@cite{sedivy-etal-1999}

Achieving Incremental Semantic Interpretation through Contextual Representation. Cognition 71(2), 109–147.

Core Argument

Visual-world eye-tracking shows that listeners draw contrastive inferences from scalar adjectives during incremental sentence processing. When a speaker says "Pick up the tall glass" in a context containing both a tall and a short glass, listeners fixate the target faster and look at the contrast-set member more than in contexts without a same-category competitor.

Three experiments (all using scalar/size adjectives: tall, short, big, small, fat, thin, long, wide) converge on this finding. The theoretical claim is that scalar adjectives trigger contrastive inferences because they are semantically restrictive — their interpretation depends on a contextually-determined comparison class, making their use pragmatically marked (informative) when a contrast set is available.

The General Discussion predicts that color adjectives, being non-restrictive (no comparison-class dependence), should NOT trigger contrastive inferences. This prediction was later tested and confirmed in Sedivy (2003, 2004), but challenged by @cite{ronderos-etal-2024} who found that color adjectives DO elicit contrastive inferences cross-linguistically.

Connection to PropertyDomain

The scalar adjectives tested all belong to PropertyDomain.size, which has requiresComparisonClass = true. The theoretical mechanism — comparison-class dependence drives contrastive inference — is thus encoded in the PropertyDomain infrastructure.

Verified Data

All F-statistics, degrees of freedom, and mean values verified against paper Tables 2–3 (Exp 1), Tables 5–7 (Exp 2), Tables 10–11 (Exp 3).

def Phenomena.Reference.Studies.SedivyEtAl1999.scalarAdjectives : List String

All three experiments used the same 8 scalar adjectives, all spatial dimension terms mapping to PropertyDomain.size.

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.scalarAdjectives = ["tall", "short", "big", "small", "fat", "thin", "long", "wide"]

def Phenomena.Reference.Studies.SedivyEtAl1999.adjDomain : Core.PropertyDomain

All tested adjectives belong to the size domain.

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.adjDomain = Core.PropertyDomain.size

inductive Phenomena.Reference.Studies.SedivyEtAl1999.TokenType : Type

Token typicality: how well the target exemplifies the adjective.

• good : TokenType
• poor : TokenType

instance Phenomena.Reference.Studies.SedivyEtAl1999.instDecidableEqTokenType : DecidableEq TokenType

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instDecidableEqTokenType x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯

def Phenomena.Reference.Studies.SedivyEtAl1999.instBEqTokenType.beq : TokenType → TokenType → Bool

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instBEqTokenType.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)

instance Phenomena.Reference.Studies.SedivyEtAl1999.instBEqTokenType : BEq TokenType

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instBEqTokenType = { beq := Phenomena.Reference.Studies.SedivyEtAl1999.instBEqTokenType.beq }

def Phenomena.Reference.Studies.SedivyEtAl1999.instReprTokenType.repr : TokenType → ℕ → Std.Format

Equations

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

instance Phenomena.Reference.Studies.SedivyEtAl1999.instReprTokenType : Repr TokenType

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instReprTokenType = { reprPrec := Phenomena.Reference.Studies.SedivyEtAl1999.instReprTokenType.repr }

inductive Phenomena.Reference.Studies.SedivyEtAl1999.Condition : Type

Contrast condition across all experiments.

• contrast : Condition
• noContrast : Condition

instance Phenomena.Reference.Studies.SedivyEtAl1999.instDecidableEqCondition : DecidableEq Condition

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instDecidableEqCondition x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯

def Phenomena.Reference.Studies.SedivyEtAl1999.instBEqCondition.beq : Condition → Condition → Bool

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instBEqCondition.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)

instance Phenomena.Reference.Studies.SedivyEtAl1999.instBEqCondition : BEq Condition

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instBEqCondition = { beq := Phenomena.Reference.Studies.SedivyEtAl1999.instBEqCondition.beq }

instance Phenomena.Reference.Studies.SedivyEtAl1999.instReprCondition : Repr Condition

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instReprCondition = { reprPrec := Phenomena.Reference.Studies.SedivyEtAl1999.instReprCondition.repr }

def Phenomena.Reference.Studies.SedivyEtAl1999.instReprCondition.repr : Condition → ℕ → Std.Format

Equations

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

structure Phenomena.Reference.Studies.SedivyEtAl1999.AnovaResult : Type

ANOVA result with by-subjects and by-items F-statistics.

• F1 : Float

  F-statistic (by subjects)

• df1 : ℕ

  Denominator df (by subjects); numerator is always 1

• F2 : Float

  F-statistic (by items)

• df2 : ℕ

  Denominator df (by items)

• significant : Bool

  Significant by subjects (p < 0.05)

def Phenomena.Reference.Studies.SedivyEtAl1999.instReprAnovaResult.repr : AnovaResult → ℕ → Std.Format

Equations

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

instance Phenomena.Reference.Studies.SedivyEtAl1999.instReprAnovaResult : Repr AnovaResult

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instReprAnovaResult = { reprPrec := Phenomena.Reference.Studies.SedivyEtAl1999.instReprAnovaResult.repr }

structure Phenomena.Reference.Studies.SedivyEtAl1999.Exp1Latency : Type

Mean latency (ms) of first eye movement to target from adjective onset. Table 2: faster target fixation with contrast.

• goodContrast : ℕ
• goodNoContrast : ℕ
• poorContrast : ℕ
• poorNoContrast : ℕ

instance Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp1Latency : Repr Exp1Latency

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp1Latency = { reprPrec := Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp1Latency.repr }

def Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp1Latency.repr : Exp1Latency → ℕ → Std.Format

Equations

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

def Phenomena.Reference.Studies.SedivyEtAl1999.exp1_latency : Exp1Latency

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp1_latency = { goodContrast := 479, goodNoContrast := 571, poorContrast := 540, poorNoContrast := 714 }

def Phenomena.Reference.Studies.SedivyEtAl1999.exp1_latency_contrast : AnovaResult

Exp 1: Main effect of contrast on target latency. F₁(1,23) = 8.29, P < 0.01; F₂(1,14) = 5.41, P < 0.05.

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp1_latency_contrast = { F1 := 8.29, df1 := 23, F2 := 5.41, df2 := 14, significant := true }

structure Phenomena.Reference.Studies.SedivyEtAl1999.Exp1CompetitorRate : Type

Proportion of trials with fixation to competitor within first two eye movements from adjective onset. Table 3.

• goodContrast : Float
• goodNoContrast : Float
• poorContrast : Float
• poorNoContrast : Float

instance Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp1CompetitorRate : Repr Exp1CompetitorRate

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp1CompetitorRate = { reprPrec := Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp1CompetitorRate.repr }

def Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp1CompetitorRate.repr : Exp1CompetitorRate → ℕ → Std.Format

Equations

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

def Phenomena.Reference.Studies.SedivyEtAl1999.exp1_competitor : Exp1CompetitorRate

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp1_competitor = { goodContrast := 0.67, goodNoContrast := 0.50, poorContrast := 0.68, poorNoContrast := 0.50 }

def Phenomena.Reference.Studies.SedivyEtAl1999.exp1_competitor_contrast : AnovaResult

Exp 1: Main effect of contrast on competitor fixation. F₁(1,23) = 5.26, P < 0.05; F₂(1,14) = 5.06, P < 0.05.

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp1_competitor_contrast = { F1 := 5.26, df1 := 23, F2 := 5.06, df2 := 14, significant := true }

def Phenomena.Reference.Studies.SedivyEtAl1999.exp2_latency_contrast : AnovaResult

Exp 2: Effect of contrast on target latency — NOT significant. F₁(1,23) = 0.54, P = 0.47; F₂(1,14) = 0.66. Unlike Exp 1, prenominal position means the noun disambiguates before the contrastive inference can speed target identification.

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp2_latency_contrast = { F1 := 0.54, df1 := 23, F2 := 0.66, df2 := 14, significant := false }

structure Phenomena.Reference.Studies.SedivyEtAl1999.Exp2CompetitorRate : Type

Proportion of trials with competitor fixation. Table 6.

• goodContrast : Float
• goodNoContrast : Float
• poorContrast : Float
• poorNoContrast : Float

instance Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp2CompetitorRate : Repr Exp2CompetitorRate

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp2CompetitorRate = { reprPrec := Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp2CompetitorRate.repr }

def Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp2CompetitorRate.repr : Exp2CompetitorRate → ℕ → Std.Format

Equations

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

def Phenomena.Reference.Studies.SedivyEtAl1999.exp2_competitor : Exp2CompetitorRate

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp2_competitor = { goodContrast := 0.28, goodNoContrast := 8e-2, poorContrast := 0.33, poorNoContrast := 0.20 }

def Phenomena.Reference.Studies.SedivyEtAl1999.exp2_competitor_contrast : AnovaResult

Exp 2: Main effect of contrast on competitor fixation. F₁(1,23) = 8.19, P < 0.01; F₂(1,14) = 6.70, P < 0.05.

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp2_competitor_contrast = { F1 := 8.19, df1 := 23, F2 := 6.70, df2 := 14, significant := true }

structure Phenomena.Reference.Studies.SedivyEtAl1999.Exp2ContrastObjRate : Type

Proportion of trials with fixation to contrast object. Table 7.

• goodContrast : Float
• goodNoContrast : Float
• poorContrast : Float
• poorNoContrast : Float

def Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp2ContrastObjRate.repr : Exp2ContrastObjRate → ℕ → Std.Format

Equations

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

instance Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp2ContrastObjRate : Repr Exp2ContrastObjRate

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp2ContrastObjRate = { reprPrec := Phenomena.Reference.Studies.SedivyEtAl1999.instReprExp2ContrastObjRate.repr }

def Phenomena.Reference.Studies.SedivyEtAl1999.exp2_contrastObj : Exp2ContrastObjRate

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp2_contrastObj = { goodContrast := 0.44, goodNoContrast := 0.10, poorContrast := 0.41, poorNoContrast := 0.11 }

def Phenomena.Reference.Studies.SedivyEtAl1999.exp2_contrastObj_contrast : AnovaResult

Exp 2: Main effect of contrast on contrast-object fixation. F₁(1,23) = 32.83, P < 0.001; F₂(1,14) = 29.93, P < 0.001.

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp2_contrastObj_contrast = { F1 := 32.83, df1 := 23, F2 := 29.93, df2 := 14, significant := true }

def Phenomena.Reference.Studies.SedivyEtAl1999.exp3_latency_contrast : AnovaResult

Exp 3: Effect of contrast on target latency. F₁(1,18) = 6.76, P < 0.05; F₂(1,17) = 4.36, P = 0.05.

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp3_latency_contrast = { F1 := 6.76, df1 := 18, F2 := 4.36, df2 := 17, significant := true }

def Phenomena.Reference.Studies.SedivyEtAl1999.exp3_competitor_contrast : AnovaResult

Exp 3: Main effect of contrast on competitor fixation. F₁(1,19) = 12.83, P < 0.01; F₂(1,15) = 11.73, P < 0.01.

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp3_competitor_contrast = { F1 := 12.83, df1 := 19, F2 := 11.73, df2 := 15, significant := true }

def Phenomena.Reference.Studies.SedivyEtAl1999.exp3_contrastObj_contrast : AnovaResult

Exp 3: Main effect of contrast on contrast-object fixation. F₁(1,19) = 70.29, P < 0.001; F₂(1,15) = 35.96, P < 0.001.

Equations

• Phenomena.Reference.Studies.SedivyEtAl1999.exp3_contrastObj_contrast = { F1 := 70.29, df1 := 19, F2 := 35.96, df2 := 15, significant := true }

theorem Phenomena.Reference.Studies.SedivyEtAl1999.scalar_contrastive_inference : exp1_competitor_contrast.significant = true ∧ exp2_competitor_contrast.significant = true ∧ exp2_contrastObj_contrast.significant = true ∧ exp3_competitor_contrast.significant = true ∧ exp3_contrastObj_contrast.significant = true

Core finding: scalar adjectives trigger contrastive inferences across all three experiments. Evidence comes from two measures: increased competitor fixation (all 3 exps) and increased contrast-object fixation (Exps 2, 3).

theorem Phenomena.Reference.Studies.SedivyEtAl1999.contrastObj_large_effects : exp2_contrastObj_contrast.F1 > 30 ∧ exp3_contrastObj_contrast.F1 > 70

Contrast-object effects are very large: the F-statistics for contrast-object fixation are the strongest effects in the paper, indicating robust contrastive inference.

theorem Phenomena.Reference.Studies.SedivyEtAl1999.scalar_requires_comparison_class : adjDomain = Core.PropertyDomain.size ∧ Core.PropertyDomain.size.requiresComparisonClass = true

All tested adjectives belong to the size domain, which requires comparison-class computation.

theorem Phenomena.Reference.Studies.SedivyEtAl1999.comparison_class_mechanism : Core.PropertyDomain.size.requiresComparisonClass = true ∧ Core.PropertyDomain.color.requiresComparisonClass = false ∧ Core.PropertyDomain.material.requiresComparisonClass = false

The Sedivy mechanism: adjective types that require comparison-class computation (size domain) trigger contrastive inferences. Adjective types that do not require comparison-class computation (color, material) are predicted not to trigger contrastive inferences.

The theoretical prediction is that requiresComparisonClass = true is a necessary condition for contrastive inference via the semantic-restrictiveness route.

theorem Phenomena.Reference.Studies.SedivyEtAl1999.size_has_intermediate_discrimination : Core.PropertyDomain.size.noiseDiscrimination = some RSA.Noise.sizeDiscrimination ∧ RSA.Noise.colorDiscrimination > RSA.Noise.sizeDiscrimination ∧ RSA.Noise.sizeDiscrimination > RSA.Noise.materialDiscrimination

The size domain (tested here) has intermediate noise discrimination (0.60), between color (0.98) and material (0.40). Despite not having the highest discrimination, scalar adjectives robustly trigger contrastive inferences — suggesting the comparison-class mechanism operates independently of perceptual discrimination.