Words Spent

One of my favorite Djikstra programming quotes is about thinking via “lines of code spent” rather than “lines of code produced”. I started using this as a philosophy in my writing too: words spent.

Background

One of the things that’s surprised me about student writing is how poorly words are spent. You’ll have a solution where the trivial boilerplate steps are painfully verbose, and then the actually important parts are missing all the critical details.

I wonder how much of this is because of crummy writing advice. In school essays, even when you have nothing meaningful to say, teachers often impose a minimum word count¹ as a “proof of work”. The implied conclusion is that writing many words is the goal. Which is same mistake that counting “lines of code produced” is.

More careful teachers might talk about being concise, and might impose a maximum word count instead (or in addition). Then students get the idea you have this elastic dial that goes from “too little detail” to “too much detail”, and you need to guess the magic “just right” to win your teacher’s approval.

But that’s also mistaken. The “right” numbers of words should depend on what you have to say, just as the number of lines of code should reflect the complexity of the program you’re trying to implement.

Proposal

My proposal is to use “words spent” — just that two-word phrase — as a guideline to steer writers away from bad habits they’ve learned in school. I’ve been using it myself recently and I think it really conveys the right instincts concisely.

Examples:

• People instinctively know to avoid costs. If you treat word count as “words spent”, you won’t try to inflate it.
  • Orwell’s advice “if a word can be cut out, always cut it out” becomes an obvious corollary. Of course you cut unneeded costs.
• People also instinctively know that costs can be necessary. If you buy a car for 5000 dollars, you don’t expect much from it.
• Cost prioritization becomes instinctive. You know you should be spending the most words on the most important parts. In principle, the reader can guess what matters by how many words are spent on it.
• Cost efficiency becomes natural too. You want good words that do more with less.

Example usage

Individual sentences

You could use “words spent” to guide individual sentences. For example, in a different blog post, I had a draft with the passage:

Olga and I were originally planning to do a variant on Slitherlinks, and we had some prototypes that felt decent by mid-February. I was trying to not be too ambitious because of the tight deadline, so I was trying to keep things simple. We had a verifier up and running too like I did for BWMS, so we felt we could iterate quickly once we got the meta specification.

but on re-reading I deleted a third of the words:

Olga and I were originally planning Slitherlinks variants, and had some decent prototypes by mid-February. I wanted to keep things simple because of the tight deadline. We had a verifier written (like with BWMS), so we could iterate quickly.

(It may be possible to overdo this. If I’m being really aggressive, almost every sentence has some fat, but at some point the sentences start to sound unnatural. I would gladly spend three extra words to make a paragraph flow better, even if those words technically don’t change the meaning of the paragraph.)

Overall picture

But I think it’s more important to use “words spent” at a higher level, rather than micromanaging individual sentences too much. Here are some examples of ways it’s improved my writing style overall:

• Sometimes in my writing you’ll see claims whose proof is the single word “Clear” or similar. I’m never terse out of laziness². It’s always because I’ve decided it’s not worth the words spent to say more.

• Conversely, when writing math exposition, I really value good examples. So I know that I should spend a lot of words on examples, and spend fewer words on formal definitions or proofs that I want to de-emphasize.

• I know pictures aren’t words, but treat them the same way. (Okay, fine: “a picture is worth 1000 words”.) Point is, good diagrams and pictures are really cost-efficient. They might take a lot of time for me to draw, but the author’s time³ is irrelevant to this discussion. You can easily convey something in a half-page diagram that even 1000 words can’t do. If so, do it. (Tables can be really good too.)

Zooming out even further, here’s some thoughts on how the Napkin and LAMV books fit into this framework.

I think I was happy with the Napkin because I was cost-efficient: tons of diagrams, emphasis on intuition over proofs, etc. That’s why Napkin is a good birds-eye view of topics, without getting bogged down by fine or technical details. It’s not a textbook — concisely conveying intuition is the goal of Napkin.

But on the other extreme, LAMV was meant to outright replace the class textbook. That was really freaking expensive, especially for the target audience, who typically really need things spelled out in full. So I think LAMV was successful exactly because it put in writing all the detailed exposition⁴ that others were simply unable or unwilling to do. In this case the enormous cost paid off.