Giorgolo & Asudeh 2012: Monads for Conventional Implicatures #
@cite{giorgolo-asudeh-2012}
Core Claim #
Conventional implicatures are modeled as Writer monad side-effects.
CI-contributing expressions (appositives, expressives) log propositions
to a side-issue dimension via write, while presupposition triggers
log conditions via check. The monadic type structure enforces
@cite{potts-2005}'s flow restriction by construction: bind's function
argument receives only the at-issue value, never the CI log.
Two-Stage Architecture #
- Compositional phase: at-issue and CI dimensions are separated; Potts's flow restrictions hold (at-issue → CI is one-way).
- Post-compositional phase: anaphora resolution and presupposition checking can freely access both dimensions.
Two Channels via Monad Transformer (Appendix A) #
The analysis requires two Writer monads combined via a monad transformer:
- Inner Writer (CI): accumulates conventional implicature propositions
- Outer Writer (Presupposition): accumulates presuppositional conditions
write and check have the same definition λt.⟨⊥, {t}⟩ (eq. 21)
but operate in different monad layers (fn. 4).
Worked Example (§5): "John, who likes cats, likes dogs also." #
- At-issue: like(john, dogs)
- CI: {like(john, cats)} — from the NRRC via
write - Presupposition: {∃z. like(john, z) ∧ z ≠ dogs} — from "also" via
check
The presupposition is satisfied by the CI content: john likes cats, and cats ≠ dogs. This satisfaction happens post-compositionally, after both Writer logs are exposed.
Relation to @cite{shan-2001} #
@cite{shan-2001} showed monads capture deep structure in NL semantics (focus, scope, questions, binding). @cite{giorgolo-asudeh-2012} apply the Writer monad specifically to CIs, arguing it is preferable to the continuation-based approach: each monad isolates one kind of side-effect, and monad transformers compose them modularly.
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Phenomena.Presupposition.Studies.GiorgoloAsudeh2012.like Phenomena.Presupposition.Studies.GiorgoloAsudeh2012.E.john Phenomena.Presupposition.Studies.GiorgoloAsudeh2012.E.cats = true
- Phenomena.Presupposition.Studies.GiorgoloAsudeh2012.like Phenomena.Presupposition.Studies.GiorgoloAsudeh2012.E.john Phenomena.Presupposition.Studies.GiorgoloAsudeh2012.E.dogs = true
- Phenomena.Presupposition.Studies.GiorgoloAsudeh2012.like x y = false
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.
- Phenomena.Presupposition.Studies.GiorgoloAsudeh2012.instBEqCIProp.beq x✝¹ x✝ = false
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Presuppositional conditions: unevaluated conditions logged by check.
- existsOtherLiked : E → E → PresupProp
∃z. like(subj, z) ∧ z ≠ obj — the presupposition of "also"
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.
- Phenomena.Presupposition.Studies.GiorgoloAsudeh2012.instBEqPresupProp.beq x✝¹ x✝ = false
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Evaluate a presuppositional condition over the finite entity domain.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Two-channel meaning: flattened Writer Presup (Writer CI A).
The outer Writer carries presuppositions; the inner carries
conventional implicatures. Figure 1's result type is:
⟨⟨at-issue, {CI-props}⟩, {presup-conditions}⟩
Instances For
Equations
Instances For
Equations
Instances For
write(t) = ⟨⊥, {t}⟩ (eq. 21): log a proposition to the CI channel.
Equations
Instances For
check(t) = lift(⟨⊥, {t}⟩) (eq. 21, fn. 4): log a condition to the
presupposition channel. Same definition as write, different layer.
Equations
Instances For
Equations
Instances For
Equations
Instances For
All lexical items not introducing CIs or presuppositions are
η-lifted: ⟦word⟧ = η(standard-meaning) (Table 1).
comma (Table 1): λj λl. j ⋆ λx. l ⋆ λf. write(f x) ⋆ λ_. η(x)
The prosodic comma introduces NRRC content as CI via write.
Both arguments are monadic (⊸* in Glue), matching Table 1's type
j ⊸* (j ⊸ l) ⊸* j.
Equations
- One or more equations did not get rendered due to their size.
Instances For
also (Table 1):
λv.λo.λs. s ⋆ λx. v ⋆ λf. o ⋆ λy. check(∃z. f z x ∧ z ≠ y) ⋆ λ_. η(f y x)
Takes verb, object, subject (all monadic per ⊸*). Checks the
presupposition that the subject verb-s something other than the
object, then returns the at-issue content f y x.
Equations
- One or more equations did not get rendered due to their size.
Instances For
"John, who likes cats" — comma writes like(john, cats) to CI.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Full sentence (example 20):
also(likes)(dogs)(john_who_likes_cats)
Equations
- One or more equations did not get rendered due to their size.
Instances For
Figure 1's result:
⟨⟨like(j, dogs), {like(j, cats)}⟩, {∃z.like(j,z) ∧ z ≠ dogs}⟩
Post-compositionally, both logs are exposed. CI content and presuppositional conditions are evaluated against the model.
All CI content is true in the model.
All presuppositions are satisfied in the model.
The CI log provides the witness for presupposition satisfaction.
The NRRC logs like(john, cats). Since like(john, cats) = true and
cats ≠ dogs, the presupposition ∃z. like(john, z) ∧ z ≠ dogs is
witnessed. This is the paper's central empirical point: the
presupposition of "also" is satisfied by CI content from the NRRC,
but this satisfaction is only computable post-compositionally
(the log produced by write cannot be examined before the monadic
computation terminates).
Why the Writer monad enforces dimensional separation #
The function in bind has type A → TwoChannel CI Presup B, not
TwoChannel CI Presup A → TwoChannel CI Presup B. It receives the
value stripped of both logs. This means:
At-issue → CI (allowed): The comma operator receives
john(the value) and uses it to construct CI contentlike(john, cats).CI → at-issue (blocked): When
alsoapplies the verb, the function receives onlyx = john,f = likes,y = dogs. It cannot see that the CI log containslike(john, cats).Presup → at-issue (blocked): Similarly, presuppositional conditions logged by
checkare invisible to subsequent at-issue computation.
This structural separation IS Potts's restriction, enforced by the type system rather than stipulated as a constraint on derivations.
Changing the NRRC content does not affect the at-issue result: the main clause function sees only the value, never the CI log.
Equations
- One or more equations did not get rendered due to their size.
Instances For
CI-contributing expressions modeled by the Writer monad — expressives, appositives, NRRCs — are exactly Class B in @cite{tonhauser-beaver-roberts-simons-2013}'s taxonomy: SCF=no (CI content can be informative), OLE=no (attributed to speaker, not attitude holder). The Writer's log threading captures this: content projects past all operators without requiring prior establishment in context.
Class B = SCF=no, OLE=no: the behavior the Writer monad models.
For intensional models where values and log entries are world-indexed
propositions, the CI channel maps directly to @cite{potts-2005}'s
TwoDimProp. The presupposition channel is orthogonal.
Project the CI channel to a TwoDimProp.
The at-issue value becomes atIssue; the conjoined CI log becomes ci.
Presuppositional content is discarded (it lives in a separate dimension).
Equations
Instances For
Structural correspondence to PostSupp #
PostSupp S A (@cite{charlow-2021}) is structurally identical to a single
Writer monad: a value paired with accumulated side-effect content, composed
via pure/bind with log sequencing via dseq. The Writer monad for CIs
and Charlow's PostSupp for modified numerals are the same pattern applied
to different side-effects (CI propositions vs cardinality tests), confirming
@cite{shan-2001}'s insight that monads capture recurring compositional
structure in natural language.