Documentation

Linglib.Phenomena.Presupposition.Studies.DegenTonhauser2022

@cite{degen-tonhauser-2022}: Are There Factive Predicates? #

@cite{heim-1983} @cite{schlenker-2009} @cite{van-der-sandt-1992}

Empirical data from "Are there factive predicates? An empirical investigation" by Judith Degen and Judith Tonhauser (Language 98(3):552–591, 2022).

Finding #

Projection is gradient across predicates, with no categorical gap between traditional factive and nonfactive classes.

The gradient patterns observed in inference judgments do not, on their own, settle whether factivity is fundamentally discrete or gradient. As Degen & Tonhauser note, "the observed gradience in projection [is] compatible with a binary factivity category" under the assumption that predicates are ambiguous between factive and nonfactive readings. @cite{grove-white-2025} subsequently show that models implementing discrete factivity fit these data better than models implementing gradient factivity.

Predicates #

This paper reuses the 20 clause-embedding predicates from @cite{degen-tonhauser-2021} and adds a four-way traditional classification.

Experiments #

Experiment 1: Projection (Certain That Diagnostic) #

Presented polar questions with clause-embedding predicates
Asked: "Is [speaker] certain that [CC]?"
Found gradient projection, no categorical distinction

Experiments 2-3: Entailment #

Inference diagnostic: "Does it follow that [CC]?"
Contradictoriness diagnostic: "[Matrix] but [not CC]" contradictory?
Found only prove/be right pattern with entailed controls

Theoretical Implications #

Definition 3a (CC is presupposed) not supported:
- Canonically factive CCs not categorically more projective
Definition 3b (CC is presupposed AND entailed) not supported:
- Set of "factive" predicates is either empty or heterogeneous
For projection analyses:
- Unclear which predicates the analyses apply to
- Need to account for gradient projection

Connection to @cite{tonhauser-beaver-roberts-simons-2013} #

This paper extends and empirically tests the Tonhauser taxonomy:

SCF (Strong Contextual Felicity) relates to whether content must be established
OLE (Obligatory Local Effect) relates to attribution under embedding
The gradient projection patterns challenge simple binary classification

inductive Phenomena.Presupposition.Studies.DegenTonhauser2022.TraditionalClass :

Traditional classification of clause-embedding predicates. This classification is challenged by the experimental results.

factive : TraditionalClass
Canonically factive: CC traditionally assumed presupposed + entailed
nonveridicalNonfactive : TraditionalClass
Nonveridical nonfactive: CC neither presupposed nor entailed
veridicalNonfactive : TraditionalClass
Veridical nonfactive: CC entailed but not presupposed
optionallyFactive : TraditionalClass
Optionally factive: CC may or may not be presupposed

Instances For

instance Phenomena.Presupposition.Studies.DegenTonhauser2022.instDecidableEqTraditionalClass :

DecidableEq TraditionalClass

Equations

Phenomena.Presupposition.Studies.DegenTonhauser2022.instDecidableEqTraditionalClass x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯

instance Phenomena.Presupposition.Studies.DegenTonhauser2022.instReprTraditionalClass :

Repr TraditionalClass

Equations

Phenomena.Presupposition.Studies.DegenTonhauser2022.instReprTraditionalClass = { reprPrec := Phenomena.Presupposition.Studies.DegenTonhauser2022.instReprTraditionalClass.repr }

def Phenomena.Presupposition.Studies.DegenTonhauser2022.instReprTraditionalClass.repr :

TraditionalClass → ℕ → Std.Format

Equations

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

Instances For

def Phenomena.Presupposition.Studies.DegenTonhauser2022.traditionalClass :

DegenTonhauser2021.Predicate → TraditionalClass

Traditional classification of each predicate.

Equations

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

Instances For

def Phenomena.Presupposition.Studies.DegenTonhauser2022.projectionRating_Exp1a :

DegenTonhauser2021.Predicate → Float

Mean certainty ratings from Experiment 1a (gradient scale 0-1). Higher = more projective (speaker more certain of CC).

Values computed from the data at github.com/judith-tonhauser/projective-probability (results/5-projectivity-no-fact), rounded to 2 decimal places. The gradient nature and ordering show no categorical factive/nonfactive gap.

Equations

Instances For

def Phenomena.Presupposition.Studies.DegenTonhauser2022.projectionRating_Exp1b :

DegenTonhauser2021.Predicate → Float

Proportion of 'yes' responses from Experiment 1b (binary choice).

Values computed from the data at github.com/judith-tonhauser/projective-probability (results/8-projectivity-no-fact-binary), rounded to 2 decimal places.

Equations

Instances For

def Phenomena.Presupposition.Studies.DegenTonhauser2022.inferenceRating_Exp2a :

DegenTonhauser2021.Predicate → Float

Mean inference ratings from Experiment 2a (gradient scale 0-1). Higher = inference to CC more strongly supported.

Values computed from the data at github.com/judith-tonhauser/projective-probability (results/4-veridicality3), rounded to 2 decimal places.

Equations

Instances For

theorem Phenomena.Presupposition.Studies.DegenTonhauser2022.optionally_factive_as_projective_as_factive :

projectionRating_Exp1a DegenTonhauser2021.Predicate.inform > projectionRating_Exp1a DegenTonhauser2021.Predicate.reveal ∧ projectionRating_Exp1a DegenTonhauser2021.Predicate.acknowledge > projectionRating_Exp1a DegenTonhauser2021.Predicate.reveal

Optionally factive predicates can be as projective as canonically factive ones.

Inform projects more strongly than reveal (a canonical factive): inform=0.81 vs reveal=0.70 (Exp 1a).

theorem Phenomena.Presupposition.Studies.DegenTonhauser2022.no_categorical_projection_gap :

projectionRating_Exp1a DegenTonhauser2021.Predicate.inform > projectionRating_Exp1a DegenTonhauser2021.Predicate.reveal

There is no categorical gap between factive and optionally factive predicates in projection.

The mean projection rating of the least projective factive (reveal: 0.70) is lower than the most projective optionally factive (inform: 0.81).

theorem Phenomena.Presupposition.Studies.DegenTonhauser2022.entailment_projection_dissociation :

inferenceRating_Exp2a DegenTonhauser2021.Predicate.beRight > inferenceRating_Exp2a DegenTonhauser2021.Predicate.know ∧ projectionRating_Exp1a DegenTonhauser2021.Predicate.know > projectionRating_Exp1a DegenTonhauser2021.Predicate.beRight

Predicates with highest entailment have lowest projection.

be_right: inference=0.96, projection=0.18 know: inference=0.93, projection=0.86

This dissociation between inference (entailment) strength and projection strength is one of the key empirical observations of the paper, undermining Definition 3b.

theorem Phenomena.Presupposition.Studies.DegenTonhauser2022.top_two_agree :

The two projection diagnostics (Exp 1a continuous, Exp 1b binary) agree on the two most projective predicates: be_annoyed and know.

theorem Phenomena.Presupposition.Studies.DegenTonhauser2022.binary_sharper_separation :

The binary diagnostic produces sharper separation: nonfactive predicates cluster near 0 in binary (< 0.10) but are above 0.15 in continuous.

Canonically factive verbs have factivePresup = true in the Fragment, matching D&T 2022's traditional classification. "be annoyed" is copular and emotive — its presupposition derives from emotive semantics, not doxastic veridicality, so factivePresup = false.

Nonveridical nonfactive predicates have factivePresup = false in the Fragment, matching D&T 2022's traditional classification.

theorem Phenomena.Presupposition.Studies.DegenTonhauser2022.traditionalClass_consistent_with_fragment (p : DegenTonhauser2021.Predicate) (v : Fragments.English.Predicates.Verbal.VerbEntry) (h : DegenTonhauser2021.toVerbEntry p = some v) :

(traditionalClass p = TraditionalClass.factive → v.factivePresup = true) ∧ (traditionalClass p = TraditionalClass.nonveridicalNonfactive → v.factivePresup = false)

The traditional classification is consistent with Fragment factivity for verbal entries: every verb classified as factive has factivePresup = true, every nonfactive has false.

theorem Phenomena.Presupposition.Studies.DegenTonhauser2022.copular_presup_matches_classification :

(DegenTonhauser2021.toPredicateCore DegenTonhauser2021.Predicate.beAnnoyed).isPresupTrigger = true ∧ (DegenTonhauser2021.toPredicateCore DegenTonhauser2021.Predicate.beRight).isPresupTrigger = false

"be annoyed" is a presupposition trigger (emotive factive), while "be right" is not. This matches D&T 2022's traditional classification: factive predicates trigger presuppositions, veridical nonfactives do not.