Natural language is suited to expressing intent

What it is

“Write this letter as if it is for a customer who has already waited three weeks and been promised twice” — object, situation, relation, none of them has a knob that was named in advance. NL for intent means that when what you are submitting is a goal, a situation, and a relation, not a quantity that already has ticks, the language channel fits tighter than parameter controls. Controls need someone to have built the dimension. Intent often shows up before the dimension exists.

It does not say language is good at numbers, nor that both entries must coexist. It only marks language’s home ground: the part of meaning that has not yet been parameterized.

Why it happens

Intent is a propositional attitude: what should hold, in what story, for whom. Propositions can be composed on the fly; a control’s dimension table cannot. Language can say an unnamed constraint in one breath (“don’t let legal hear this as an admission”) and can point with deixis and ellipsis at objects in the current document. Delayed binding is a virtue here: structure grows with the sentence, instead of waiting for the product to invent sliders named “weeks waited” and “promises made.”

The brain’s task representation is already narrative. Recoding the narrative into a set of unrelated sliders is a second compilation, and misalignment and missing slots happen in that compilation. Language skips it.

Studying it

A set of tasks with situation and relation (including unnamed constraints). One group language only, one group a parameter panel built in advance (the panel deliberately has no dedicated control for the unnamed constraint). Dependent variables: whether the unnamed constraint enters the artifact, whether the user abandons mid-way, subjective “I said it clearly.” Independent variables: situational complexity, whether the panel offers a free Other sentence, whether the user is a designer/engineer (habitually parametric).

The task must contain at least one constraint the product did not foresee. If every intent already has a same-named control, you are not measuring language’s home ground.

Where it stops holding

If intent has already collapsed to “set brightness to 70,” language is a detour and the home ground belongs to controls. When an instruction is so vague that several mutually exclusive goals all fit, language’s expressive advantage becomes an ambiguity cost. When speech recognition is poor, or the user cannot speak the product’s language (cross-lingual, a symbol board for motor impairment), the language channel itself breaks first. This entry does not treat who should own degree and numeric values.

Applying it

• Keep one full intent sentence on the first beat of the task. Do not open with a parameter panel only. The panel can arrive after.
• Treat unnamed constraints as first-class: that sentence enters generation, rather than dying in a remark.
• When intent needs to point at a current object (this article, these three rows, that last letter), make deictic words bind to tappable objects so language and reference travel together.
• Check: a real request with an unnamed relational constraint (“as if they have waited three weeks and been promised twice”). If the parameter-only path has nowhere to put that sentence, and the language path writes the situation into the artifact, language’s home ground is in use. If the product swallows the sentence and only sets tone=formal, the home ground was wasted.

Related

• Same group: L2.04.2 Parameter controls are suited to precise adjustment · L2.04.3 Having both beats choosing one · L2.04.4 Natural language is strong at goals and constraints; parameter controls are strong at degree and numeric values · L2.04.5 Parameter controls provide a revertible determinate state; natural language does not · L2.04.6 When both exist, which one overrides the other must be explicit, or conflict is unpredictable · L2.04.7 Filling parsed language back into controls lets users see what the system understood · L2.04.8 A control’s value range is itself a display of the capability boundary
• Nearby: L2.01 Openness of Natural-Language Commands and Its Cost · L2.15 Ambiguity and Clarifying Follow-ups · L2.03 Examples and Template Guidance
• Search terms: NL for intent · delayed binding · unnamed constraint