Notes on writing in two languages

I spend my working hours writing English: commit messages, documentation, reports about what a model did and why. At night I write मुक्तक in Nepali.

I assumed for a long time that these were the same skill pointed in two directions. They are not.

English asks me to be unambiguous

Technical English rewards a particular shape of sentence: one claim, stated plainly, with the uncertainty marked explicitly. The model reached 91% validation accuracy on a held-out set of 4,000 images. Every word is load bearing. If a sentence can be read two ways, it is a defect.

Nepali verse asks the opposite

A मुक्तक works because a line can be read two ways. The reader arrives at the last line and the first one quietly changes meaning behind them. Ambiguity is not a defect there. It is the mechanism.

Documentation fails when the reader has to think. A poem fails when they do not.

What each has taught the other

Verse has made my technical writing shorter. Four lines forces you to find out which words were doing nothing, and that habit transfers. Most first drafts of a README are twice as long as the final one for the same reason a first draft of a मुक्तक is.

Technical writing has made my verse more concrete. Abstractions are easy and cheap in poetry: grief, time, distance. Naming a specific thing is harder and almost always better. The same instinct that makes me write 4,000 images instead of a large dataset now makes me reach for a particular street rather than the idea of a city.

They are still different kinds of thinking. But they sharpen the same edge.