@cite{engelhardt-etal-2006}

@cite{sedivy-etal-1999} @cite{grice-1975} @cite{dale-reiter-1995}

Do Speakers and Listeners Observe the Gricean Maxim of Quantity? Journal of Memory and Language 54(4), 554–573.

Core Argument

Three experiments test whether speakers and listeners follow the Gricean Maxim of Quantity — the requirement that contributions be as informative as required but not more informative than required.

• Exp 1 (production, N = 24): Speakers over-describe ~31% of the time, using unnecessary modifiers (color adjectives, relative clauses, prepositional phrases) when a bare NP suffices.
• Exp 2 (judgment, N = 72): Listeners rate over-descriptions no worse than concise descriptions, but rate under-descriptions significantly worse.
• Exp 3 (eye-tracking, N = 48): Over-descriptions cause implicit processing costs — longer first-pass reading times on the head noun and more regressions back to the unnecessary modifier.

Conclusion: the language processing system is "only moderately Gricean." Speakers routinely violate the "not more informative" half of Quantity, listeners are explicitly tolerant of these violations but implicitly sensitive to them.

Connection to Contrastive Inference

The eye-tracking processing cost for unnecessary modifiers is the complement of @cite{sedivy-etal-1999}'s contrastive inference benefit from necessary modifiers. Both effects confirm that the comprehension system is sensitive to modifier presence and its pragmatic licensing.

Verified Data

Eye-tracking F-statistics and reading times verified against running text (§4). Judgment t-test significance levels verified against running text (§3). Production proportions and judgment mean ratings read from Tables 1–2 (not restated in running text).

structure Phenomena.Reference.Studies.EngelhardtEtAl2006.ProductionRate : Type

Over-description and modification rates from Exp 1 (Table 1). In 1-referent displays, speakers used modified NPs 31% of the time when a bare NP would have been sufficient.

• modified : Float

  Proportion of modified NP productions

• se : Float

  Standard error

instance Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprProductionRate : Repr ProductionRate

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprProductionRate = { reprPrec := Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprProductionRate.repr }

def Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprProductionRate.repr : ProductionRate → ℕ → Std.Format

Equations

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

def Phenomena.Reference.Studies.EngelhardtEtAl2006.exp1_target_1ref : ProductionRate

1-referent target: 31% over-description (Table 1).

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.exp1_target_1ref = { modified := 0.31, se := 5e-2 }

def Phenomena.Reference.Studies.EngelhardtEtAl2006.exp1_target_2ref : ProductionRate

2-referent target: 90% modification — necessary (Table 1).

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.exp1_target_2ref = { modified := 0.90, se := 4e-2 }

def Phenomena.Reference.Studies.EngelhardtEtAl2006.exp1_goal_noCompetitor : ProductionRate

No-competitor goal: 19% over-description (Table 1).

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.exp1_goal_noCompetitor = { modified := 0.19, se := 5e-2 }

def Phenomena.Reference.Studies.EngelhardtEtAl2006.exp1_goal_competitor : ProductionRate

Competitor goal: 92% modification — necessary (Table 1).

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.exp1_goal_competitor = { modified := 0.92, se := 3e-2 }

structure Phenomena.Reference.Studies.EngelhardtEtAl2006.JudgmentResult : Type

Judgment result: between-subjects t-test comparing two instruction types. Rating scale 1–5 (1 = "very bad", 5 = "very good"). The three participant groups each saw a different instruction version, so comparisons use independent-sample t-tests.

• conciseRating : Float

  Mean rating for concise instruction

• altRating : Float

  Mean rating for alternative instruction (over-described or modified)

• df : ℕ

  Degrees of freedom

• tStat : Option Float

  t-statistic; none when paper reports "< 1"

• significant : Bool

  Significant at p < .05

instance Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprJudgmentResult : Repr JudgmentResult

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprJudgmentResult = { reprPrec := Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprJudgmentResult.repr }

def Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprJudgmentResult.repr : JudgmentResult → ℕ → Std.Format

Equations

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

def Phenomena.Reference.Studies.EngelhardtEtAl2006.exp2_target_1ref : JudgmentResult

1-referent target: over-description NOT rated worse than concise. Concise 3.93 vs over-described 3.89, t₁(46) < 1 (§3).

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.exp2_target_1ref = { conciseRating := 3.93, altRating := 3.89, df := 46, tStat := none, significant := false }

def Phenomena.Reference.Studies.EngelhardtEtAl2006.exp2_target_2ref : JudgmentResult

2-referent target: under-description rated significantly worse. Concise (= under-described) 3.31 vs modified 3.88, t₁(46) = 3.69, p < .001 (§3).

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.exp2_target_2ref = { conciseRating := 3.31, altRating := 3.88, df := 46, tStat := some 3.69, significant := true }

def Phenomena.Reference.Studies.EngelhardtEtAl2006.exp2_goal_noCompetitor : JudgmentResult

No-competitor goal: over-description NOT rated worse. Concise 3.56 vs over-described 3.59, t₁(46) < 1 (§3).

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.exp2_goal_noCompetitor = { conciseRating := 3.56, altRating := 3.59, df := 46, tStat := none, significant := false }

def Phenomena.Reference.Studies.EngelhardtEtAl2006.exp2_goal_competitor : JudgmentResult

Competitor goal: under-description rated significantly worse. Concise (= under-described) 3.36 vs modified 3.69, t₁(46) = 2.01, p < .05 (§3).

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.exp2_goal_competitor = { conciseRating := 3.36, altRating := 3.69, df := 46, tStat := some 2.01, significant := true }

@[reducible, inline]

abbrev Phenomena.Reference.Studies.EngelhardtEtAl2006.AnovaResult : Type

Reuse ANOVA result structure from @cite{sedivy-etal-1999}. Both studies use by-subjects (F₁) and by-items (F₂) ANOVAs on eye-tracking data.

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.AnovaResult = Phenomena.Reference.Studies.SedivyEtAl1999.AnovaResult

structure Phenomena.Reference.Studies.EngelhardtEtAl2006.ReadingMeasure : Type

Eye-tracking reading measure with mean values and ANOVA.

• modified : Float

  Mean for modified (over-described) condition

• bare : Float

  Mean for bare (concise) condition

• anova : AnovaResult

  ANOVA result

instance Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprReadingMeasure : Repr ReadingMeasure

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprReadingMeasure = { reprPrec := Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprReadingMeasure.repr }

def Phenomena.Reference.Studies.EngelhardtEtAl2006.instReprReadingMeasure.repr : ReadingMeasure → ℕ → Std.Format

Equations

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

def Phenomena.Reference.Studies.EngelhardtEtAl2006.exp3_headNoun_firstPass : ReadingMeasure

Head noun first-pass reading time: modified 531ms vs bare 489ms. Listeners spend significantly longer reading the head noun after an unnecessary modifier. F₁(1,47) = 9.31, p < .01; F₂(1,23) = 7.32, p < .05 (§4, non-matching condition).

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.exp3_headNoun_firstPass = { modified := 531, bare := 489, anova := { F1 := 9.31, df1 := 47, F2 := 7.32, df2 := 23, significant := true } }

def Phenomena.Reference.Studies.EngelhardtEtAl2006.exp3_postNoun_regressions : ReadingMeasure

Post-noun regressions back to adjective: modified 14.6% vs bare 7.3%. Listeners regress to re-read the unnecessary modifier significantly more often. F₁(1,47) = 17.74, p < .001; F₂(1,23) = 14.15, p < .001 (§4, non-matching condition).

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.exp3_postNoun_regressions = { modified := 14.6, bare := 7.3, anova := { F1 := 17.74, df1 := 47, F2 := 14.15, df2 := 23, significant := true } }

def Phenomena.Reference.Studies.EngelhardtEtAl2006.exp3_headNoun_firstFixation : AnovaResult

Head noun first-fixation duration: marginal by subjects, significant by items. F₁(1,47) = 3.19, p < .08; F₂(1,23) = 5.12, p < .05 (§4, non-matching condition).

Equations

• Phenomena.Reference.Studies.EngelhardtEtAl2006.exp3_headNoun_firstFixation = { F1 := 3.19, df1 := 47, F2 := 5.12, df2 := 23, significant := false }

theorem Phenomena.Reference.Studies.EngelhardtEtAl2006.violation_mapping : Theories.Pragmatics.GriceanMaxims.QuantityViolation.underInformative.submaxim = Theories.Pragmatics.GriceanMaxims.QuantitySubmaxim.Q1 ∧ Theories.Pragmatics.GriceanMaxims.QuantityViolation.overInformative.submaxim = Theories.Pragmatics.GriceanMaxims.QuantitySubmaxim.Q2

Q1 and Q2 violations map to under- and over-description.

theorem Phenomena.Reference.Studies.EngelhardtEtAl2006.q1_q2_asymmetry : ¬exp2_target_1ref.significant = true ∧ ¬exp2_goal_noCompetitor.significant = true ∧ exp2_target_2ref.significant = true ∧ exp2_goal_competitor.significant = true

The two sub-maxims of Quantity behave asymmetrically in explicit judgments: Q1 violations (under-description) are penalized, but Q2 violations (over-description) are not.

theorem Phenomena.Reference.Studies.EngelhardtEtAl2006.q2_implicit_processing_cost : exp3_headNoun_firstPass.anova.significant = true ∧ exp3_postNoun_regressions.anova.significant = true

Despite explicit tolerance of Q2 violations, over-descriptions incur implicit processing costs: both first-pass reading times and regression rates are significantly elevated.

theorem Phenomena.Reference.Studies.EngelhardtEtAl2006.over_description_slows_reading : exp3_headNoun_firstPass.modified > exp3_headNoun_firstPass.bare

Over-descriptions slow head-noun reading: 531ms > 489ms.

theorem Phenomena.Reference.Studies.EngelhardtEtAl2006.over_description_doubles_regressions : exp3_postNoun_regressions.modified > exp3_postNoun_regressions.bare

Over-descriptions double regression rate: 14.6% vs 7.3%.

theorem Phenomena.Reference.Studies.EngelhardtEtAl2006.speakers_distinguish_necessity : exp1_target_2ref.modified > exp1_target_1ref.modified ∧ exp1_goal_competitor.modified > exp1_goal_noCompetitor.modified

Speakers are sensitive to discriminability: necessary modification is near-ceiling (~90%) while unnecessary modification is ~31%.

theorem Phenomena.Reference.Studies.EngelhardtEtAl2006.complementary_eye_tracking : SedivyEtAl1999.exp1_competitor_contrast.significant = true ∧ exp3_headNoun_firstPass.anova.significant = true ∧ exp3_postNoun_regressions.anova.significant = true

Both this study and @cite{sedivy-etal-1999} show that adjective modifiers affect eye-movement patterns, but in complementary ways:

• @cite{sedivy-etal-1999}: necessary modifiers trigger contrastive inferences, speeding target identification (competitor fixation).
• This study: unnecessary modifiers impose a processing cost, slowing head-noun reading and increasing regressions.

Both effects are significant, confirming that the comprehension system is sensitive to modifier presence and its pragmatic licensing.

theorem Phenomena.Reference.Studies.EngelhardtEtAl2006.moderately_gricean : exp1_target_1ref.modified > 0.2 ∧ ¬exp2_target_1ref.significant = true ∧ exp2_target_2ref.significant = true ∧ exp3_headNoun_firstPass.anova.significant = true ∧ exp3_postNoun_regressions.anova.significant = true

The paper's main conclusion: the language processing system is "only moderately Gricean." Evidence:

1. Speakers violate Q2 31% of the time (over-description)
2. Listeners don't explicitly penalize Q2 violations (tolerant)
3. But listeners implicitly detect Q2 violations (processing cost)
4. Listeners DO penalize Q1 violations (under-description)

Q1 (informativeness) is enforced; Q2 (non-redundancy) operates implicitly rather than explicitly.

theorem Phenomena.Reference.Studies.EngelhardtEtAl2006.supports_noBrevity : exp1_target_1ref.modified > 0.2 ∧ ¬exp2_target_1ref.significant = true ∧ exp2_target_2ref.significant = true ∧ DaleReiter1995.BrevityInterpretation.noBrevity.strength = 0

@cite{dale-reiter-1995} argue that Q2 should be interpreted as "No Brevity" — speakers use a fixed preference order and include any discriminating attribute without optimizing for brevity. This study provides direct empirical support:

1. Speakers over-describe 31% of the time (Q2 violated in production)
2. Over-descriptions are not penalized in judgment (Q2 tolerated)
3. Under-descriptions ARE penalized (Q1 enforced)

This matches the No-Brevity regime: Q1 is enforced, Q2 is not.