Meyer and Feiman (2021): Priming Reveals Similarities and Differences Between Implicatures #
This file formalizes the implicature spectrum that [MF21] propose to reconcile
their structural-priming results. The core implicature mechanism has two sub-computations,
generating alternatives and negating them, and each may run online or have its output
stored; alternatives generated online cannot have been pre-negated, so the spectrum has
three points (Spectrum, spectrum_iff): both steps online, the alternatives stored and
negated online, or the enriched reading stored as a second lexical entry. Two readings
prime each other when they share an online sub-computation that can be primed (Primes).
Three experiments find that some and number words prime each other while free-choice
disjunction primes neither; the paper places some at the fully online point and number
words at the stored-alternatives point, so the shared computation behind the effect is the
negation of alternatives (negate_primeable), and the free-choice data then rule out the
recursive-implicature derivation of free choice, which negates alternatives, while
admitting the alternative-asserting mechanism of [BLF17] exactly when generation
cannot be primed and modal-specific accounts such as [Sim05] unconditionally
(fits_iff).
Implementation notes #
A derivation is represented by the finite set of sub-computations it runs online; the
paper's hypothesis for some and number words with free choice left open is spectrum,
and Fits compares an account's predicted between-category priming with the observed
pattern. Within-category priming, the higher enrichment rate for number words, and the
experiments' picture-similarity controls are not represented.
References #
Equations
- MeyerFeiman2021.instDecidableEqStep x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
A point on the implicature spectrum: the sub-computations of the core mechanism that run online, the others having their output stored, where alternatives generated online cannot have been pre-negated.
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
The spectrum has three points: both steps online, stored alternatives negated online, and a stored enriched reading.
Two derivations prime each other when some sub-computation that can be primed runs online in both.
Equations
- MeyerFeiman2021.Primes primeable o o' = ∃ s ∈ primeable, s ∈ o ∧ s ∈ o'
Instances For
Equations
- MeyerFeiman2021.instDecidableEqCategory x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Between-category priming observed: some and number words prime each other in both directions, and free-choice disjunction primes neither and is primed by neither.
Equations
Instances For
An account assigns each category the sub-computations its enriched reading runs online; it fits the data when it predicts between-category priming exactly where observed.
Equations
- MeyerFeiman2021.Fits primeable account = ∀ (a b : MeyerFeiman2021.Category), a ≠ b → (MeyerFeiman2021.primed a b = true ↔ MeyerFeiman2021.Primes primeable (account a) (account b))
Instances For
The paper's hypothesis, with the free-choice derivation fc left open: some generates
and negates its alternatives online, and number words store their alternatives, the count
list, and negate them online.
Equations
Instances For
The only sub-computation online for both some and number words is negation, so the priming between them shows that negating alternatives can be primed.
The hypothesis fits the data exactly when negation can be primed and the free-choice derivation shares no primeable sub-computation with some.
The recursive-implicature derivation of free choice ([Fox07]) negates alternatives, so it shares the primeable computation with some and number words and cannot fit the data.
The alternative-asserting mechanism of [BLF17] shares only generation with some, so it fits the data exactly when generation cannot be primed.
A mechanism specific to modals and disjunction ([Sim05]) runs no implicature sub-computation, so it fits the data whenever negation can be primed.