@cite{denic-homer-rothschild-chemla-2021} #
The influence of polarity items on inferential judgments. Cognition 215, 104791.
The puzzle #
The standard story is unidirectional: monotonicity determines where PIs are grammatical. If the arrow goes only from monotonicity to PIs, then PI presence should be inert — it shouldn't change how listeners perceive the monotonicity of the environment. This paper shows it does.
Design #
Four experiments (§3–§7):
- Environments: UE, DE, NM (Exps 1–4); DN added in Exps 3–4
- PI conditions: NPI (any/ever/at all), PPI (some), no-PI
Participants rated sentence pairs (related by a narrowing substitution) for both UE-ness and DE-ness on 7-point scales. The directional rating (UE − DE) gives a single bipolar measure.
Core findings #
- NPI in NM: directional ratings shift toward DE (significant across all exps)
- PPI in DN: directional ratings shift toward UE (significant in Exp 3)
- UE and DE: no PI effects (ceiling/floor)
- Scalar side-effect not the sole mechanism: the NPI effect holds across NPIs with heterogeneous scalar properties (e.g., "ever" lacks domain widening; §10.2)
Significance for linglib #
The finding that PIs distinguish DN from UE challenges ContextPolarity,
which maps both to .upward. The full EntailmentSig preserves the
distinction (DN = addMult, simple UE = mono), suggesting PI-sensitive
code should use EntailmentSig paths rather than ContextPolarity.
The four monotonicity environments tested.
A refinement of Core.NaturalLogic.ContextPolarity that keeps DN distinct
from UE. ContextPolarity maps both to .upward; the paper shows PIs
distinguish them empirically.
- UE : MonotonicityEnv
- DE : MonotonicityEnv
- NM : MonotonicityEnv
- DN : MonotonicityEnv
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Whether an environment is globally upward-entailing. DN and UE both return true — the crux of the coarsening problem.
Equations
- Phenomena.Polarity.Studies.DenicEtAl2021.MonotonicityEnv.UE.isGloballyUE = true
- Phenomena.Polarity.Studies.DenicEtAl2021.MonotonicityEnv.DN.isGloballyUE = true
- Phenomena.Polarity.Studies.DenicEtAl2021.MonotonicityEnv.DE.isGloballyUE = false
- Phenomena.Polarity.Studies.DenicEtAl2021.MonotonicityEnv.NM.isGloballyUE = false
Instances For
The four PI conditions in the experimental design.
- npi : PICondition
- ppi : PICondition
- noPISimple : PICondition
- noPIComplex : PICondition
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
A finding about PI influence on directional ratings within an environment. The key observable: does PI presence significantly shift ratings relative to the no-PI baseline, and in which direction?
- experiment : ℕ
- environment : MonotonicityEnv
- piCondition : PICondition
- significant : Bool
Significant difference from no-PI baseline?
- shiftDirection : Option MonotonicityEnv
- evidence : String
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
NPI in NM: shifts directional ratings toward DE.
Equations
- One or more equations did not get rendered due to their size.
Instances For
PPI in DN: shifts directional ratings toward UE (Exp 3, §6.3).
Equations
- One or more equations did not get rendered due to their size.
Instances For
NPI in UE: no effect (ceiling).
Equations
- One or more equations did not get rendered due to their size.
Instances For
NPI in DE: no effect (floor).
Equations
- One or more equations did not get rendered due to their size.
Instances For
PPI in UE: no effect.
Equations
- One or more equations did not get rendered due to their size.
Instances For
PPI in DE: no effect.
Equations
- One or more equations did not get rendered due to their size.
Instances For
NPI in DN: no significant effect (Exp 3, §6.3).
Equations
- One or more equations did not get rendered due to their size.
Instances For
PPI in NM: no significant effect.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
NPI effect in NM replicates in Experiment 3 (§6.3). The NPI effect holds across NPIs with heterogeneous scalar properties (any = domain widening, ever = non-domain-widening), which argues against scalar side-effects as the sole mechanism (§10.2).
Equations
- One or more equations did not get rendered due to their size.
Instances For
The complementary pattern: each PI type affects exactly the environment where its associated polarity is ambiguous.
- NPIs (associated with DE) affect NM but not DN
- PPIs (associated with UE) affect DN but not NM
- npi_affects_NM : Bool
- npi_affects_DN : Bool
- ppi_affects_NM : Bool
- ppi_affects_DN : Bool
- complementary : self.npi_affects_NM = true ∧ self.npi_affects_DN = false ∧ self.ppi_affects_NM = false ∧ self.ppi_affects_DN = true
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Instances For
Equations
Instances For
Map the paper's environments to canonical entailment signatures.
Equations
- Phenomena.Polarity.Studies.DenicEtAl2021.envSignature Phenomena.Polarity.Studies.DenicEtAl2021.MonotonicityEnv.UE = some Core.NaturalLogic.EntailmentSig.mono
- Phenomena.Polarity.Studies.DenicEtAl2021.envSignature Phenomena.Polarity.Studies.DenicEtAl2021.MonotonicityEnv.DE = some Core.NaturalLogic.EntailmentSig.antiAdd
- Phenomena.Polarity.Studies.DenicEtAl2021.envSignature Phenomena.Polarity.Studies.DenicEtAl2021.MonotonicityEnv.DN = some Core.NaturalLogic.EntailmentSig.addMult
- Phenomena.Polarity.Studies.DenicEtAl2021.envSignature Phenomena.Polarity.Studies.DenicEtAl2021.MonotonicityEnv.NM = none
Instances For
The DE environment maps to a DE-side signature.
isGloballyUE is derived from the signature map, not stipulated.
Any composition of two DE-side signatures is UE under ContextPolarity.
ContextPolarity cannot distinguish ANY doubly-negative environment
from simple UE.
EntailmentSig preserves DN–UE distinctions that ContextPolarity erases.
UE strength distinguishes DN from UE.
DN has no DE strength.
Central theorem: PPI effect reveals a distinction that
ContextPolarity provably cannot make.
NPI effects target environments with no determinate signature.
PPI effects target environments with composite UE signatures.
The NPIs tested are classified as NPIs in the Fragment lexicon.
The PPI tested is classified as a PPI in the Fragment lexicon.
The tested NPIs are all strengthening (@cite{israel-2001} Figure 1) but have heterogeneous semantic bases — the mechanism isn't purely about domain widening (§10.2).
Mechanism constraints: effect persists across semantically heterogeneous NPIs (different base forces) and without domain widening. The heterogeneity that matters for the non-scalar argument is in baseForce (existential vs temporal vs degree), not in scalar direction — all three are strengthening per @cite{israel-2001}.
Predict NPI significance from signature.
Equations
- Phenomena.Polarity.Studies.DenicEtAl2021.predictNPIEffect env = match Phenomena.Polarity.Studies.DenicEtAl2021.envSignature env with | none => true | some val => false
Instances For
Predict PPI significance from signature.
Equations
- One or more equations did not get rendered due to their size.
Instances For
NPI prediction matches all 4 experimental environments.
PPI prediction matches all 4 experimental environments.
Algebraic predictions agree with all 8 empirical data points.
Observed pattern matches signature-based predictions.
Shift directions match PI polarity.
Significant findings' environments match predicted environments.
Mechanism verdict linked to data file.
NPI items match Fragment lexicon entries by surface form.