imperfection might be what distinguishes work created by humans from that of machines.

I ended up reversing many of the corrections the AI-spellchecker had made to my novel.

My novel’s opening sentence remains, according to my British partner, too long.