Griceans claim that SIs are a type of pragmatic inferences called implicatures that involve scales, e.g. ⟨or, and⟩, ⟨some, (most,) all⟩, ⟨possible, certain⟩
Some neo-Griceans (e.g., Geurts 2010) use the term quantity implicature
For grammaticalists (e.g., Chierchia, Fox & Spector 2012), SIs are technically not 'implicatures', but they still nevertheless use the term 'scalar implicature'
2.1 Primary vs. secondary implicatures
2.2 The Symmetry Problem
2.3 Embedded scalar implicatures
Not all meaning is encoded in linguisitc expressions
One naturally concludes from (1A) that the speaker does not speak Korean.
But this inference is arguably not part of the meaning of the linguistic expressions used
We often conclude more than what is said.
Paul Grice pioneered the study of pragmatic inferences
Idea: With the assumption that the speaker is cooperative, one may draw extra inferences based on their (linguistic or non-linguistic) behaviour
Grice called such pragmatic inferences implicatures
It's reasonable to assume that conversational agents are generally cooperative
Implicatures arise in non-linguistic communication as well
What does it mean to be a cooperative conversational agent?
Grice's (tentative) answer = To follow the four Griceran maxims:
A cooperative agent's behavior that seems to flout a maxim must have a reason, e.g.
In the Gricean approach to SIs, the Maxim of Quantity plays a crucial role
The Maxim of Quantity is about informativity
(Neo-)Griceans claim that SIs are implicatures, a type of pragmatic inference generated via reasoning based on cooperativity
Recipe for SI:
According to this theory, an SI arises via counterfactual reasoning about a hypothetical utterance of a more informative alternative
Since the early 2000s, SIs have been investigated experimentally, as well as theoretically (see Chemla & Singh 2014a,b, Noveck 2018 for overviews)
Some relevant findings
Short = 900 ms; Long = 3000 ms
Memory task sandwitching sentence verification (à la Bott & Noveck 2004):
Results (N = 56)
Gricean recipe: Upon hearing φ (e.g., "The linguist went to Italy or France"), Hearer reasons uttering the stronger alternative ψ (e.g. "The linguistic went to Italy and France) should have violated the Maxim of Quality
But the preceived SI is typically stronger than this: Speaker has evidence for the falsity of ψ (secondary implicature)
Neo-Griceans (Sauerland 2004, Spector 2006, etc.) assume that a primary implicature gets strengthened to a secondary implicature via an auxiliary assumption, Opinionatedness (alt. Experthood, Competence)
1. + 3.
But some experimental evidence that secondary SIs are drawn even if Opinionatedness does not hold (i.e., the speaker is known to be uncertain)
Conditions
Scalars: some, almost, two, PLURAL
Classical Gricean theories are naive about alternatives
The crucial alternative is formed with "all", but one could think of another alternative formed with "some but not all" or "only some"
These alternatives are called symmetric alternatives
Symmetric alternatives cannot be both negated consistently, while maintaining the truth of the prejacent
One might think that the symmetry between "all" and "some but not all" can be broken by the difference in structural complexity (see Katzir 2007)
But not all symmetry problems can be explained by structural complexity (Romoli 2013, Trinh & Haida 2015, Breheny et al. 2018)
Gricean derivations of SIs only apply at the utterance level
But SIs seem to be able to take scope under logical operators (Chierchia 2004, Chierchia, Fox & Spector 2012, etc.)
Geurts & Pouscoulous 2009 failed to observe evidence for embedded SIs
Chemla & Spector 2011 found some evidence
Debate: Geurts & Van Tiel 2013, Cummins 2014, Van Tiel 2014, Potts, Lassiter, Levy & Frank 2014, Franke, Schlotterbeck & Augurzky 2017, Van Tiel, Noveck & Kissine 2018
Difficulties
"Every letter is connected with some of its circles" (in French)
Literal reading (true in L, W, S):
∀x(Lx → ∃y(●y ∧ Cxy))
Global reading (true in W, S):
(∀x(Lx → ∃y(●y ∧ Cxy)) ∧ ¬∀x(Lx → ∀z(●z → Cxz))
Local reading (true in S):
∀x(Lx → (∃y(●y ∧ Cxy) ∧ ¬∀z(●y → Cxz)))
Chemla & Spector 2011 claim the difference between Weak and Strong as suggesting that there is a local SI reaidng ('the more readings are true, the truer the sentence sounds')
E.g., the global reading is true and the local reading false, when 3 students flunked some but not all, 1 flunked all
Note that these sentences might lack the global reading altogether
NB: The global and local readings amount to the same thing for 'exactly one'
The grammatical approach to SI was developed mainly to deal with embedded SIs
SIs are triggered by a phonologically null operator Exh (alt:
"Every linguist met some of the philosophers"
Exh(Every linguist met some of the philosophers)
Every linguist λx Exh(x met some of the philosophers)
Every linguist is such that
The disjunct-alts are used to derive ignorance inferences (= primary implicatures)
They also sometimes give rise to SIs
The disjunctive-alts are stronger (and hence non-weaker) than the prejacent, but negating both of them independently will contradict the prejacent
Fox 2007 puts forward innocent exclusion: Negate as many alternatives as possible, while maintaining consistency with the prejacent
E.g.
In most theories of SI, generation of a scalar implicature involves:
Theories mostly differ with respect to the negation mechanism
Every theory needs a recipe for determining alternatives
Experimental topics
A)
B)
C)
Theoretical topics
D)
E)
F)