Gradient At-Issueness and Projectivity #
@cite{roberts-2012} @cite{tonhauser-beaver-degen-2018}
At-issueness and projectivity are gradient, not binary, and anti-correlated. @cite{tonhauser-beaver-degen-2018} report r = .85 (9 expression types) and r = .99 (12 predicates) for the positive correlation between not-at-issueness and projectivity, equivalently a negative correlation between at-issueness and projectivity. This module lifts at-issueness from a binary enum to a bounded rational degree with threshold semantics, mirroring the pattern used for gradable adjectives (degree > θ → positive meaning).
Design #
All four degree/threshold types are Rat01 (= ↥(Set.Icc (0 : ℚ) 1))
from Core.Scales.Scale. The type aliases document intent but share
infrastructure (order instances, smart constructors, etc.).
| Adjective | At-issueness |
|---|---|
Degree max (Fin) | AtIssuenessDegree (Rat01) |
Threshold max | AtIssuenessThreshold (Rat01) |
positiveMeaning d θ | isAtIssue d θ |
A degree of at-issueness ∈ [0, 1]. 0 = fully backgrounded (not at-issue), 1 = fully at-issue.
Instances For
A degree of projectivity ∈ [0, 1]. 0 = no projection, 1 = obligatory projection.
Instances For
Contextual threshold for at-issueness classification. Content with degree above this threshold counts as at-issue.
Instances For
Contextual threshold for projectivity classification.
Instances For
Content is at-issue when its at-issueness degree exceeds the threshold.
Mirrors positiveMeaning from Adjective/Theory.lean.
Equations
Instances For
Content is projective when its projectivity degree exceeds the threshold.
Equations
Instances For
Binary at-issueness, recoverable from gradient degree + threshold.
This is the same enum as in ProjectiveContent.lean, now derived.
- atIssue : AtIssuenessClassical
- notAtIssue : AtIssuenessClassical
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Recover binary classification from gradient degree and threshold.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Default threshold at 0.5, matching the midpoint of the [0, 1] scale.
Instances For
Default projectivity threshold at 0.5.
Instances For
A pair of at-issueness and projectivity ratings for a single expression.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Anti-correlation property: across a list of expression ratings, higher at-issueness systematically co-occurs with lower projectivity.
@cite{tonhauser-beaver-degen-2018} report r = .85 across 9 expression types (Exp 1a) and r = .99 across 12 clause-embedding predicates (Exp 1b), for the positive correlation between not-at-issueness and projectivity. This is equivalent to an anti-correlation between at-issueness and projectivity.
This structure captures the deterministic (monotone) version of the Gradient Projection Principle. The statistical version uses Pearson r.
- pairs : List GradientPair
The paired ratings
Instances For
Qualitative connection between QUD and at-issueness: content that doesn't distinguish QUD cells has low at-issueness.
@cite{roberts-2012}: at-issue content is content that addresses the QUD. Content that yields the same truth value across all worlds within each QUD cell is backgrounded (not at-issue).
The full measure-theoretic version would compute mutual information between content and QUD partition.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Both at-issueness and projectivity are closed-bounded scales on [0, 1].
Instances For
Construct a degree from a rational in [0, 100], normalizing to [0, 1]. Useful for entering experimental ratings.
Equations
- Core.Discourse.AtIssueness.ofPercent n h0 h1 = ⟨n / 100, ⋯⟩