Natural language is strong at goals and constraints; parameter controls are strong at degree and numeric values

What it is

“For legal, must not sound like an admission, must mention order 8472” is goals and constraints. “One hundred fifty words, formal band, table not paragraphs” is degree and value. The goals-versus-magnitude split cuts the two channels’ home grounds one notch finer: language owns propositions (what should hold, what must not), controls own choices that already sit on a dimension. Coexistence only says both sides must be present. This cut says which class of meaning each owns, so the two entries do not fight to express the same thing and jointly drop the other.

“Suited to intent” is still coarse. Intent mixes propositions with numbers smuggled in. Picking the numbers out of the proposition and handing them to controls is what lets language actually do the stretch it is good at.

Why it happens

Propositions compose, negate, and point at objects. Quantities order, step, and compare. Expressing a negative constraint (“must not sound like an admission”) in controls means a switch or a banned-word list, and composition explodes as soon as it gets complex. Expressing “one hundred fifty words” in language re-estimates every decode. The mismatch is at the type layer: stuffing the composable into a finite enum, stuffing the orderable into an adjective.

People mix both in one spoken sentence. The UI’s job is not to ban the mix. It is to split the mixed sentence before send and route each class to its channel. A clean split is what later adjustment needs in order to know whether to move a slider or to edit the ban.

Studying it

Pre-label the same requests into proposition slots and quantity slots. Watch, on a hybrid surface, which class people actually put in which channel, and whether the product routes by type. Independent variables: whether the UI does a visible split before send (“I put word count in length”), whether quantities may remain written in the sentence. Dependent variables: type hit rate (propositions through language, quantities through controls), dual-write conflict rate, share of later adjustments that find the right channel.

Use a mismatched UI as control: numbers forced into sentences, bans implemented as switches. The mismatched group should slow significantly at the adjustment stage — behavioral evidence that the split failed.

Where it stops holding

Some constraints are enums (output language, file format): they look like propositions and still belong on controls; the split follows type, not “does it look like a sentence.” Some degree words have no dimension yet (“a bit sharper” for prose); leave them in language rather than inventing a slider for the sake of the split. “Must not admit fault” in legal work is a proposition; putting it on a tone slider is a type error. This entry does not treat who overrides whom in a conflict.

Applying it

• Before send, echo extracted numbers and bands onto the matching controls, and leave bans and objects in a language summary. Let the user correct a wrong split.
• Teach the surface: word count goes to length, tightening a ban goes to that sentence or a banned-word list. Do not let both entries secretly edit the same field undeclared.
• Promote a new quantity to a control only after people have repeatedly said it in language. Do not pre-build a slider for every adjective.
• Check: one mixed request (goal + word count + one ban). The split should drop word count into the length control and keep the ban on the proposition side. Then ask for “just a bit shorter” and “tighten that ban.” If both attempts go to the same box, the split has not entered adjustment habit.

Related

• Same group: L2.04.1 Natural language is suited to expressing intent · L2.04.2 Parameter controls are suited to precise adjustment · L2.04.3 Having both beats choosing one · 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: L2.06 Granularity of Control · L2.15 Ambiguity and Clarifying Follow-ups · L2.01 Openness of Natural-Language Commands and Its Cost
• Search terms: goals-versus-magnitude split · proposition versus scalar · channel-type mapping