NSM / semantic primes (writing rule)

What the rule is

Natural Semantic Metalanguage (NSM) posits a small universal vocabulary of irreducible meanings (semantic primes). Writing under this rule means: introduce compressed terms only after a reductive paraphrase built from everyday relational language close to those primes (THERE IS, SOMEONE, DO, HAPPEN, THINK, KNOW, WANT, INSIDE, BEFORE, AFTER, BECAUSE, IF, CAN, NOT, PART OF, KIND OF, …).

What good looks like

  • A definition that a careful non-specialist could restate without the original jargon token.

  • No circular dictionary loops (“garbage collection is the collection of garbage”).

  • Technical labels appear after the paraphrase, then get reused for brevity.

Failure modes for LLM writing

  • Fake simplicity: short sentences that still depend on undefined abstractions.

  • Prime cosplay: sprinkling words like “want” and “inside” without actually reducing the concept.

  • Over-reduction that becomes false (dropping essential constraints to keep the paraphrase cute).

Micro-example

Weak: “A mutex provides mutual exclusion for critical sections.”

Stronger (NSM-flavored): “There is a thing two programs must not do at the same time. If one is doing it, the other cannot do it until the first has finished. People call that kind of control a mutex.”