Silence as a low-prior alternative #
@cite{bergen-levy-goodman-2016}
WithSilence U := Option U is the standard wrapper that adds a "silence"
element to any utterance type. none represents silence; some u represents
a paper-utterance.
Following @cite{bergen-levy-goodman-2016}, silence has universal extension
(liftMeaning makes it true at every world), giving it the smallest lex
value (1/card W) under the standard extensionOf-based literal listener.
This disfavors silence at the softmax when stronger utterances are honest.
At observations where no non-silent utterance is honest, silence absorbs
all probability — dissolving the "no qOk witness" defect that would
otherwise make (access, word) ∉ {sensible} cases vacuous in PMF
formalizations of @cite{goodman-stuhlmuller-2013}-style models.
Main definitions #
WithSilence U—Option U, wherenone= silence.liftMeaning m— extendsm : U → W → BooltoWithSilence U → W → Bool, with silence true at every world.
Per-paper specialisations #
The "cover hypothesis is universally satisfied because silence is always
qOk" theorem is paper-specific (because qOk depends on the paper's
observation type and compatibility predicate). Each consumer paper proves
its own cover_silent as a 1-line application of liftMeaning_none.
Silence-extended utterance type. some u is a paper-utterance;
none is the "say nothing" alternative.
Definitionally Option U so it inherits all standard instances
(DecidableEq, Fintype, Repr, Inhabited) without manual derivation.
Equations
- RSA.WithSilence U = Option U
Instances For
Lift a meaning function to handle silence. Silence is "vacuously honest" — true at every world (commits to nothing).
This is the canonical lifting that makes silence the universally-qOk fallback: at any observation, the only worlds the speaker considers are those compatible with the obs, and silence is true at all of them trivially.
Equations
- RSA.liftMeaning m (some u) x✝ = m u x✝
- RSA.liftMeaning m none x✝ = true