Writing a More Perfect Sentence

I subscribe to Ingrid Burrington's Perfect Sentences delightful newsletter that brings brilliantly crafted lines from books, articles, news stories, or even tweets. They're presented without context and just stand alone letting you chew on their craftily prepared morsels bereft from their original piece. Sometimes, Ingrid gives a hint of context but the title of the piece is often enough to appreciate the sentence better.

Reading these sentences week after week made me wish one of my sentences would be included and I wish I could write so adroitly yet succinctly. I was reading aspirationally. My default writing is to ramble on or overexplain hence brevity comes difficult to me. But I thought, why not use Burrington's newsletter to better understand why I like some of the sentences and better still, learn from that knowledge to improve my own writing.

Always a quantitative scientist, I have done very little qualitative analysis much less understand the nuances of language and rhetoric. This felt like exactly the kind of thing AI should be good at. Before you recoil and close the tab, let me explain.

I don't intend to ask Claude or ChatGPT to review the newsletters contents and then edit my writing in line with its style. Not because it's unethical, though it edges close, but because it would be useless: these sentences reflect Burrington's taste, not mine. Stripped of context, the forced imitation would come out deranged, like a strap-on in a Shakespearean play.

I wanted to learn more about what made those sentences so brilliant and if you could even classify them by type and context to use how to use them effectively in your own writing for your own words. Basically, it would be like an writing and rhetoric course. Until I'm retired and can enroll in community college class on the topic, I've to make do with what AI provides. So fasten your seat belts and follow me on my journey to writing better sentences.

First, I scraped the archive from the newsletter's archive page to a folder in my Obsidian vault. She has sent nearly 200 editions so far that contain 2,083 sentences in total. Initially, I thought it would make sense to contextualize the sentences with their sources so that they make more sense but realized that apart from the title of the essay, I didn't want too much context since I wanted the sentences to stand on their own. Second, since this was Burrington's taste, I had to marked the sentences I wanted to keep. This took some iteration on the selection because I was selecting far too many of them. So I went back and reviewed the sentences from the perspective of "I wish I had written that".

Here, the AI was helpful. It helped to transform the text from the text files in Obsidian to a CSV spreadsheet with a sentence per row with a single column of whether I selected it or not. Since it's a pain to review text, even if there are single sentences, within Excel and I'm sure that's how we did it in the good 'ol days, this would take me forever. Claude helped me build a local HTML tool: one sentence at a time, keep or skip with a keystroke, writing straight to the CSV, with resume & shuffle features built in for whole-archive coverage. It also allowed me to add a re-skim mode to tighten early loose picks to the "I wish I had written that" strict bar.

I ended with a clean gold set of 106 sentences under the strict bar at an 18 to 19 percent rate from the overall dataset. Claude mentioned declining marginal benefits for doing more so I stopped there and then let AI do its magic. Another script measured the shape of my 'keeps' against the full archive as a control and found that my taste is in fact plain, not ornate, contrary to what I had thought. Every showy marker was under-represented. No questions, no exclamations, no parentheticals, almost no semicolons. I seemed to favor clean, medium-length, declarative sentences. Or as Claude said, "Your taste is not about punctuation or subject. It is about what a plain sentence does with an idea."

I asked Claude to then classify the sentences into recurring thematic element of rhetoric or writing style. I don't know any so asked it to use the domain knowledge in its training to categorize. After some digital humming, Claude spat out six categories, like a literary vending machine: Yoked analogy, flat aphorism, deadpan absurdity, lyric image, compressed power critique, and the turn.

1. The yoked analogy. Your signature. A comparison that bolts together two domains that do not belong together, and the distance is the point. This is the single most common move in your set.

• San Francisco as "a very pretty, well-upholstered car with a growing colony of possums under the hood."
• "The Dunning-Kruger effect is the hemophilia of dynastic capitalism."
• "matt yglesias looks like if a newspaper could take a shit."
• the Supreme Court compared to how Batman thinks of Superman.

2. The flat aphorism. A portable general truth, stated as plain fact, present tense, no hedging. Quotable on its own.

• "What looks like discipline is often just impatience dressed up as process."
• "Sooner or later, we trust the disappointment more than the dream."
• "No one wants to be food, but it feels somehow more demeaning to be gum."

3. Deadpan absurdity. A grim or ridiculous thing delivered in a flat, composed register. The calm tone is the joke.

• "Grazer could not be immediately reached for comment, as she is a wild animal living in a remote national park in Alaska."
• the museum of your worst day, attended "with a bunch of tourists from Denmark."
• "just because the clown car goes very fast does not mean it isn't full of clowns."

4. The lyric image. Concrete sensory description, usually place, nature, or body, that lands as poetry without reaching for it.

• "the sky over the palm trees is the color of dirty water from a janitor's bucket."
• "Sun-faded pumpjacks languorously sip oil up from underground."
• "It is a soundless immensity with mountains in it."

5. Compressed power critique. A hard claim about institutions, war, money, or media, packed tight.

• America as "a complicated system of PR and political logistics" fighting over "the world's most deadly compilation of Excel spreadsheets."
• "grief is also america's biggest export."
• "those fluent in the good grammar of civility have deployed decorum to mask agendas of cruelty."

6. The turn. A setup that pivots on a single distinction, so the second half reframes the first.

• "telling the truth was way different from finding the truth."
• "It does not set out to embarrass anybody, but if history embarrasses someone, so be it."

Another script coded each 'keep' with a primary and optional secondary classification. Yoked was my top category at 30 percent. Eighty-five of 106 keeps carried two craft-moves.. My signature pairings were deadpan plus yoked, lyric plus yoked, and aphorism plus power. These are all qualitative analyses techniques that you usually train undergraduate or graduate students and get them to code it.

I found 106 to be a small sample and wanted to review and 'keep' more sentences to train my "taste." Claude offered another script which was basically a local embedding model that shortlisted unreviewed sentences per craft-move which I then confirmed as real hits in the chat and dropped the false positives. This took my gold set from 106 to 162, coded by the six craft-moves identified earlier.

Now comes the application part. I fed selected recent posts from my blog to Claude and asked it to matched my attempts at writing against the six moves. The findings:

Deadpan is your strength. You reach for it and land it. This matches your taste.

Aphorism is rare in your writing but lands when you try ("bloodlust dressed up as patriotism"). You can do it; you just do not reach for it.

Power you argue over paragraphs instead of compressing into one portable line, though you nail it when you do.

The yoked analogy is the big gap: your most-admired move and your least-produced one.

A tic surfaced: your keeps contain zero questions, yet your posts habitually end on a rhetorical question. Your taste is declarative; your habit is interrogative.

It turns out I rarely write the kind of sentence I love reading. What I'm actually good at is deadpan. So now comes the learning part.

When asked how I could improve my writing based on the sentences I liked from the newsletter, it offered the following advice:

Three moves, in order of payoff:

1. Build one yoked analogy per post, on purpose. This is your most-admired, least-used move. When you hit an abstract claim (an economy, a policy, a technology), stop and ask what unlike thing it resembles. You did it with Circe. Do it deliberately, once per piece, until it is reflex.
2. Compress one argument into an aphorism. In every political or reflective post, find the paragraph that carries the point and try to crush it into one portable line. You already prove you can: "bloodlust dressed up as patriotism." Make it a habit, not an accident.
3. Kill the question ending. Replace at least half your closing questions with a declarative. Your 'keeps' never end on a question; your best closers ("Well, do you now?") work because they are rare and pointed, not because they are questions. End on a statement and see if it lands harder.

Admittedly, that was solid advice for writing something that I liked to read from others. It even offered to show me an example of how to do that with an existing post with suggestions for three sentences or parts of the post to incorporate three of the moves mentioned above. The resulting post was definitely better, more succinct, and a pleasure to read.

I found this exercise immensely satisfying and educational. I learned specifics about types of good writing; something that I hadn't paid attention to. I knew what I liked to read but never why. This exercise also demystified the art of good writing and emphasized like any other craft, it needs to be honed and perfected through deliberative practice.

Of course, just writing great sentences doesn't make for a great text. You still need a solid structure, engaging content, and a strong argument. But that's a separate craft. For now, it's enough to know why a sentence works.