Quantity implicatures

Quantity implicatures
Overview of the course

1 Quantity implicatures
2 Nonce inferences or defaults?
3 “Embedded” implicatures
4 Free choice permission
5 Non-pragmatic alternatives

Quantity implicatures
Overview of the course

1 Quantity implicatures
Grice
Quantity implicatures
Scalar implicatures
Disjunction
2 Nonce inferences or defaults?
3 “Embedded” implicatures
4 Free choice permission
5 Non-pragmatic alternatives

Quantity implicatures
 Grice

Quantity implicatures
Quantity implicatures
Grice on conversational implicatures

Philosophical background.
Not everything that is conveyed by way of a linguistic
utterance φ is part of φ’s meaning.
An important part of the information conveyed is inferred
on the assumption that the speaker seeks to be cooperative.
Example:
A: I am out of petrol.
B: There is a garage round the corner.
; For all B knows, the garage is open and has petrol to sell.
Such inferences Grice calls “conversational implicatures”.
This is not a psychological theory. Rather, Grice is
concerned primarily with speakers’ commitments.

Quantity implicatures
Inference to the best explanation

Conversational implicatures are not deductive inferences.

Rather, they are inferences to the best explanation (aka
abductive inferences). Cf. diagnostic reasoning.
Even if much of the literature suggests otherwise, there
couldn’t be a mechanical procedure for computing
conversational implicatures.

Quantity implicatures
Conversational implicatures are:

derived from:
an utterance made by a speaker S (“what S has said”)
+ the assumption that S is trying to be cooperative.
cancellable:
B: There is a garage around the corner, but I’m not sure it is
open.
non-linguistic: they are licensed by general principles of
rational cooperative behaviour.
Methodological corollary:
+ All else being equal, an explanation in terms of
conversational implicature is better than any alternative
explanation in linguistic terms.

Quantity implicatures
Cancellability

Cancellability is a useful heuristic for testing if a given

inference might be a conversational implicature.
But #1: Conversational implicatures aren’t the only kind
of cancellable inferences:

(1) Harry doesn’t know he got promoted.

(2) Harry doesn’t know he got promoted, because he
didn’t: it was just a rumour.

But #2: “Cancellability” is a bit of a misnomer.

Quantity implicatures
Implicatures aren’t really cancellable

Explicit cancellation:

(3) There is a garage round the corner, but it’s closed.

It’s not at all clear that an implicature is cancelled in this case

(certainly not from the speaker’s point of view).

Implicit cancellation:

(4) [During a treasure hunt:] The prize is either in the garden

or in the attic.

Here it’s probably better to say that the implicature never arose
in the first place.

Quantity implicatures
Summing up

Conversational implicatures are:

pragmatic inferences that aren’t encoded, but rather flow
from universal principles of cooperative behaviour
inferences to the best explanation, i.e. abductive rather
than deductive
“cancellable”

Quantity implicatures
Quantity implicatures

Quantity implicatures
Quantity implicatures

 Q-implicatures are a kind of conversational implicature.

 Q-implicatures are inferred on the basis of the assumption that
the speaker is trying to make his utterance sufficiently informa-
tive “for the current purposes of the exchange.”
↑
tricky bit

Quantity implicatures
Varieties of Q-implicature

Exhaustivity effects in question answering

A: Who did you see?
B: Fred.
; The speaker saw only Fred.
Scalar implicatures
Fred ate some of the figs.
; Fred didn’t eat all the figs.
Conditional perfection
If you scratch my back, I’ll give you a kiss.
; Only if you scratch my back will I give you a kiss.
Free choice permission
You can have an apple or a pear.
; You can have an apple.
; You can have a pear.

Quantity implicatures
The Standard Recipe for deriving Q-implicatures

(1) A: Who did you see?

(2) B: Fred.

1 Rather than saying (2), B could have said:

(2*) Fred and . . .
Why didn’t she do so?
2 The most likely explanation is that B doesn’t believe that
(2*) is true: ¬BelB (2*).
3 B is likely to have an opinion as to whether (2*) is true:
BelB (2*) ∨ BelB ¬(2*).
4 Thus, it follows that BelB ¬(2*): B believes that she saw
only Fred.

Quantity implicatures
The Standard Recipe for deriving Q-implicatures

(1) A: Who did you see?

(2) B: Fred.

1 Rather than saying (2), B could have said:

(2*) Fred and . . .
Why didn’t she do so?
2 The most likely explanation is that B doesn’t believe that
(2*) is true: ¬BelB (2*). ← primary/weak implicature
3 B is likely to have an opinion as to whether (2*) is true:
BelB (2*) ∨ BelB ¬(2*). ← Bivalence Assumption
4 Thus, it follows that BelB ¬(2*): B believes that she saw
only Fred. ← secondary/strong implicature

Quantity implicatures
The Standard Recipe for deriving Q-implicatures

(3) Fred ate some of the figs.

1 Rather than saying (3), S could have said:

(3*) Fred ate all the figs.
Why didn’t he do so?
2 The most likely explanation is that S doesn’t believe that
(3*) is true: ¬BelS (3*). ← primary/weak implicature
3 S is likely to have an opinion as to whether (3*) is true:
BelS (3*) ∨ BelS ¬(3*). ← Bivalence Assumption
4 Thus, it follows that BelS ¬(3*): S believes that Fred didn’t
eat all the figs. ← secondary/strong implicature
+ Ergo: “some” doesn’t entail “not all”.

Quantity implicatures
Bivalence, Competence, and No Opinion

Weak Q-implicatures are always of the form ¬BelS (φ).

Strong Q-implicatures are always of the form BelS (¬φ).
The Bivalence Assumption is always of the form
BelS (φ) ∨ BelS (¬φ).
S is competent about φ iff the following hold:
– if φ then BelS (φ)
– if ¬φ then BelS (¬φ)
Competence entails Bivalence but not v.v.
If S has No Opinion about φ, then ¬BelS (φ) ∧ ¬BelS (¬φ).
Weak implicatures express less than No Opinion.
No Opinion is the negation of Bivalence.

Quantity implicatures
Bivalence, Competence, and No Opinion

non-belief
¬BelS (φ)
z }| {
BelS (φ) BelS (¬φ) ¬BelS (φ) ∧ ¬BelS (¬φ)
belief disbelief no opinion
| {z }
BelS (φ) ∨ BelS (¬φ)
bivalence

Quantity implicatures
Weak implicatures aren’t always strenghtened

(4) Bonnie had an apple or a pear.

Implicatures:

(5) a. ¬BelS (Bonnie had an apple)

b. ¬BelS (Bonnie had an pear)

But since the Bivalence Assumption doesn’t hold in this case

(why not?), we don’t get:

(6) a. BelS ¬(Bonnie had an apple)

b. BelS ¬(Bonnie had an pear)

Quantity implicatures
Reminder: Q-implicatures are abductive inferences

A realises that the figs were injected with strychnine and

suspects that somebody may have eaten some:
(7) A: Did anyone eat of the figs?
B: Fred ate some.

Speaker is contemplating a priceless antique clock:

(8) What’s this thing doing on the mantlepiece?

Parent at the breakfast table:

(9) I see that somebody hasn’t eaten his porridge, yet.

+ In these cases, Q-implicatures are (at the very least) less

likely to arise: the Standard Recipe doesn’t apply here.

Quantity implicatures
Relevance

Simplistic notion of Q-implicature:

/ If S says φ and ψ is stronger than φ, then it is implicated

that ¬BelS (ψ).

This entails, e.g., that (10a) implicates ¬BelS (10b):

(10) a. Clyde is a dope.

b. Clyde is a dope and Jupiter is larger than Saturn.

We need some notion of relevance:

, If S says φ and ψ is relevant and stronger than φ, then it is

implicated that ¬BelS (ψ).

Quantity implicatures
But what is relevance?

This is (ahem) a difficult question, but here’s a first stab at an

answer:

Anything that is of potential interest to the hearer is

relevant.

How does the speaker decide what is of interest to the hearer?

There are at least 3 different ways:
S can assume that all discourse goals are of interest to H.
H may have personal interests that S knows about.
S can assume that everything of general interest is of
interest to H.

Quantity implicatures
Relevance I: discourse goals

In the context of a treasure hunt: Grice (1975)

(11) The prize is either in the garden or in the attic.

Or compare: Matsumoto (1995)

(12) A: What have you done with that mail?

B: I’ve typed it.
(13) A: Did you handwrite that letter?
B: No. I typed it.

Quantity implicatures
Relevance II: hearers’ personal interests

(14) A: I tried to call you yesterday. Where were you?

B: I was playing golf with a friend of mine.

B’s utterance is likely to implicate that his golf partner is

not among A and B’s mutual friends,
even if this information is not relevant to the immediate
purpose of the discourse.

Quantity implicatures
Relevance III: general interest

(15) A: Where is Wilma?

B: She went out to buy a piece of furniture.

B’s answer is likely to implicate that he doesn’t know (or

care) what kind of furniture Wilma went out to buy.
This information does not seem relevant to the immediate
purpose of the discourse, and it need not be relevant to A’s
personal interests.
(Though maybe it is assumed by fiat that anything of
general interest is of interest to the hearer).

Quantity implicatures
Relevance III: general interest (cont.)

It is possible that conventions (of some kind) decide what is the

minimum level of specificity that may be expected.

Cf.

(16) a. She went out to buy a piece of furniture.

b. She went out to buy a sofa.
(17) a. I saw an animal on the lawn this morning.
b. I saw a dog on the lawn this morning.

Quantity implicatures
Relevance III: general interest (cont.) Matsumoto (1995)

(18) Michio is Takashi’s brother.

6; S doesn’t know whether M. is older or younger than T.
(19) Kochira wa Takashi-kun no kyoodai no Michio-kun desu.
this top Takashi-Mr. gen brother gen Michio-Mr. cop
; S doesn’t know whether M. is older or younger than T.

Background: Unlike English, Japanese also has a system of

sibling terms that encode sex as well as seniority:

ani “older brother” otooto “younger brother”

ane “older sister” imooto “younger sister”

Apparently, there is a general expectation that, by default,

sibling terms are at least as specific as this.

Quantity implicatures
So, what’s relevance now?

Uuuuh . . .
Whatever relevance is, it is likely to be a multi-faceted
thing.
Perhaps, as I have suggested, relevance is hearer-centred.
With few exceptions, the R-issue is avoided in the
literature on Q-implicature.

Quantity implicatures
 Scalar implicatures

Quantity implicatures
Varieties of Q-implicature

Exhaustivity effects in question answering

A: Who did you see?
B: Fred.
; The speaker saw only Fred.
Scalar implicatures
Fred ate some of the figs.
; Fred didn’t eat all the figs.
Conditional perfection
If you scratch my back, I’ll give you a kiss.
; Only if you scratch my back will I give you a kiss.
Free choice permission
You can have an apple or a pear.
; You can have an apple.
; You can have a pear.

Quantity implicatures
Synopsis

Scalar implicatures have received more attention than any

other type of Q-implicature.
Theories of scalar implicature have been deeply influenced
by Horn’s (1972) dissertation work.
This has led to a somewhat distorted perception of
Q-implicatures.

Quantity implicatures
Horn scales

A Horn scale is a sequence of expressions ordered according

to strength:
hsome, many, most, alli
hclever, brillianti
hwarm, hot, scaldingi
...
Scales are used to generate, by means of simple
substitution, alternative sentences the speaker could have
uttered:
Let φ(α) be a sentence, where α is an expression in φ.
Then φ(β) is the same sentence, except that α is replaced
with β.
φ(β) is an alternative to φ(α) iff β occurs to the right of α
on some Horn scale.

Quantity implicatures
Horn scales: examples

1 Sentence: Wilma had many of the pears

Horn scale: hsome, many, most, alli
Alternatives:
Wilma had most of the pears
Wilma had all of the pears
Scalar implicatures:
BelS ¬(Wilma had most of the pears)
BelS ¬(Wilma had all of the pears)
2 Sentence: Barney’s essay is okay
Horn scale: hokay, brillianti
Alternative: Barney’s essay is brilliant
Scalar implicature: BelS ¬(Barney’s essay is brilliant)
3 Sentence: Betty had an apple or a pear
Horn scale: hor, andi
Alternative: Betty had an apple and a pear
Scalar implicature: BelS ¬(Betty had an apple and a pear)

Quantity implicatures
The generative method

One might call this the “generative” method for dealing

with scalar implicatures.
This method is much more problematic than it may seem:
It gives rise to pseudo-issues.
It doesn’t generalise to other Q-implicatures.
It may give a false impression of what Q-implicatures
are.

Quantity implicatures
Downward entailment

In a downward entailing (DE) environment, entailment relations

are reversed:

(1) a. Fred’s car is navy blue ⇒ Fred’s car is blue

b. Fred’s car isn’t navy blue ; Fred’s car is blue
c. Fred’s car isn’t blue ⇒ Fred’s car isn’t navy blue

; Negation creates a DE environment.

(2) a. All [cars] [are navy blue] ⇒ All [cars] [are blue]
b. All [blue cars] [are cheap] ⇒ All [navy blue cars] [are cheap]

; In a sentence of the form “All φ are ψ”, φ is a DE environment,

but ψ isn’t.

Quantity implicatures
Downward entailment and scalar implicatures

Since scalar implicatures are based on the relative strength of

scalar expressions, and downward entailment affects relative
strength, downward entailment and scalar implicature are
bound to interact.

(3) Fred read many of these novels

entailment: Fred read some of these novels
implicature: ¬(Fred read all of these novels)
(4) Fred didn’t read many of these novels
entailment: Fred didn’t read all of these novels
implicature: ¬(Fred didn’t read some of these novels)
= Fred read some of these novels

Quantity implicatures
“Scale reversal”

It is often said that Horn scales are reversed under

negation and in other DE environments.
This is wrong: it is entailment relations that reverse.
The assumption that scales are ordered in terms of
strength is redundant: alternatives are ordered in terms of
entailment anyway:
Wilma had all of the pears.
⇒ Wilma had most of the pears.
⇒ ⇒ Wilma had some of the pears.
+ The idea that scalar implicatures are generated from Horn
scales yields a distorted view of implicatures.

Quantity implicatures
Scalar implicatures with Horn sets

Replace Horn scales with Horn sets, like {all, some, most,
many}, {brilliant, clever}, etc.
Let φ(α) be a sentence, where α is an expression in φ.
φ(β) is an alternative to φ(α) iff
α, β ∈ H, where H is a Horn set, and
φ(β) is stronger than φ(α) (i.e. φ(β) entails φ(α) but not
vice versa).
Examples:
Sentence: Betty ate many of the tarts
Alternatives: Betty ate most of the tarts
Betty ate all of the tarts
Sentence: Betty didn’t eat many of the tarts
Alternatives: Betty didn’t eat some of the tarts

Quantity implicatures
Scalar implicatures with Horn sets

Lesson #1:
For the purpose of calculating Q-implicatures, strength is
measured at sentence level, not below.
Lesson #2:
We don’t need scales: The best way of implementing the
generative method uses sets rather than scales.

Quantity implicatures
Assorted worries about Horn scales/sets

Horn scales/sets may be viewed as way of dodging the

problem of relevance. Basically, this problem is stipulated
out of existence.
What are Horn scales/sets, and where do they come from?
The Horn-style analysis of scalar implicatures suggests that
a generative theory of Q-implicatures is possible. However,
such a theory will not work for other Q-implicatures:
E.g.: Bonnie stole the apples or the pears.
; ¬BelS (Bonnie stole the apples)
; ¬BelS (Bonnie stole the pears)
Finally:
What does this have to do with conversational implicature?

Quantity implicatures
 Disjunction

Quantity implicatures
Disjunction

Disjunctions belong to the stock-in-trade examples of

sentences that have (scalar) Q-implicatures associated with
them.
But things aren’t as straightforward as is often thought.
Working hypothesis: the lexical meaning of “or” is
inclusive: “ϕ or ψ” means that at least one of ϕ and ψ is
true.

Quantity implicatures
Two types of inference associated with “or”

(1) Wilma is dating Albinoni or Boccherini. (“A or B”)

1 Exclusivity inferences:
; ¬BelS (A and B) (weak exclusivity)
; BelS ¬(A and B) (strong exclusivity)
2 Ignorance inferences:
; ¬BelS (A) and ¬BelS (¬A)
; ¬BelS (B) and ¬BelS (¬B)

Quantity implicatures
Two problems with exclusive “or”

1 It is not clear that strong exclusivity is derivable as a

Q-implicature.
2 The scalar analysis of exclusive “or” doesn’t scale up to
sentences with more than two disjuncts.

Quantity implicatures
A problem with strong exclusivity

This is how it should go:

(1) Wilma is dating Albinoni or Boccherini.

1 Rather than saying (1), S could have said:

(1*) Wilma is dating Albinoni and Boccherini.
Why didn’t S do so?
2 The most likely explanation is that S doesn’t believe
that (1*) is true: ¬BelS (1*).
3 Bivalence: BelS (1*) ∨ BelS ¬(1*).
4 Ergo: BelS ¬(1*).

But: How plausible is the Bivalence Assumption in this case?

Quantity implicatures
A problem with n-ary disjunctions (n ≥ 1)

(2) Wilma is dating Albinoni, Boccherini, or Corelli.

The exclusive reading of (2) should be:

“Only one of A, B, and C is true.”
It’s not clear how this can be derived as a Q-implicature.

Quantity implicatures
A problem with n-ary disjunctions (n ≥ 1)

Idea #1:
Generalise “or” to an n-ary connective:
or(φ1 , . . . , φn ) means that at least one of φ1 , . . . , φn is true.
Ditto for “and”:
and(φ1 , . . . , φn ) means that all of φ1 , . . . , φn are true.
But this yields inferences that are too weak:
“A, B, or C” implicates merely that BelS (¬and(A, B, C)),
i.e., S believes that not all of A, B, and C are true.

Quantity implicatures
A problem with n-ary disjunctions (n ≥ 1)

Idea #2:
Stick to the binary analysis of the connectives:
The underlying form of “A, B, or C”
is either “A or (B or C)” or “(A or B) or C”.
Hence, the stronger alternatives to “A, B, or C” are
A and (B and C) (A and B) and C
either: A and (B or C) or: (A and B) or C
A or (B and C) (A or B) and C
Note: This requires multiple substitutions.

Quantity implicatures
A problem with n-ary disjunctions (n ≥ 1)

A and (B and C)
Take the first set of alternatives: A and (B or C)
A or (B and C)

A B C ruled out by:

1 1 1 ¬(A and (B and C))
1 1 0 ¬(A and (B or C))
1 0 1 ¬(A and (B or C))
+1 0 0 ¬(A or (B and C))
0 1 1 ¬(A or (B and C))
+0 1 0 —
+0 0 1 —
0 0 0 “A, B, or C”

Quantity implicatures
A problem with n-ary disjunctions (n ≥ 1)

Conclusion:
The generative method doesn’t seem to work: in the
general case, exclusivity inferences don’t seem to be scalar
implicatures.
How bad is this?

Quantity implicatures
Ignorance inferences

(3) Wilma is dating Albinoni or Boccherini. (“A or B”)

Ignorance inferences:
; ¬BelS (A) and ¬BelS (¬A)
; ¬BelS (B) and ¬BelS (¬B)

These are not derivable as scalar implicatures, because the

relevant alternatives are shorter than (3):
Wilma is dating Albinoni.
Wilma is dating Boccherini.

Quantity implicatures
A pseudo-dichotomy

The generative view on (scalar) implicatures has led to the

widespread opinion that exclusivity inferences are very
different from ignorance inferences.
Apart from the fact that it is doubtful that exclusivity
inferences can be analysed as scalar implicatures, this
dichotomy is not very appealing.

Quantity implicatures
A deflationist view on scalar implicatures

Lengthwise, there are three ways a sentence can relate to

any of its alternatives: it can be (i) longer, (ii) shorter, or
(iii) the same length.
In the case of (iii), we can speak of “scalar implicatures”.
That’s all.

The key concept in a theory of Q-implicatures is

that of alternatives, not scales.

Quantity implicatures
Where are we now?

1 Quantity implicatures
Grice
Quantity implicatures
Scalar implicatures
Disjunction
2 Nonce inferences or defaults?
3 “Embedded” implicatures
4 Free choice permission
5 Non-pragmatic alternatives

Quantity implicatures