A reader stopped on the fifth word of my novel and asked whether it had been translated from Swedish.
It has not. I wrote it in English. But it is a better thing to receive than a compliment, and it is the same question I have been circling since I was seventeen.
California, 1987
I spent a year in the USA as an exchange student. Somewhere in the middle of it the thing I had gone there for finally happened: I stopped translating and started thinking in English directly.
Here is what I wrote about it fifteen years later, in the foreword of my thesis:
At first I was delighted, this was great, I could interact with my environment much faster when I no longer had to constantly translate every sentence into Swedish, think up a response and then translate that back to English. The joy was not long lasting though; I soon found that I had become less witted. My limited English vocabulary made it impossible to think certain thoughts; I simply lacked the words.
This taught me an important lesson about the power of language: If your language (or your grasp of it) is not up to the task, you put your intelligence and creativity at risk.
Faster and stupider, and they arrived together, in the same week, as one improvement. The loss felt like fluency.
The foreword opens with Dijkstra, which is still the best sentence anyone has written on this:
The tools we use have a profound (and devious!) influence on our thinking habits, and, therefore, on our thinking abilities.
Devious is the load-bearing word. Not restrictive. Devious. A limited vocabulary does not announce itself as a limited vocabulary. It shows up as having fewer ideas, and you attribute it to yourself.
The same lesson, in languages I chose
That paragraph is most of the reason my career looks the way it does.
I went to Uppsala because I wanted a more powerful language and they still taught Lisp. In 1994 I found Erlang, and I stayed long enough to write a compiler for it. What kept me was never the performance work. It was that I could think about concurrency directly, in the language, without first translating it into something the language was able to hold. The translation step was gone, and this time nothing went missing inside it.
In 2003 I went to EPFL to work with Martin Odersky on Scala, and 1987 came straight back. I could not think intelligently in Scala. The language was too complex and my grasp of it too shallow, and from the inside those two are indistinguishable. That is the whole problem in one sentence. You cannot tell the difference between a language that will not express your idea and an idea you do not actually have.
Anyone who has moved between languages has felt some version of this. The thing you want to say in one is unsayable in the next, so you say a nearby thing instead, and after a while you stop wanting to say the first thing at all. The language does not block the thought. It just makes some thoughts more expensive than you are willing to pay for, which over a long enough period is the same as blocking it.
Seven years of not writing a novel
I started my novel on 9 November 2018. That is the date on the first commit,
and the commit contains no novel. It has a README with one line in it, "A
thriller about a blockchain development", and a .gitignore that excludes
*.beam, erl_crash.dump and rebar, because I created the repository from
an Erlang project template and never changed it. The book's repository has now
spent seven years carefully ignoring compiled BEAM files that have never
existed.
I had the first scene in a paper notebook and a rule I had given myself: no second book until the first one is finished. The first one was The BEAM Book, about the virtual machine that runs the language I had spent a career on. It took considerably longer than I expected.
So the novel waited six years and four months in a repository with two files in it. Not a language problem. Just a promise, and I am glad I kept it.
The BEAM Book came out in 2025. I started writing the novel again more or less as it went to press.
Then I did it to myself
Chapter one now exists in 109 commits, and I have read the whole log.
The early drafts were overwritten in the ordinary way that early drafts are overwritten. There is a ceiling fan that rattles "like a dying breath". There are sirens wailing "like ghosts in the humid African night". There is a sentence explaining that panic was a luxury he could not afford, immediately after a paragraph of ragged breathing and clenched fists, in case the reader had missed it.
All of that deserved to go, and over the next seventeen months it went. The recent commit messages are almost all subtraction: cut the decorative tails at chapter endings, cut recaps and closers, cut named emotions, break up the composure inventory. Every one of those commits was right. I would make all of them again.
Here is the page as it stood the day my reader wrote to me:
Frost stared at the pottery shard buried in the man's throat. He made himself breathe, once, all the way down. The rest of the vase lay in blue-and-white pieces between the bed and the door. His right hand would not stop shaking. He pressed it flat against his thigh and waited for the fingers to obey him. The room smelled of sweat and damp plaster.
Better than what came before it. Obviously better. Now measure it. Those sentences run 21, 9, 15, 7, 15, 8, 6, 11, 12, 7 words. After the opening one, nothing goes above fifteen and nothing drops below six. Every sentence is a flat declarative. Nothing is subordinate to anything else. There is no long sentence anywhere for a short one to land against, so nothing lands.
Across the whole chapter: 239 sentences, mean length 9.3 words, 35 per cent of them under six.
That is what my reader was reacting to, and he was right without having any of this in front of him. It was never the vocabulary. It was the cadence. Translated prose often reads exactly like that, because translation carries the meaning across and drops the rhythm.
What this actually is
At seventeen I lost expressive range because I did not have the words. Thirty-nine years later I lost it again by optimising the words away, one defensible commit at a time, until there was no variation left to optimise.
Both times I could feel the loss without being able to find it. In California I knew I had become less witted; in the book I could measure the flatness. The cause stays hidden because you do not notice the sentences you can no longer write. You write the ones you can, and they seem fine, and they are fine one at a time.
Engineers will recognise the shape without caring about prose at all. It is local optimisation against a metric that is not the objective. Every step improves the number you are looking at. The system gets worse along an axis nobody is measuring, and it stays worse until somebody who does not share your assumptions looks at it cold and says something is off.
That is what code review is supposed to be and mostly is not, because the reviewer usually shares the assumptions. It is also why the useful reviewer is so often the new one, who has no idea why anything is the way it is and asks about it anyway.
The fix took four joins. The opening now runs 21, 9, 38, 6, 8, 10, 11, then three sentences of one word each, then 22. I added almost nothing back. The staccato was correct for a man in shock. An effect that is always on is not an effect.
The word that survived everything
There is one more thing in the log.
My reader's actual question was about the fifth word: what does "shard" mean here. It is a fine word, and it turned out to be the most durable one in the manuscript. It sits in the first sentence of the first draft. It survived the protagonist being renamed, the chapter order being rebuilt, the loss of about two thirds of the words around it, and 109 commits by a man removing everything he could justify removing. I never once considered cutting it. It was precise, it was load bearing, and it was the correct word for a broken piece of fired clay.
It did not survive one reader asking what it meant.
Its neighbours outlived it. "Buried" and "pottery" are both still on the page, from that same first sentence. "Pottery" did better than survive: it is the last word of the chapter and it is the chapter's title.
So the word I never considered cutting was the word doing the damage, and the word next to it, which I also never thought about, had quietly become the thing the chapter is named after. I had opinions about hundreds of decisions on that page. Those two were never decisions at all, and they are the two that mattered.
A coda for the compiler people
There is a joke in this that I did not plan.
I have spent twenty-five years on the idea that a language bounds what its users can think. Then I wrote a book whose central crime is a distinction the machine has no vocabulary for, and I did not notice the connection until a stranger asked me whether my English was really Swedish.
The words you do not have are the ones you cannot miss.
The thesis is Efficient Implementation of Concurrent Programming Languages, Uppsala 2002. The novel is The Chain, published under Frank Able, out 9 September. The first chapter is free at readthechain.com, and the Kindle edition is on Amazon for pre-order.
