Documentation

Linglib.Semantics.Quantification.Numerals.Roundness

Graded Numeral Roundness (k-ness Model) #

[Kri07a] [Sig88] [WWL+23] [JP01] [Cum15]

Framework-agnostic infrastructure for graded numeral roundness, following [Sig88], [JP01], and [WWL+23].

A number n has k-ness if it lies in [JP01]'s set [k × (1–9 × 10ⁿ)] (their p. 198): n = m × k × 10^b with 1 ≤ m ≤ 9 — so 10-ness is the k = 1 family (divisors 10, 100, …), per their own example "70 has only 10-ness". Their original allows b ≥ 0; following [WWL+23] (fn. 3) the search starts at b ≥ 1, which drops the single digits from 10-ness and 15, 45, … from 5-ness (cf. Studies/JansenPollmann2001.lean for the original and the divergence).

The 6 properties, ordered by strength as frequency predictors in [WWL+23]'s negative binomial regression (strongest first): 10-ness (β = 4.46), 2.5-ness (β = 3.84), 5-ness (β = 3.39), 2-ness (β = 2.74), multiple of 10 (β = 2.45), multiple of 5 (β = 0.06); the 2-ness and multiple-of-10 credible intervals overlap.

Main definitions #

k-ness primitives #

n has integer k-ness: n = m × k × 10^b for some b ≥ 1 and 1 ≤ m ≤ 9 ([JP01]). The witness search is bounded at b ≤ 10, valid for n < 10¹¹.

Equations
Instances For

    n has 2.5-ness: n = m × 2.5 × 10^b for b ≥ 1, 1 ≤ m ≤ 9 — equivalently 2n = m × 5 × 10^b. Search bounded as in HasKness.

    Equations
    Instances For

      Roundness score #

      The six graded roundness properties of [Sig88] and [JP01] — multiple of 5, multiple of 10, 2-ness, 2.5-ness, 5-ness, 10-ness — counted equally. The count predicts numeral frequency and pragmatic behavior ([WWL+23]).

      Count of true roundness properties (0–6). Higher = rounder.

      Equations
      • One or more equations did not get rendered due to their size.
      Instances For

        Maximum possible roundness score.

        Equations
        Instances For

          Roundness grade (binned score) #

          Binned roundness grade for use in width/tolerance functions.

          Collapses the 0–6 score into 4 levels to avoid duplicating step-function logic across Theory files.

          Instances For
            Equations
            • One or more equations did not get rendered due to their size.
            Instances For
              @[implicit_reducible]
              Equations

              Classify a number into a roundness grade.

              Equations
              • One or more equations did not get rendered due to their size.
              Instances For

                Context-sensitive roundness #

                Count k-ness-like properties relative to a non-standard base.

                For base b, checks divisibility by b, 2b, 5b, and 10b — mirroring the standard k-ness properties but on a different scale.

                Examples:

                • contextualRoundnessScore 48 12 = 2 (48 ÷ 12 = 4, 48 ÷ 24 = 2)
                • contextualRoundnessScore 120 12 = 4 (divides by 12, 24, 60, 120)
                Equations
                • One or more equations did not get rendered due to their size.
                Instances For

                  Context-sensitive roundness: compose default k-ness with a non-standard base.

                  On a base-12 (dozens) scale, 48 = 4 × 12 is "round" even though its default k-ness score is 0. On base-60 (minutes), 120 = 2 × 60 is round.

                  The contextual score derives from actual divisibility properties relative to the base (not a flat bonus), paralleling how standard k-ness derives from divisibility by 2/2.5/5/10 × powers of 10.

                  Equations
                  Instances For

                    Per-datum verification #

                    Multiples of 10 have roundness score ≥ 2 (multiple-of-5 and multiple-of-10 both hold). The keystone for downstream sorry-free proofs.