Journal of Logic, Language and Information 13: 491–519, 2004.

C 2004 Kluwer Academic Publishers. Printed in the Netherlands.
491

Exhaustive Interpretation of Complex Sentences

ROBERT VAN ROOIJ and KATRIN SCHULZ

Institute for Logic, Language and Computation, Universiteit van Amsterdam, Nieuwe Doelenstraat
15, 1012 CP Amsterdam, The Netherlands
E-mail: r.a.m.vanrooij, [email protected]

(Received 7 June 2004)

Abstract. In terms of Groenendijk and Stokhof’s (1984) formalization of exhaustive interpretation,

many conversational implicatures can be accounted for. In this paper we justify and generalize this
approach. Our justification proceeds by relating their account via Halpern and Moses’ (1984) non-
monotonic theory of ‘only knowing’ to the Gricean maxims of Quality and the first sub-maxim of
Quantity. The approach of Groenendijk and Stokhof (1984) is generalized such that it can also account
for implicatures that are triggered in subclauses not entailed by the whole complex sentence.

Key words: Circumscription, conversational implicatures, exhaustive interpretation, non-monotonic

reasoning, pragmatics

1. Introduction
One of the most influential pragmatic theories of this century is the theory of con-
versational implicatures proposed by Grice (1967). It has not only been applied to
various semantical problems, but also received considerable attention in philosophy
and the social sciences. The main purpose of this theory was to defend a simple,
truth-conditional approach to semantics, particularly to the meaning of sentential
operators and quantificational phrases. Traditionally, the semantic meaning of nat-
ural language expressions like ‘and’, ‘or’, ‘every’, ‘some’, ‘believe’, and ‘possibly’
has been analyzed in terms of their intuitive analogs in classical logic:‘∧’, ‘∨’, ‘∀’,
‘∃’, ‘
’, and ‘’, respectively. However, in many contexts these expressions receive
interpretations that are different from what is predicted by this approach to their
semantics. It turned out to be extremely difficult to come up with an alternative
semantic theory that can account for the observed interpretations. This led some
ordinary language philosophers such as Ryle and Strawson even to question the
logical approach to natural language semantics in general.
According to Grice (1967), the mistake in this line of reasoning is the assumption
that the problematic interpretations have to be explained by semantics only. He
proposes to single out within the ‘total significance’ of a linguistic utterance the class
of conversational implicatures. Grice takes conversational implicatures (from now
on: implicatures) to be not part of the semantic meaning of an utterance, but to be due
492 R. VAN ROOIJ AND K. SCHULZ

to principles of pragmatics. More particularly, they are inferences an interpreter can

draw from taking the speaker to behave rationally in a cooperative conversational
situation. According to Grice, this means that the speaker is assumed to obey
certain rules that govern such behavior: the maxims of conversation. The idea,
then, is to account for the interpretation of sentential operators and quantificational
expressions in terms of both their semantic meaning as described in classical logic
and a set of conversational implicatures.
While Grice’s notion of conversational implicature is generally accepted, his
proposal concerning the way these inferences are determined is still under debate.
One central issue is the question whether (i) the conversational implicatures of
an utterance are generated globally, after the grammar assigned a meaning to it,
or, whether (ii) the generation refers to intermediate states of the grammar-driven
semantic derivation. Following Grice’s theory one should adopt the first position.
However, it has been argued that a global derivation is not able to account for
the implicatures actually observed (Landman, 2000; Chierchia, ms.). The central
argument brought forward by defenders of a local derivation is the behavior of
implicatures in complex sentences, where the expression whose interpretation is
to be explained is embedded under other sentential operators or quantificational
expressions. For instance, the semantic meaning of numerals such as ‘100’ is often
analyzed as ‘at least 100’ and then conversational implicatures are taken to be
responsible for the ‘exactly’-reading these expressions often receive. Chierchia
now claims that globalists cannot explain why sentence (1) is normally interpreted
as implying that John believes that his colleague makes not more than $100 an
hour, hence, why the numeral in scope of the belief-operator receives an ‘exactly’-
interpretation.

(1) John believes that his colleague makes $100 an hour.

A closer investigation of the argumentation of the localists Landman (2000)

and Chierchia (ms.) reveals that they discuss only one particular approach to a
global description of certain conversational implicatures: the simple scalar ap-
proach. Theories that fit into this scheme assume that sentences can be associated
with expression scales (ordered sets of expressions). They derive the conversational
implicatures of an utterance of sentence s as follows. If s contains an item i from
an expression scale that s can be associated with, let s  be a sentence one obtains by
replacing i in s by another element of this scale that is ranked higher than i. Then s
conversationally implies not s . Such a kind of derivation is, for instance, proposed
in Horn (1972). The conversational implicatures these theories aim to describe are
now generally called – after this approach – scalar implicatures. To give a concrete
example of a derivation, the simple scalar approach can, for instance, account for
the exactly-readings of numerals occuring in simple sentences such as ‘John’s col-
league makes $100 an hour’. Assume that the sentence is associated with the scale
containing the numerals and ordered by increasing height. Then its utterance is
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 493
predicted to conversationally imply that John’s colleague does not earn more than
$100 an hour. Together with the asserted meaning that John’s colleague earns at
least $100 an hour we derive the exactly interpretation. The simple scalar approach
easily gets into trouble with examples such as (1). The only implicature derivable
this way is that John did not believe that his colleague makes more than $100 an
hour. This does not give us the exactly-reading of the embedded numeral that we
intuitively perceive.
However, the argumentation of localists such as Landman and Chierchia would
only be conclusive if they could show that all global accounts get into this kind of
trouble. But the simple scalar approach that they criticize is not the only possible
theory of this kind. A quite different global account has been introduced by Groe-
nendijk and Stokhof (1984). Even though they address the exhaustive interpretation
of answers, and not directly conversational implicatures, their description of ex-
haustivity is able to account for many phenomena analyzed under the latter heading,
in particular for scalar implicatures. Except for its appealing predictions, this pro-
posal also overcomes other shortcomings of previous approaches to conversational
implicatures, such as the neglect of contextual interactions and dependence on the
conceptually difficult notion of expression scales/alternatives. Recently, it has been
shown (van Rooij and Schulz, submitted) how some well-known problems faced
by Groenenendijk and Stokhof’s (1984) account can be overcome by using results
from decision theory and dynamic semantics.
In this article we will study whether this approach can deal with conversational
implicatures of complex sentences. We will see that while it easily accounts for some
of the counterexamples to the simple scalar approach brought forward by localists,
other predictions it makes are not satisfying. We will then develop a generalization
of the approach that can deal with the problematic cases.
At the same time, the generalization will address another open question. While
the work of Groenendijk and Stokhof (1984) and van Rooij and Schulz (submit-
ted) provides us with a powerful formal description of exhaustive interpretation
and many conversational implicatures, neither of these works gives us a satis-
fying theory of the conceptual status of the inferences that they describe. Are
they part of the semantic meaning? Are they products of pragmatic rules? Can
they be explained by Grice’s theory, hence, as due to taking the speaker to obey
the maxims of conversation? As we will see, the generalization of Groenendijk
and Stokhof’s (1984) approach we are going to develop can be interpreted as
formalizing some of the maxims of conversation.1 Thereby it links Groenendijk
and Stokhof’s (1984) exhaustivity operator to Grice’s theory of conversational
implicatures.
The rest of the paper is organized as follows. The next section will be ded-
icated to a discussion of the subtle data of implicatures in complex sentences.
We will then introduce Groenendijk and Stokhof’s (1984) approach to exhaustive
interpretation in Section 3 and discuss the predictions it makes concerning implica-
tures of complex sentences. Afterwards, a new pragmatic interpretation function is
494 R. VAN ROOIJ AND K. SCHULZ

defined that tries to capture parts of Grice’s theory of conversational implicatures.

We will show that it comprises Groenendijk and Stokhof’s account as a special
case. Section 5 is devoted to the application of the introduced framework to various
problems involving conversational implicatures of complex sentences. We conclude
with a discussion of the results.

2. Conversational Implicatures of Complex Sentences: The Data

A general problem one always has to face when discussing conversational implica-
tures is that the observations on which the whole subject is based are rather subtle
and controversial. As the reader will agree with us very soon, this gets even worse
if it comes to implicatures of complex sentences.2 It is widely accepted that con-
textual features – in particular in what kind of exchange we are involved and what
is relevant at the present state of conversation – have a great impact on the issue
of which implicatures are generated. For instance, Hirschberg (1985) argues con-
vincingly that question–answer sequences are important for the analysis of scalar
implicatures, and, just like Groenendijk and Stokhof, gives some examples where
an implicature does not arise when the scalar term used is part of the answer’s
background (see example (11) in the sequel). We will take this observation seri-
ously by restricting our discussion to implicatures that arise in a particular type of
conversation: cooperative exchange of information. Furthermore, we will make the
information structure of the context explicit by taking all examples to be answers
to overt questions. We will choose the questions such that the expressions whose
interpretation is to be explained by implicatures will always occur in that part of
the sentence that could have been used as term-answer. In this way we make sure
that it contributes to the new, relevant information of the sentence. For instance, we
are only interested in the implicatures induced by (1) when uttered in the context
of a question like ‘How much does John believe his colleague makes?’.
As already mentioned in the introduction, the specific implicatures that have
been used by localists to support their point belong to a particular rather well-
studied group of conversational implicatures: the scalar ones. These inferences
are traditionally associated with Grice’s first sub-maxim of Quantity (we will call
implicatures due to this maxim Quantity1-implicatures). Example (1) falls in this
group. Quantity1-implicatures of a sentence s are, roughly speaking, sentences of
the form ¬s  where s  is an alternative to s that is in some sense stronger than s itself.
Controversial in the literature is the issue of what epistemic force these sentences
¬s  should actually be generated with. Roughly speaking, the issue is whether
Quantity1-implicatures should receive a strong or a weak reading. Proponents of
the existence of a strong reading argue either with Horn (1972) that it is indeed ¬s 
that is conversationally implied (we will call this the factive strong reading), or with
Gazdar (1979) (for scalar implicatures) that it is implicated that the speaker knows
or believes ¬s  (what we will call the epistemic strong reading). In the latter case, the
derivation of ¬s  is taken to be due to other rules such as veridicality of knowledge.
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 495

Figure 1.

Proponents of the existence of a weak reading have either argued that sometimes
no Quantity1-implicature is generated at all (the factive weak reading, see Gazdar
(1979) for scalar items under negation) or that one only infers that the knowledge
of the speaker is limited with respect to ¬s  . Here a distinction should be made
between the inference that the speaker thinks it is possible that ¬s  , and, hence,
does not know/believe that s  (the epistemic weak reading, see Soames (1982) for
scalar implicatures) and the inference that the speaker takes both ¬s  and s  to be
possible, and, hence, does not know or does not believe whether ¬s  or s  (what we
will call the ignorance reading, see Gazdar (1979) on clausal implicatures3 ). The
different readings of the implicatures ¬s  are summarized in Figure 1 with some
associated names.
Let us now discuss some reported observations concerning Quantity1-
implicatures of complex sentences using this vocabulary. The critical data we will
discuss here fall roughly in three groups: (i) the scalar item occurs in the scope
of a negation; (ii) the scalar item occurs in the scope of an existential quantifier;
and (iii) the scalar item occurs in the scope of an all-quantifier (such as a belief
operator). This is not a complete classification of the examples that have been
brought forward against global theories of implicatures. But the classification does
capture a wide range of these examples4 and we cannot discuss all of them in one
paper. The reader is invited to try the account we will propose to the other cases by
herself.
The first context we are going to discuss is one of negation. Look at the following
examples:

(2) (A: What did John eat?)

B: John didn’t eat the apples or the pears.
(3) (A: How many apples did John eat?)
B: John didn’t eat three apples.

In the literature, mainly two readings are reported for such examples:5 (a) a
factive weak reading, according to which no Quantity1-implicature is present if
the scalar item occurs under negation (see, e.g., Gazdar, 1979; Hirschberg, 1985;
Landman, 2000); and (b) a reading where the sentence raises factive strong implica-
tures, for (3), for instance, that John did not eat less than two apples (e.g., Atlas and
Levinson, 1981; Levinson, 2000; Chierchia, ms.). According to our informants the
496 R. VAN ROOIJ AND K. SCHULZ

answers in (2) and (3) normally imply that the speaker cannot provide a complete
answer and the given response is the best she can do. Hence, they report epistemic
weak or ignorance implicatures. Some informants can also get the strong factive
inferences but others rigorously exclude them.
A second group of examples brought forward by localists can be character-
ized as existence-quantifying contexts.6 We start with the simple case of multiple
disjunction.

(4) (A: Who knows the answer?)

B: Peter, Mary, or Sue.

Intuitively, the answer given in (4) has an interpretation according to which only
one of the three persons knows the answer. Notice that the sentence B uses in her
response only counts as being complex if one assumes that multiple disjunction
constructions are based on the iterative application of a binary disjunction operator.
But given that the opinions on the question how to analyze such constructions still
diverge and that many global theories do have a problem with this example no
matter whether they assume an analysis with one n-ary or two occurrences of a
binary ‘or’, we thought it being a good idea to discuss this example here.7 The next
example is discussed in Landman (2000).

(5) (A: Who invited whom?)

B: Three boys invited four girls.

Landman analyzes the semantic meaning of the cumulative reading of (5) as

follows: ∃e ∈ ∗ INVITE : ∃x ∈ ∗ BOY : card(x) = 3 ∧ ∗ Agent (e) = x ∧ ∃y ∈ ∗ GIRL
: card(y) = 4 ∧ ∗ Theme(e) = y. Hence, the groups of boys and girls are introduced
in the scope of an existential quantifier over events. According to Landman the
factive strong implicature that should be described is that no more than three boys
invited a girl and not more than four girls were invited by a boy. He takes this to
be a problem for global accounts given that the noun phrases are interpreted under
the scope of an existential quantifier.
Another example, adapted from Chierchia (ms), is the following:

(6) (A: What did John eat?)

B: John ate the apples or some of the pears.

Here, the scalar item ‘some’ occurs under ‘or’. According to Chierchia the
answer should get a reading according to which John either ate the apples, or some,
but not all of the pears. This is again a factive strong inference. He claims that a
global account cannot make this prediction.
Finally, we discuss some examples where classical scalar items occur in the
scope of an all-quantification.
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 497
(7) (A: Who kissed whom?)
B: Every boy kissed three girls.

According to Landman the answer of B in (7) has the factive strong implicature
that every boy kissed no more than three girls. A similar intuition he reports for (8).

(8) (A: Who does Bill believe were at the party?)

B: Bill believes that there were four boys at the party.

Again, we should obtain the implicature that Bill believes that there were no
more than four boys at the party. Chierchia makes similar observations. How-
ever, opinions diverge whether these implicatures are indeed generally observed in
all-quantifying contexts. An anonymous referee questioned whether a sentence like
‘Every admirer of Dickens read Bleak House or Great Expectations.’ does come
with the implicature that no admirer read both of the books. While we admit that
these implicatures do not have to occur, we think nevertheless that they represent
reasonable readings – particularly in the kinds of context we use. For instance, (9)
seems to us to come with a reading implying that every student took not all three
courses, Semantics 1 and Phonology 1 and 2.

(9) A: Which courses did your students take?

B: Every student took Semantics 1 or Phonology 1 and 2.

3. Implicatures in Non-Monotonic Logic

The examples we have discussed in the last section have been used by Landman
(2000) and Chierchia (ms.) to argue against a global approach to conversational im-
plicatures. However, as we have pointed out in the introduction, their argumentation
is not conclusive because they only showed for one particular global theory that it
fails to make the correct predictions. In this section we will introduce a promising
alternative global approach, the description Groenendijk and Stokhof (1984) pro-
pose for exhaustive interpretation, and discuss the predictions this account makes
for the implicatures of complex sentences.

3.1. CIRCUMSCRIPTION
According to Grice (1989), one of the defining features of conversational im-
plicatures is that they may be cancelled. This is still by far the most com-
monly used property to identify these inferences. But calling implicatures can-
celable is nothing but calling them non-monotonic inferences. This suggests that
techniques and results from non-monotonic logic are useful for the analysis of
implicatures.
498 R. VAN ROOIJ AND K. SCHULZ

There is one approach that successfully uses such techniques to account for a
particular class of conversational implicatures – though without noticing the con-
nection to non-monotonic logic: the proposal of Groenendijk and Stokhof (1984)
(from now on abbreviated as G&S).8 Actually, they did not intend to describe
implicatures, their aim was to account for the particular way we often interpret
answers: we take an answer such as ‘Peter’ to question ‘Who called yesterday?’
not only as conveying that Peter was among the callers, but additionally that he
is the only person who called yesterday. This reading is known as the exhaustive
interpretation of the answer. It is a well-know fact – also illustrated by the para-
phrase of the exhaustive interpretation just given – that this mode of interpretation
is closely connected with the way we understand sentences containing ‘only’. How-
ever, ‘only’-paraphrases are also often given to reinforce implicatures. This holds,
in particular, for inferences that are analyzed as scalar implicatures. To give an
example, in a context where it is relevant how many cookies Paula ate, (10a) is
quite generally reported to come with the cancelable inference that Paula did not
eat all of the cookies. This meaning can also be expressed by (10b) – but now it is
no longer cancelable.

(10) (a) Paula ate some of the cookies.

(b) Paula ate only [some] F of the cookies.9

Given this connection between exhaustive interpretation, the meaning of ‘only’,

and scalar implicatures, it should not come as a surprise that as far as G&S are
successful in accounting for the exhaustive interpretation of answers, they can also
describe many classical scalar implicatures (and the meaning of ‘only’) – but of
course, now dependent on the particular question the sentence is meant to answer.10
Before we illustrate the descriptive power of the approach with some examples, let
us first quickly review their proposal.
G&S describe the exhaustive interpretation as the following interpretation func-
tion, taking as arguments (i) the predicate B of the question, and (ii) the meaning
of the term-answer, or focus, F to the question.11

exh(F, B) =de f F(B) ∧ ¬∃B  ⊆ D : F(B  ) ∧ B  ⊂ B

Van Benthem (1989) first observed that this function can be seen as instantiating
one of the first and best-known mechanisms to describe non-monotonic inferences:
predicate circumscription, introduced by McCarthy (1980). Predicate circumscrip-
tion is an operation that maps theories A and predicates P to the following theory
called the circumscription of P with respect to A.12

cir c(A, P) ≡de f A ∧ ¬∃P  ⊆ D : A[P  /P] ∧ P  ⊂ P

EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 499

It is obvious that exh(F, B) can be obtained from circ(A, P) by taking B for P

and A to be F(B), hence instead of the term-answer, the sentential answer.13
For our purposes it is important to notice the following model-theoretic analog of
circumscription: interpretation in minimal models. In order to make the connection,
we have to enrich the model theory for classical predicate logic by defining an order
on the class W of possible models of our predicate logical language in the following
way: a model v is said to be more minimal than model w with respect to some
predicate P, v < P w, in case they agree on everything except the interpretation
they assign to P and here it holds that P(v) ⊂ P(w). In this setting fact 1 is a
well-known result.

FACT 1.

∀w ∈ W : w |= cir c(A, P) ⇔ w |= A ∧ ¬∃v ∈ W : v |= A ∧ v < P w.

This fact shows that we can equivalently describe predicate circumscription by the
interpretation function circW (A, P) defined as follows:

cir c W (A, P) =de f {w ∈ W |w |= A ∧ ¬∃v ∈ W : v |= A ∧ v < P w}

3.2. PROSPECTS AND PROBLEMS OF THE CIRCUMSCRIPTION ACCOUNT

In terms of G&S’s exhaustivity operator, or of circumscription, quite a number of
conversational implicatures (including scalar ones) can be accounted for straight-
forwardly. Besides the obvious result that from the answer ‘P(a)’ to a question with
question-predicate P we derive that a is the only object that has property P, we also
derive (i) for ‘John ate three apples’ that John ate exactly three apples;14 for P as
question-predicate (ii) the exclusive reading of a disjunctive sentence like ‘P(a) ∨
P(b)’; (iii) the implicature that not everybody has property P from the assertion that
most have; (iv) the so-called conversion-inference that every P-thing is a Q-thing,
if the answer is ‘Every Q is a P’; and (v) the biconditional reading of ‘John will
come if Mary will go’, if this sentence is given as answer to the polar question ‘Will
John come?’. Another pleasing property of an exhaustivity analysis of implicatures
is that it predicts that it depends on the context, or question-predicate, whether we
observe these inferences. If, for instance, the scalar item occurs in the question-
predicate P instead of in the focus F of the answer, as for instance in example (11),
no implicatures are predicted.15

(11) A: Do you have some apples?

B: Yes, I have some apples.
500 R. VAN ROOIJ AND K. SCHULZ

This may account (at least partially) for the often cited context- and relevance-
dependence of implicatures and the observed factive weak readings of sentences
containing scalar items. All these predictions are appealing and they show that
this approach outperforms many other accounts of Quantity1-implicatures. Note,
furthermore, that G&S’s description of exhaustive interpretation (and this is even
more true for circ(A,P)) is a global account of implicatures, because it can be
assumed to work on the output of the grammar, or on some kind of discourse
representation.16
Still, there are some serious limitations of an analysis of implicatures in terms
of circumscription or exhaustive interpretation.17 First, it is quite obvious that such
an analysis cannot account for the context-dependency of exhaustive interpretation
on other factors than the predicate of the question, which plays, for instance, a
role for phenomena such as domain restriction, or answers that receive a mention-
some, instead of a mention-all reading. This is inevitable given the functionality
of exh as defined by G&S. The restricted functionality of the operation also causes
other problems: because circumscription (or exhaustification) works immediately
on the semantic meaning of an expression, it is predicted that if two sentences
have the same semantic meaning, they will give rise to the same implicatures as
well. This, however, does not seem to be the case. It is, for instance, standardly
assumed in generalized quantifier theory (adopted by G&S) that ‘three men’ has
the same semantic meaning as ‘at least three men’. But sentences in which the
former occurs seem to give rise to an ‘at most’ implicature, while the latter do
not.
In van Rooij and Schulz (submitted) solutions to these and some other problems
are proposed by bringing the approach of G&S (particularly in the form of circ(·))
together with some independent developments in natural language semantics and
pragmatics. For instance, by adopting dynamic semantics (e.g., Kamp, 1981; Heim,
1982), which allows more fine-grained distinctions than static semantics does, one
can account for at least part of the functionality problem. As for the context depen-
dence of exhaustive answers, it is proposed in van Rooij and Schulz (submitted)
that the ordering between worlds should not be defined in terms of the extensions
the question-predicate has in different worlds, but rather in terms of the utility
or relevance of the propositions that express what those extensions are in those
worlds. In this way we make the exhaustification operator more sensible to the
beliefs and preferences of the agents involved, and can account for, among others,
both mention-all and mention-some readings of answers. Furthermore, this may
also help us to get an even better grasp of the context (and relevance) dependence
of implicatures.
In this paper we are interested in the issue to what extent this global account
of implicatures can deal with implicatures of complex sentences. So, let us start to
check the examples discussed in Section 2. It proves to be the case that some of the
observations that are claimed by Landman (2000) and Chierchia (ms.) to be out of
reach of global accounts are correctly predicted immediately.
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 501
First, even though this is not an example from localists, notice that by exhaustive
interpretation or circumscription we straightforwardly get the correct prediction for
(4) that only one of the three disjuncts is true. This is even independent of the issue
which position is taken with respect to the functionality of ‘or’. In the same way,
circ predicts for (5) and (6) the reported factive strong implicatures. The approach
can also deal with the all-quantification examples (7) and (9). We discuss shortly
the second one. Consider the sentence ∀x(P(x, a) ∨ P(x, b)) in the context of
a question ?x, y[P(x, y)] (or ?y[∀x P(x, y)]).18 Applying circ will minimize the
number of tuples x, y for which P(x, y) holds. In particular, it will minimize for
every x the number of y such that P(x, y). For example, (9), this number is smallest
if every student either took Semantics 1 or Phonology 1 and 2, but not both. Hence,
it is implied that no student took all three courses.
However, G&S’s approach is not able to deal with the other examples we have
discussed in Section 2, and also the improvements on this account proposed in
van Rooij and Schulz (submitted) are of no help here. These were the cases where
the relevant expression occurs, for instance, under negation or a verb of belief.
G&S were already well-aware of the problem concerning negation: by taking the
exhaustive interpretation of ¬P(a) with respect to predicate P, we end up with the
wrong result that in the actual world P has an empty extension. Some improvement
can be made by proposing that if a negation occurs in the answer then it is not
P that is circumscribed, but the complement P̄ (as proposed by von Stechow and
Zimmermann, 1984). In this way we can account for the readings of (2) and (3)
with factive strong implicatures. However, as we have already discussed above,
while some speakers of English can get these inferences, many others claim that
they do not. They understand an answer like ‘Not Peter’ to the question ‘Who
called yesterday?’ as stating that Peter did not call yesterday and that as far as
the speaker knows other individuals might have called, and, hence, only get a
reading with epistemic weak implicatures. Furthermore, sentences concerning the
belief-state of the speaker or of other agents such as (1) are problematic for the
reason that an account of exhaustive interpretation in terms of circumscription is
purely extensional. circ totally ignores information about P in other (epistemic)
possibilities than the actual world. To give another example where this causes a
problem, take (12).

(12) A: Who knows the answer?

B: Peter and possibly Mary.

Following our informants, B’s response can have two different readings. Ac-
cording to the first, it gives rise to the factive strong inference that except for Peter
and perhaps Mary, nobody knows the answer. According to the second, the an-
swer exhausts the knowledge of the speaker in the sense that she knows that Peter
knows the answer, she has some evidence that also Mary knows the answer, but
for all other people they may as well not know the answer. Hence, in this case the
502 R. VAN ROOIJ AND K. SCHULZ

interpreter only derives epistemic weak implicatures. For the reason given above,
circ can account for neither of these readings.

4. All that the Speaker Knows: Setting the Stage

In this section, a generalization of the circumscription account to exhaustive inter-
pretation is introduced that will help us to account for the problematic implicatures
of complex sentences. But before we start we should decide first which of the re-
ported readings we actually want to describe. At the beginning of Section 2 we
distinguished five readings of Quantity1-implicatures that can be found in the liter-
ature: the factive strong, the epistemic strong, the factive weak, the epistemic weak,
and the ignorance reading. However, they may not all exist independently of each
other. For instance, some of the readings entail others. The latter may, therefore,
be distinguished as an independent reading only because some additional infer-
ences were ignored. There is, for instance, some evidence that at least in some
cases when a factive weak reading is diagnosed what was actually observed was an
epistemic weak interpretation.19 Furthermore, we have seen in Section 3 that circ
can account for weak factive readings if the ‘trigger’ of the implicature occurs in
the question-predicate or background of the answer. We will propose that in those
cases discussed here, where the expression that triggers an implicature appears in
the focus-or term-answer-part of an answer to an explicitly asked question we do
not have to distinguish a weak factive reading. Likewise, there is some evidence that
also the ignorance reading is due to a misinterpretation of the data. For instance,
according to Gazdar’s (1979) formal account of clausal implicatures at page 59, the
sentence ‘My sister is either in the bathroom or in the kitchen’ has the implicatures
the speaker does not know that her sister is in the bathroom, the speaker does not
know that her sister is not in the bathroom, the speaker does not know that her sister
is in the kitchen, the speaker does not know that her sister is not in the kitchen.
Earlier on at page 50, however, he describes the implicatures of the sentence as
being I don’t know that my sister is in the bathroom and I don’t know that my sister
is in the kitchen and thus only reports epistemic weak implicatures. Based on such
observations we decided to assume that for the cases discussed here, we can also
dial out the ignorance reading. Finally, we would like to claim that there is no factive
strong implicature without the inference that the speaker knows the implicature to
be true. Then, under the additional assumption that the factive strong implicatures
come about via the verdicality of knowledge, we can restrict our considerations to
one strong reading: the epistemic strong reading of implicatures.
Given this analysis, what we have to model is a function that makes the interpre-
tation of a sentence dependent on the question-predicate and allows both epistemic
strong and weak implicatures. Furthermore, we have seen that we already have with
circ a promising description of the factive consequences of the epistemic strong
inferences. Hence, any approach to the epistemic strong reading should turn out to
predict factive inferences that are strongly related to those of circ.
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 503
The proposed reduction to two context-dependent readings of the implicatures
of an answer nicely reflects a central tenor in the literature on implicatures. Many
authors acknowledge the existence of both a weak and a strong reading of Quantity1-
implicatures – even though they may disagree on the exact form of these readings.
There are also some divergences with respect to the question what determines
which reading should be predicted in a certain context. Recall from Section 2 that
the Quantity1-implicatures of a sentence s are based on sentences s  that are, in
some sense, stronger alternatives to s. According to Gazdar (1979), it is the form of
s  that decides the epistemic force with which a Quantity1-implicature is generated.
He distinguishes between two classes: scalar and clausal Quantity1-implicatures.
Scalar implicatures are based on sentences s  that imply s and that are obtained
from s by replacing an item in s by an alternative from a certain expression scale.
The actual implicature is that the speaker knows ¬s  (epistemic strong reading). A
clausal implicature is based on a sub-sentence s  of s that is not decided by s. Here,
the actual implicature is that the speaker does not know ¬s  and does not know s 
(ignorance reading). According to such an approach, an utterance may raise at the
same time both weak and strong Quantity1-implicatures.
Gazdar has often been criticized for the prediction that scalar implicatures al-
ways have to have strong epistemic force. Instead, it has been argued by many
students of conversational implicatures that scalar implicatures should be gener-
ated primarily with weak epistemic force. Only in contexts where the speaker is
assumed/ believed to know that ¬s  if ¬s  is true, or, in other words, if the speaker
is taken to be competent on s  , the epistemic strong reading is derived (see, among
others, Soames, 1982; Leech, 1983; Horn, 1989; Matsumoto, 1995; Green, 1995).
Apart from intuitions, one other argument given for this analysis is that in the
Gricean derivation often used to explain scalar implicatures one explicitly has to
make this additional assumption on the competence of the speaker; the maxims
alone are not strong enough to derive the strong reading.20,21 We agree with these
authors that, indeed, scalar implicatures can, depending on the context, be gener-
ated with either strong or with weak epistemic force and that Gazdar’s predictions
are not adequate here. An additional pleasing property of such an analysis is that
Gazdar’s unmotivated difference between the epistemic force of scalar and clausal
implicatures is weakened. Now both, scalar and clausal implicatures are primarily
generated with a weak epistemic reading. However, the distinction between the two
kinds of implicatures thereby does not disappear. For one thing, clausal implicatures
are still claimed to have the stronger ignorance reading. For another, both classes
of implicatures are described by two different generation processes for s  . This is
not very convincing given that Gazdar ascribes both types of inferences to the same
maxim: the first sub-maxim of Quantity.
Schulz (2003) shows how some developments in non-monotonic logic, namely
the work of Halpern and Moses (1984) on the concept of ‘only knowing’, recently
generalized by van der Hoek et al. (1999, 2000), can be used to improve on Gazdar’s
account of clausal implicatures. One of the advantages of the approach is that it is
504 R. VAN ROOIJ AND K. SCHULZ

much closer to Grice’s formulation of the first sub-maxim of Quantity than Gazdar’s
account. This raises the following question: given that we have an approach that
nicely describes clausal implicatures, can we extend this approach so that it can
also deal with scalar implicatures and the different epistemic readings observed?
This is the topic of the present section.
We will start by applying this new approach to clausal implicatures to the situa-
tion at hand. In order to do so, we have to introduce some technical machinery. The
relevant aspects of natural language interpretation will be modeled using a formal
language L of modal predicate logic that is generated from predicate and function
symbols of various types, variables, the logical connectors ¬, ∧, and ∀ (we will
use φ ∨ ψ and ∃x.φ to abbreviate ¬(¬φ ∧ ¬ψ) and ¬∀x.¬φ , respectively), and
for the moment we are satisfied with having one modal operator
in our language
that refers to the knowledge-state of the speaker.
φ should be read as the speaker
knows that φ. φ ≡de f ¬
¬φ expresses in turn that the speaker takes it as possible
that φ. L is the set of all sentences (hence, formulas containing no free variables)
that can be constructed from these primitives in the standard way. We will occasion-
ally refer to the basic language L0 ⊆ L, which is the set of sentences containing
no modal operators.
L-formulas are interpreted with respect to states s and assignments g. A state
is a tuple M, w consisting of a model M = (W, R, D, V ) (where W is a set
of points (possible worlds), R a binary relation on W, D a set of individuals, and
V an interpretation function for our non-logical vocabulary) and a point w ∈ W .
R[w] denotes the set {v ∈ W |R(w, v)}. We will use a strong version of the unique
domain assumption: not only do we take the domain D to be the same for all
worlds in a state, but we assume in addition that all states have the same domain.
Truth of a formula φ ∈ L with respect to a state s = M, w and an assignment
g is defined in standard ways. We will give here only the definition of truth for a
modal formula and assume the reader’s familiarity with the standard definitions:
M, w, g |=
ψ iffde f for all worlds v such that v ∈ R[w] it holds that M, v, g |= ψ.
When we talk about the truth of a sentence, the assignment will be dropped. We
will work with a restricted class of states: those states where the relation R is an
equivalence relation. This is a standard way to turn R into a relation that can rep-
resent the knowledge-state of the speaker: she is assumed to be fully introspective,
and her beliefs are taken to be true. Let S be the class of states that fulfill this
restriction.
We will now directly start from Grice’s theory of conversational implicatures
and try to capture the central ideas of his maxims Quantity1 and Quality in terms of
a pragmatic interpretation function for answers: this function will map sentences on
the set of states where the speaker knows what she claims (she believes her utterance
and has evidence for its truth) and provides all the relevant knowledge she has –
she gives the best (i.e., most informative) answer she can, given her knowledge.22
To account for the ‘maximally informative’ part, we enrich the class of states S
for our language with an order that compares how much relevant information the
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 505
speaker has in different states. For the application at hand the order is intended to
compare how much the speaker knows about the extension of predicate P in different
states. Our pragmatic interpretation function selects among those states where the
speaker knows that her utterance is true only the minimal elements with respect to
this order, hence, those states where the speaker knows least about the extension
of P.23

DEFINITION 1. (The Epistemic Weak Reading) Given a sentence A of L and a

predicate P, we define the pragmatic meaning eps1S (A, P) of A with respect to P
and a set of states S as follows:

eps1S (A, P) =de f {s ∈ S|s |=
A ∧ [∀s  ∈ S : s  |=
A → s 
s  ]}

But how to define the order 
? The concept is very simple. When does a
speaker know more about the extension of the question-predicate P? If she takes
less possible extensions of P to be compatible with her beliefs. And because of the
epistemic weak reading only positive information about P counts (hence, knowl-
edge that P(a) but not knowledge that ¬P(a)), it is sufficient to call s1 = M1 , w1 
as least as small as s2 = M2 , w2  if for every epistemic possibility in s2 the speaker
distinguishes an epistemic possibility in s1 where the extension of P is smaller than
or equal to the extension of P in s2 . Hence, we define:

DEFINITION 2. For all

s1 = M1 , w1 , s2 = M2 , w2  ∈ S :
s1 
s2 iffde f ∀v2 ∈ R2 [w2 ]∃v1 ∈ R1 [w1 ]
V1 (P)(v1 ) ⊆ V2 (P)(v2 )
s1 ∼
=
s2 iffde f s1 
s2 and s2 
s1

To illustrate the workings of the order and also of eps1S (·), assume that D has
only two elements, a and b. Then, the extension of P in each point v of every state
s = M, w ∈ S can only have four different values. We use this observation to
define the following function f on R[w]:

f (v) = 0 iffde f V (P)(v) = ∅, f (v) = b iffde f V (P)(v) = {b},

f (v) = a iffde f V (P)(v) = {a}, f (v) = ab iffde f V (P)(v) = {a, b}.

We can then classify states according to which of those four cases the speaker
considers possible. For X ⊆ {0, a, b, ab} define [X] = {M, w|x ∈ X ⇔ ∃v ∈
R[w] : f (v) = x}. For instance, [{a, ab}] stands for those states where the speaker
506 R. VAN ROOIJ AND K. SCHULZ

Figure 2. An arrow from some set of states y to some set of states x represents that for all s1 ∈
x, s2 , ∈ y: s1  P s2 . Not very surprising: for states s1 , s2 ∈ [X ], X ⊆ {0, a, b, ab}: s1 ∼
=
s2 .
This picture shows that the 
-smallest states in S are those where the speaker takes it to be
possible that neither a nor b has property P. The maximum is constituted by those states where
the speaker knows that both a and b have property P.

distinguishes at least one epistemic possibility where a but not b is in the extension
of P and one epistemic possibility where both a and b are in P. With these definitions
at hand we can now represent the structure 
imposes on S – see Figure 2.
We can now use Figure 2 to calculate the pragmatic interpretation of some ex-
amples. Let us for the moment make the simplifying assumption that the speaker
knows which individual bears which name and that there is only one name
for each individual. Hence, we can identify the individuals with their names.
To calculate eps1S (P(b), P), we first select the states where the speaker knows
P(b) : [{b}] ∪ [{ab}] ∪ [{b, ab}]. Then, Figure 2 helps us to select the minima
among these states. We end up with eps1S (P(b), P) = [{b}] ∪ [{b, ab}]. Hence,
according to the interpretation function eps1 , the speaker considers it (at least)
possible that a does not have property P. In general, under the assumption made
about the relation between names and individuals, eps1S (P(b), P) implies that for
all x ∈ D, x = b the speaker takes ¬P(x) to be possible.
Applied to the sentence P(a) ∨ P(b), we obtain as the pragmatic meaning the
states that are of type [{a,b}] or [{a, b, ab}]. Hence, applying the pragmatic in-
terpretation function eps1 allows the interpreter to conclude from an utterance of
the form P(a) ∨ P(b) that the speaker neither knows P(a) nor P(b). This captures
exactly Gazadar’s (1979: 50) reported intuition concerning clausal implicatures of
a disjunction. Futhermore, we obtain the inference ¬
(P(a) ∧ P(b)) which accord-
ing to Soames (1982), among others, is the (epistemic weak) scalar implicature of
such a sentence. In sum, at least for these two examples, it looks as if eps1 really
allows us to describe the epistemic weak inferences we wanted to account for. And,
as the second example shows, we can describe with one and the same operation
both scalar and clausal implicatures.
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 507
However, so far we have done nothing to account for the strong reading that
Quantity1-implicatures sometimes receive. To describe this occasional strengthen-
ing of eps1 , we want to use the intuition expressed so often in the literature that the
epistemic strong reading is obtained in case the speaker is taken to be competent, or
be an authority. For our setting this means to take the speaker to know the answer
to the question asked. In Zimmermann (2000), Groenendijk and Stokhof’s (1984)
analysis of ‘knowing whether’ is used to define what it means for a speaker to be
competent with respect to a predicate P.

DEFINITION 3. (Competence) A speaker is competent in state M, w S (where

M = W, R, D, V ) with respect to a predicate P iffde f ∀v ∈ R[w].∀x ∈ D : w ∈
P(x) ⇔ v ∈ P(x).

Given the intuition that competence should play a role in the derivation of the
strong reading of Quantity1-implicatures, a first idea that comes to mind is that
we simply have to apply eps1 to the set COMP of states where the speaker is
competent with respect to P to obtain the epistemic strong reading. Unfortunately,
this will not work. According to this approach the epistemic strong reading can
occur only in situations where the speaker is taken to be competent, i.e., if her
utterance is interpreted with respect to COMP (or a subset of COMP). However,
there are sentences that have epistemic strong implicatures but cannot stem from a
speaker that is (i) competent in the sense just defined, and (ii) obeying the maxims
Quantity1 and Quality as interpreted by eps1 . A good example for such a sentence
is a disjunction like P(a) ∨ P(b). Let us calculate epsCOMP1 (P(a) ∨ P(b), P). A
speaker competent on P knows whether P(a) holds and whether P(b) holds. Hence,
if she believes P(a) ∨ P(b) she can only be in one of the following three types of
states: [{a}], [{b}] or [{ab}]. Now, take a look at Figure 2 again. Applying eps1
means that we have to select the 
-minima among these states. In this case the
speaker can be in different types of minimal knowledge-states: [{a}] and [{b}].
But then the speaker was withholding the relevant information which of P(a) and
P(b) does in fact hold. Hence, it is obvious for the interpreter that the speaker must
be breaking the maxim Quantity1. eps1 reflects this by assigning to P(a) ∨ P(b) on
COMP the empty interpretation: the sentence is predicted to be pragmatically not
well-formed. This certainly does not match our intuitions. Such a sentence can be
read as implying that the speaker knows that not both P(a) and P(b) hold, which
is an epistemic strong implicature (while raising at the same time the epistemic
weak implicature that the speaker does not know which of the disjuncts is true).
Given this result we have to conclude that epsCOMP
1 does not provide an adequate
description of the epistemic strong reading of implicatures.
Where should we locate the mistake? Did we choose the wrong formalization of
the maxims? That does not seem to be the case, because we do get the right results
for the epistemic weak inferences. Therefore, it might be a better idea to keep this
part and rethink the role of competence. Consider our example again. The inference
508 R. VAN ROOIJ AND K. SCHULZ

we were after was
¬(P(a) ∧ P(b)). But to derive this, we do not have to assume
that the speaker is fully competent on the extension of P – the assumption that
caused the counterintuitive interpretation above – it is enough to assume that she
knows whether or not the conjunction P(a) ∧ P(b) is true. This suggests modeling
the strong reading by adding competence only as far as this is consistent with the
assumption that the speaker obeys the maxims Quantity1 and Quality, an idea that
can also be found in Spector (2003).24 Thus, (i) competence should be treated as
something that is maximized, and (ii) competence should be maximized after the
application of eps1 .
To model maximizing competence we will use the same strategy as for mini-
mizing knowledge. We take  to denote the order defined as follows:

DEFINITION 4. For all

s1 = M1 , w1 , s2 = M2 , w2  ∈ S :
s1  s2 iffde f ∀v1 ∈ R1 [w1 ]∃v2 ∈ R2 [w2 ]
V1 (P)(v1 ) ⊆ V2 (P)(v2 )
s1 ∼
= s2 iffde f s1  s2 and s2  s1

We can use the same notation introduced earlier to visualize the structure
that  imposes on S in case there are only two individuals a and b – see
Figure 3. One can see that the lower a state is in the hierarchy imposed by  ,
the more individuals the speaker knows in s that they do not have property P. To
model maximizing competence we will thus select minima with respect to this
order.
We define a new pragmatic interpretation function, eps2S (A, P), that strengthens
eps1S (A, P) by selecting among the states in eps1S (A, P) those that are minimal with
respect to  .

DEFINITION 5. (The Epistemic Strong Reading) Given a sentence A of L and a

predicate P, we define the pragmatic meaning eps2S (A, P) of A with respect to P
and a set of states S as follows:

eps12 (A, P) = de f{s ∈ S|s |=
A ∧ [∀s  ∈ S : s  |=
A

→ [s 
s  ∧ (s ∼=
s  → s  ≺ s)]]}


= s ∈ eps1S (A, P)|∀s  ∈ eps1S (A, P) : s  ≺ s

In this interpretation function the application of the order 
has priority over
 . The latter only comes to work if the former does not see any difference between
two states.25 This captures our earlier conclusion that maximizing competence
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 509

Figure 3.  works quite opposite to 
. The  -biggest states are those where the speaker
takes it to be possible that both a and b have property P. The minimum is constituted by those
states where the speaker knows that both, a and b, do not have property P.

should only be executed as far as it does not conflict with the maxims Quantity1
and Quality as formalized in eps1 .
To illustrate the working of eps2S (·), let us calculate the pragmatic interpreta-
tion assigned by eps2 to the sentence P(b) with respect to P and S. Again, we
assume for simplicity that the speaker knows which individual bears which name
and that there is only one name for each individual. We already know that the

-minimal states where
P(b) holds are eps1S (A, P) = [{b}] ∪ [{b, ab}]. Us-
ing Figure 3 we can now select the  -minimal states among the elements of
eps1S (A, P) and end up with eps1S (P(b), P) = [{b}]. Hence, according to this prag-
matic interpretation function the speaker knows that P(b) and ¬P(a) – the second
part is the epistemic strong implicature we wanted to derive. With the veridicality of
knowledge we obtain eps2S (P(b), P) |= P(b) ∧ ¬P(a) – exactly the predictions of
cir cS (P(b), P)!26
Let us discuss one more example. eps2S (P(a)∨ P(b), P) is the set of  -minima
among eps1S (A, P) = [{a, b}]∪[{a, b, ab}], which is the set [{a, b}]. In these states
the speaker takes both P(a) and P(b) as possible. Hence, eps2S (P(a) ∨ P(b), P)
entails the epistemic weak clausal implicatures P(a) and P(b) (as did eps1 ). We
also derive that the speaker knows that P(a) and P(b) are not both true at the same
time. Thus, together with the veridicality of knowledge this interpretation function
can account for the exclusive interpretation of ‘or’: eps1S (P(a) ∨ P(b), P) |=
¬(P(a) ∧ P(b)), as was also predicted by circS (P(a) ∨ P(b), P).
In the two cases discussed above the pragmatic interpretation function eps2
makes very promising predictions: we obtain the intended epistemic strong readings
of certain implicatures which, together with the veridicality of knowledge, imply
the inferences of circ. Of course, we would rather like to establish the adequacy of
eps2 in some generality. Let us see what we can achieve here. The following fact
is quite easy to prove.
510 R. VAN ROOIJ AND K. SCHULZ

FACT 2. For A ∈ L0 and P a predicate of L, eps2S (A, P) implies circumscription

of the predicate P with respect to A and S:

eps2S (A, P) ⊆ cir cS (A, P).

Proof. Assume that for s = M, w ∈ S, s ∈ eps2S (A, P). By reflexivity of R

it follows that s |= A. Assume additionally that s ∈ / cir cS (A, P). In consequence
∃s = M , w  ∈ S : s |= A and V of s is defined in w  as is V of s in w except
     

that V  (P)(w ) ⊂ V (P)(w).

We choose an s ∗ = M ∗ , w ∈ S where M ∗ is like M except that the valuation
function V ∗ deviates from V of M as follows: if for v ∈ R[w] : V (P)(w) ⊆
V (P)(v), then V ∗ evaluates the non-logical vocabulary in v as does V  of M  in
w  ; in all other points of R[w], V and V ∗ assign the same interpretation to the
non-logical vocabulary.27 We will show that (i) s ∗ |=
A, (ii) s ∗ 
s and (iii)
s ∗ ≺ s. Thus, s ∗ falsifies that s is in eps2S (A, P). This would prove the claim.
Ad (i). Let us introduce the notation N , u ≡0 N  , u   iffde f the valuation
functions of the models N and N  agree on the interpretation of the non-logical
vocabulary in u and u  . For every v ∈ R[w] it is either the case that M ∗ , v ≡0
M, v or (in case V (P)(w) ⊆ V (P)(v))M ∗ , v ≡0 M  , w  . It is easy to see that
if for two states N , u ≡0 N  , u   then they make the same sentences φ ∈ L0
true. Because A ∈ L0 , s |=
A and s  |= A we can conclude s ∗ |=
A. Ad
(ii). We have to show that ∀v ∈ R[w]∃v ∗ ∈ R[w] : V ∗ (P)(v ∗ ) ⊆ V (P)(v).
Take v ∗ = v. By construction either V ∗ (P)(v) = V (P)(v) or (in case V (P)(w) ⊆
V (P)(v)), V ∗ (P) (v) = V  (P)(w ) ⊆ V (P)(v). Ad (iii). To show that s ∗  s and,
hence, ∀v ∗ ∈ R[w]∃v ∈ R[w] : V ∗ (P)(v ∗ ) ⊆ V (P)(v) take again v = v ∗ . Finally,
if s  s ∗ then we would have ∀v ∈ R[w]∃v ∗ ∈ R[w] : V (P)(v) ⊆ V ∗ (P)(v ∗ ).
But take v = w. By construction there will be no v ∗ in R ∗ [w∗ ] where P is as least
as big as in w. Hence, s  s ∗ .

Fact 2 shows that if applied to an answer that contains no modal operators, the
new pragmatic interpretation function eps2 will give us all the inferences we also
obtain by using predicate circumscription – a quite nice result. It does not extend
to arbitrary A ∈ L. Consider, for instance, A ≡ P(a). By circumscription we
obtain: cir cS (A, P) |= ¬P(a). It is easy to see that this inference is not sup-
ported by eps2S (A, P).28 But, as emphasized earlier, the predictions of circ for
sentences containing modal operators are not in accordance with our intuitions.
Therefore, this restriction in Fact 2 to modal-free formulas is a blessing rather than a
curse.
However, with the results of Fact 2 we are only half of our way. Of course,
we do not want that eps2S allows additional inferences that destroy the nice pre-
dictions made by circumscription for the factive strong reading. Hence, we would
like to establish something like the following: if A, φ ∈ L0 then eps2S (A, P) |=
φ ⇔ cir cS (A, P) |= φ. This, however, does not hold. Recall that the order ≤ P
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 511
on which circ relies relates two states s1 = M1 , w1 , s2 = M2 , w2  ∈ S if (i)
V1 (P)(w1 ) ⊆ V2 (P)(w2 ) and (ii) V1 and V2 assign in w1 and w2 respectively the
same interpretation to all other elements of the nonlogical vocabulary. The condi-
tions imposed by 
and  are in some sense weaker: here we have to find pairs of
epistemic possibilities that only have to fulfill the first, (i), of these two conditions.
This results in stronger predictions for pragmatic inferences. In particular, in case A
contains other items of the non-logical vocabulary besides P one may now obtain
pragmatic information about the denotation of the items. The additional condition
(ii) has been added to the definition of circumscription to explicitly prevent that
circumscribing P also imposes additional restrictions on the interpretation of the
non-logical vocabulary apart from P. But how far is this relevant to model the
exhaustive interpretation of answers? In a context where it is known that if the
weather was fine Peter was there, exhaustive interpretation of the answer ‘Mary
(was there)’ to a question ‘Who was there?’ intuitively allows the inference that the
weather was not fine. Exactly for the reason described above, standard circumscrip-
tion cannot account for this inference. This can be taken as evidence that to model
the factive strong reading we do not want to exclude that minimizing P influences
the interpretation of other items. Hence, we should rather adopt the following order
for s1 = M1 , w1 , s2 = M2 , w2  ∈ S : s1 ≤2P s2 iffde f V1 (P)(w1 ) ⊆ V2 (P)(w2 ),
and use cir c2S (A, P) to model strong factive readings.29


cir c2S (A, P) =de f s ∈ S|s |= A ∧ ¬∃s  ∈ S : s  |= A ∧ s  ≤2P s

For this version of circumscription we can indeed establish not only an analogue
claim to Fact 2 but additionally the following result:

FACT 3. Let A be an element of L0 and P a predicate of L. Assume that for all

s ∈ S such that s |= A there is some s  ∈ cir c2S (A, P) such that s  ≤2P s. Then we
have:

∀φ ∈ L0 : eps2S (A, P) |= φ ⇒ cir c2S (A, P) |= φ.

Proof. We will show that for every s ∈ cir c2S (A, P) there is some s  ∈
esp2S (A, P) such that s ≡0 s  . This proves the claim. Take an arbitrary
s = M, w ∈ cir c2S (A, P). We choose s  = M  , w   as follows: (1) ∀t ∈
cir c2S (A, P)∃v  ∈ R  [w ] : t ≡0 M  , v  , (2) ∀v  ∈ R  [w  ]∃t ∈ cir c2S (A, P) :
t ≡0 M  , v  , and (3) s ≡0 s  .30 It follows immediately from (3) that we only
have to show that s  ∈ eps2S (A, P), thus (i) s  |=
A and (ii) ∀s  ∈ S : s  |=

A ⇒ [s 
s ∧ [s ∼
  
=
s  ⇒ s  ≺ s  ]] to conclude the proof of the
claim.
Ad (i). This is an immediate consequence of the definition of s  and the fact that
A ∈ L0 . Ad (ii). Take an arbitrary s  ∈ S such that s  |=
A. We show first that
it cannot be the case that s  
s  . If it were, we would have ∃v  ∈ R  [w  ]∀v  ∈
512 R. VAN ROOIJ AND K. SCHULZ

R  [w  ] : V  (P)(v  ) ⊆ V  (P)(v  ). By (1) it follows ∃v  ∈ R  [w  ]∀M, w ∈

cir c2S (A, P) : V (P)(w) ⊆ V  (P)(v  ). This contradicts the assumption that
∀s ∈ S : s |= A ⇒ [∃s  ∈ S : s  ∈ cir c2S (A, P)∧s  ≤2P s]. Hence, s  
s  . Now
we have to show also that the following cannot hold: s  ∼ =
s  ∧ s  ≺ s  . From
 ∼
s =
s it follows that ∀v ∈ R [w ]∃v ∈ R [w ] : V (P)(v  ) ⊆ V  (P)(v  ). Fur-
       

thermore, if s   s  then ∃v  ∈ R  [w ]∀v  ∈ R  [w  ] : V  (P)(v  ) ⊆ V  (P)(v  ).

Together, this gives ∃v  ∈ R  [w  ]∃v  ∈ R  [w  ] : V  (P)(v  ) ⊂ V  (P)(v  ). By (2)
it follows that ∃M, w ∈ cir c2S (A, P)∃v  ∈ R  [w ] : V  (P)(v  ) ⊂ V (P)(w),
what contradicts the definition of cir c2S (A, P). Hence, s  ∼ =
s  → s  ≺ s  . This
concludes the proof of (ii).

So far, we have introduced two pragmatic interpretation functions that seem to

model adequately the epistemic weak (eps1 ) and the epistemic strong reading (eps2 )
of Quantity 1-implicatures. The first function eps1 was motivated directly by Grice’s
theory of conversational implicatures and intended to describe conversational impli-
catures due to the maxims of Quality and the first sub-maxim on Quantity. eps2 was
obtained by strengthening eps1 with an additional principle to maximize the com-
petence of the speaker. We have seen that the factive inferences of eps2 are closely
related to the inferences of predicate circumscription, which provided a promising
description of the factive strong reading of Quantity 1-implicatures. In sum, our
approach meets all the general requirements any analysis of these implicatures has
to fulfill that we formulated at the beginning of this section.
Before we start to discuss the adequacy of this formalization for the implica-
tures of complex sentences not treated correctly by circumscription, some general
remarks are in order.
First, concerning the epistemic force with which an implicature is generated, a
central question in the debate is what determines which reading should be predicted
in a certain context. According to the approach developed here, the difference is
based on whether or not an additional assumption of competence can be made. For
the sentences discussed until now, we think that this assumption is made as long
as it is consistent with what the interpreter already knows and the assumption that
the speaker is obeying the maxims of Quality and Quantity1. Hence, we predict the
strong reading to occur in those contexts where such a competence assumption can
be made. Otherwise only the weak reading is obtained.
Secondly, one of the central advantages of any rigorous formalization is that
it clarifies the consequences of one’s ideas. This is also very true for the work
presented here. When we were discussing some examples for the predictions of
eps1 and eps2 , we made the simplifying assumption that we can identify constants
and individuals. Let us now drop this assumption. If we then, for instance, calculate
eps1S (P(e), P) in a context S with two individuals a and b where the speaker may
but need not know which individual is denoted by e, the states selected are those in
[{a, b}] and [{a, b, ab}] where the speaker does not know whether e denotes a or b.
This may at first sight not be a very intuitive result, but it is a consequence that is to
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 513
be expected given the assumptions underlying our formalization. If the interpreter is
interested in the true individuals that have property P, then a speaker who said P(e)
and knows who is denoted by e is withholding this relevant information from the
hearer. Therefore eps1 concludes that the speaker cannot have this knowledge. This
shows that our formalization may still miss relevant variables for the calculation of
implicatures – something to study in further work.

5. Complex Sentences
Fact 2 of the previous section shows that implicatures triggered by sentences of L0
that we could account for by means of circumscription can be described as well in
terms of our epistemic notion of pragmatic interpretation. But our richer machinery
allows us, additionally, to predict exhaustivity effects – particularly connected with
the beliefs of the speaker – that could not be accounted for in terms of circumscrip-
tion, or of G&S’s operator exh. This will be illustrated in the present section.

5.1. POSSIBILITY STATEMENTS

Consider again example (12), here repeated as (13), a statement that explicitly refers
to the belief state of the speaker.

(13) A: Who knows the answer?

B: Peter and possibly Mary.

In Section 3 we reported two readings for this sentence: one reading according
to which the speaker knows that Peter knows the answer, but does not know of any
other individual that he or she knows the answer, and a second one that says that the
speaker knows that Peter knows the answer, she does not know that Mary knows
the answer, but she does know that all other individuals besides Mary and Peter
do not know the answer. The application of circ to a sentence like P( p) ∧ P(m)
will predict the inference that Mary did not know the answer – which is obviously
inadequate. eps1S (P( p) ∧ P(m), P) correctly describes the first reading, including
the ‘scalar’ implicature from P(m) to ¬
P(m). Furthermore, by applying eps2S
we are also able to account for the second reading of the example.
What about examples involving a ‘scalar’ expression under the scope of a posi-
bility statement, like (P( p) ∨ P(m))? By minimization with respect to 
, we
conclude that for every individual c the speaker does not know c has property P.
We also infer that ¬
(P( p) ∨ P(m). When additionally minimizing with respect
to  , one obtains that the speaker even knows for each c that it does not have
property P, and that ¬  (P( p) ∧ P(m)) holds. Thus, we obtain the scalar impli-
cature from ‘or’ to ‘not and’, but now under the scope of the possibility operator.
When comparing these results with the observations of the localists Landman and
Chierchia on examples where the ‘implicature-triggering’ expression stands in the
514 R. VAN ROOIJ AND K. SCHULZ

scope of an existential quantifier, it turns out that we predict exactly the reading
they claimed a global approach cannot account for.

5.2. NEGATION
As has been discussed already in Section 3, negation is a problem for G&S’s
approach, whether or not we use a ‘scalar’ expression in its scope: both ¬P(a) and
¬(P(a) ∨ P(b)) receive by exh or circ the interpretation that no individual actually
has property P. From Fact 2 stated above it follows that negation is a problem for
eps2 as well.
To solve this problem, we propose elsewhere (van Rooij and Schulz, submitted)
to follow the suggestion of von Stechow and Zimmermann (1984) that for the ex-
haustive interpretation of negative sentences we should not minimize the extension
of P, but rather that of the complement of P, i.e., P̄, hence, to calculate cir cS (·, P̄)
instead. As we suggested already in our discussion of Section 3, however, we be-
lieve that negative sentences (just like positive ones) might have two pragmatic
readings. The first reading – which only some people seem to get – is correctly
described by cir cS (·, P̄) and, hence, also by eps2S (·, P̄). In terms of circ, however,
we were not able to account for the epistemic weak reading, according to which a
sentence like ¬P(a) gives rise to the inference that for every other individual the
speaker considers it possible that it has property P. But this reading can now be
described correctly by eps1S (·, P̄). To ‘explain’ the fact that for ‘negative’ sentences
the strong reading is more exceptional than for their positive counterparts, we sug-
gest that negation functions as a trigger signaling not only that it is the extension of
P̄ that is at issue, but also that the interpreter should (normally) not try to maximize
the speaker’s competence.

5.3. BELIEF
To account for implicatures of belief attributions we have to extend our formal
framework so that it can also express facts about the epistemic states of other agents.
In general this can be easily done: we add to our language L modal operators
i /i
and extend the states with respect to which this new language is interpreted with
accessibility relations Ri for every
i . A sentence
i φ should then be read as agent
i believes φ. The question how to extend our pragmatic interpretation functions is
somewhat more tricky. We will simplify things a bit here. Let us assume that we have
in D only finitely many individuals and a finite set N of names for these individuals
such that every name denotes exactly one individual. In this case there is another
language-oriented way to define our orderings 
and  . Let L(P) ⊆ L be the
sub-language defined by the BNF ϕ :: = P(a)(a ∈ N )|ϕ ∨ ϕ|ϕ ∧ ϕ. Furthermore,

L(P) is defined as the set {
φ|φ ∈ L(P)} and L(P) as {φ|φ ∈ L(P)}. In this
simplified setting, the following connection can be proven.
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 515
FACT 4. ∀s1 , s2 ∈ S:

s1 
s2 ⇔ ∀φ ∈ L(P) : s1 |=
φ → s2 |=
φ
s1  s2 ⇔ ∀φ ∈ L(P) : s1 |=  φ → s2 |=  φ

In order to account for implicatures triggered by expressions occurring in belief

attributions, we have to consider another language in terms of which we define the
ordering relations. The idea is now not to take L(P) as the basic language, but rather
for each agent j the language L j (P) =de f {
j φ|φ ∈ L(P)} ∪ { j φ|φ ∈ L(P)}. We
define two new orders 
, j and , j as follows: s1 
, j s2 iffde f ∀φ ∈ L j (P) :
s1 |=
φ → s2 |=
φ and s1  , j s2 iffde f ∀φ ∈ L j (P) : s1 |= φ → s2 |= φ.
Now take a belief attribution like ‘John believes that P(a) that we represent by

j P(a). One reading we obtain is by looking at the minimal state which verifies

j P(a) with respect to the order 
, j . On this reading we cannot infer much
more from the sentence than its semantic meaning. Another reading we obtain by
interpreting the belief attribution as the minimal state which verifies
j P(a) with
respect to the prioritized order defined in terms of 
, j and , j . According to
the resulting interpretation, the speaker knows that John believes that only a has
property P. It is this latter reading that accounts for the inferences in (1) and (8) as
reported in Sections 1 and 2 of this paper. In analogy with ‘scalar’ terms occurring
under possibility statements, we obtain the standard exhaustive interpretation, but
now under the scope of a belief operator. Hence, again we correctly predict those
readings that Landman (2000) and Chierchia (ms.) take to be problematic for global
approaches.

6. Conclusion
In this paper we introduced two related pragmatic interpretation functions for an-
swers to overt questions. The first function, eps1 , was motivated directly by Grice’s
theory of conversational implicature and intends to describe conversational impli-
catures due to the maxims of Quality and the first sub-maxim of Quantity. The
second, eps2 , was obtained by strengthening eps1 with an additional principle to
maximize the competence of the speaker. In both cases the definition makes crucial
use of the work of Halpern and Moses (1984) on ‘only knowing’ and its recent
generalization by van der Hoek et al. (1999, 2000).
These two functions predict two different pragmatic readings for answers to
overt questions. eps1 describes the conversational implicatures that are always
obtained if the speaker is taken to obey the maxim of Quality and the first sub-
maxim of Quantity. The predicted inferences have weak epistemic force and say,
roughly, that for certain stronger statements the speaker does not know whether
they are true. The interpretation function eps2 strengthens eps1 in the contexts
where the interpreter can make additional assumption about the competence of
516 R. VAN ROOIJ AND K. SCHULZ

the speaker. This strengthening has the result that some of the inferences of eps1
are now generated with a strong epistemic force, claiming that the speaker knows
that certain claims that are stronger than the statement made by the speaker are
not true. As these paraphrases show, the implicatures predicted by our approach
are closely related to many other descriptions of these inferences given in the
literature. The specific advantages of our proposal, we claim, are, on the one hand,
its rigorous formal outset that allows for clear, testable predictions, and, on the
other, that it provides a unified account of implicatures due to the first sub-maxim of
Quantity.
Apart from these points, this paper contributes in two other respects to the
research on conversational implicatures. First, in section 4 we have seen that
eps2 generalizes Groenendijk and Stokhof’s (1984) description of exhaustive in-
terpretation, which was already a very promising approach to a wide class of
implicatures. Thereby it links their proposal to Grice’s theory of conversational
implicatures and gives this more descriptive approach a conceptual, explana-
tory foundation. Second, the proposed formalization represents a strong argu-
ment against defenders of a local approach to conversational implicatures like
Landman (2000) and Chierchia (ms.). For a wide range of examples it falsifies
their claim that global approaches cannot account for the implicatures of complex
sentences.

Acknowledgements
Part of this paper was presented at the 8th Mathematics of Language conference
in Bloomington (2003), as well as at some other occasions. We are grateful to all
those who asked useful questions. We thank Johan van Benthem, Jeroen Groe-
nendijk, Martin Stokhof, Ede Zimmermann, and two anonymous referees for their
comments on our work on exhaustive interpretation over the last few years. In
particular, we would like to thank Paul Dekker for a last-moment proof-reading.
Robert van Rooij’s research for this paper has been made possible by a fellow-
ship of the Royal Netherlands Academy of Arts and Sciences (KNAW). Katrin
Schulz is financially supported by the NWO-project Logic meets Psychology:
non-monotonicity.

Notes
1. Part of this observation can also be found in Spector (2003).
2. Though we take it to be one of the advantages of Grice’s pragmatic theory that it can explain this
diversity of intuitions.
3. Clausal implicatures are another class of inferences Gazdar takes to be due to the first sub-maxim
of Quantity. We will come back to them in Section 4.
4. One may even argue, the most frequent ones.
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 517
5. Though they are not always discussed in the context of such a question.
6. We understand here under existence-quantifiers also ‘or’, which quantifies over propositions, and
modal existential quantifiers such as ‘possibly’.
7. Though Merin (1994) already observed that by a slight (though disputable) adaption of Gazdar’s
(1979) analysis, examples like (4) could be accounted for. As it turns out, a slight modification of
Horn’s (1972) analysis of scalar implicatures would do the trick as well. These modifications will
not be of great help, however, for most other complex sentences discussed in this paper. More
recently, Sauerland (2004) proposed yet another modification of traditional analyses of scalar
implicatures to account for (4) and (6). But also this analysis will not be able to account for the
set of data discussed in this paper.
8. But see also Wainer (1991) for a more explicit use of non-monotonic reasoning techinques.
9. The notation [·] F means that the relevant item is focussed, i.e., intonationally marked.
10. Notice, by the way, that what we called an exhaustive interpretation in this paper is explicitly
treated by Harnish (1976) as a Quantity1-implicature.
11. D stands for the domain of individuals. Even though the operation is described for n-ary predicates,
we simplify and assume B to be of type e, t.
12. A[P /P] is the theory that is obtained by replacing all occurrences of P in A with P .
13. However, the two approaches are not equivalent. One thing to notice is that G&S took exh to
be a description of an operation on semantic representations while circ(A,P) is an expression
in the object language. Second, one anonymous referee called our attention to the fact that the
two operations make indeed different predictions in case there are occurrences of the question-
predicate in the focus- or term-answer-part F. The circumscription of A w.r.t. P minimizess P in
all occurences of A = F(P). The operation of G&S does so for the background-occurrence only.
To see the differences, take the answer ‘Men that wear a hat’ to a question ‘Who wears a hat?’.
Circumscribing this answer has the result that the extension of ‘wears a hat’ is empty – which
is not the intuitive reading. exh correctly predicts that exactly those people wear a hat that are
men that wear a hat. In this paper we will assume that the question-predicate will not occur in the
focus or term-answer-part.
14. Given an at least-semantics for numerals and in the context of a question ‘How many apples did
John eat?’.
15. At least, in case the scalar item is not focussed in the question itself.
16. There is a strong fraction of semanticists that have argued that the output of the grammar are
meanings structured in focus and background. For circ we would only need the question-predicate
accessible in the context.
17. For an elaborate discussion see, among others, Groenendijk and Stokhof (1984), Wainer (1991),
and van Rooij and Schulz (submitted).
18. Where ‘?x[P(x)]’ represents the question ‘Who has property P?’.
19. For example, in Zimmermann (2000) the author distinguishes an open list reading for dis-
junctions/conjunctions ‘expressing undecidedness or uncertainty whether the list is exhaustive’
(p. 261), but he models this interpretation as a factive weak reading.
20. Consider, for instance, the derivation given by Levinson: ‘The speaker S has said s; if S was in
a disposition to [. . .] assert s  then he would be in breach of the first maxim of Quantity if he
asserted s. Since I the addressee assume that S is cooperating, and therefore will not violate the
maxim of Quantity without warning, I take it that S wishes to convey that he is not in a position
to state that the stronger s  holds and indeed knows that it does not hold.’ (Levinson, 1983,
pp. 134–135, slightly modified, italics added by the authors).
21. An interesting additional argument is due to Soames (1982). To avoid making some false predic-
tions, Gazdar (1979) has to assume that clausal implicatures are added to the beliefs of the
interpreter before scalar implicatures. But he does not give any independent motivation for
why the generation of implicatures has to be ranked this way. As Soames points out, if one
518 R. VAN ROOIJ AND K. SCHULZ

assumes that scalar implicatures have epistemic weak force and that the strong readings are due
to additional beliefs about the competence of the speaker, one can do without this additional
assumption.
22. Hence, ‘relevance’ is here interpreted as relevant to the question to which the interpreted sentence
is intended as answer.
23. Obviously, this interpretation function is an instance of interpretation in preferential structures
and, hence, the basis of a non-monotonic notion of entailment.
24. Gazdar’s (1979) idea, recently modified by Sauerland (2004), to generate strong
¬s  implicatures
(his scalar ones) only in as far as they are compatible with (the set of) weak implicatures of the
form
¬s  (his clausal ones) is, of course, closely related as well. However, Gazdar does not
motivate this strengthening by appealing to the informedness of the speaker.
25. Thus, eps2S (·) falls under the heading of Prioritized Circumscription.
26. At this point we straightforwardly extend the definition of cir c W to the operation cir c S on
states of our language L of modal predicate logic. This comes down to the following adapted
definition of ≤ P . We say for two states s1 = M1 , w1 , s2 = M2 , w2  that s1 ≤ P s2 if the
respective interpretation functions V1 and V2 agree in w1 and w2 on everything except possibly
the interpretation of P and V1 (P)(w1 ) ⊆ V2 (P)(w2 ). We implicitly used cir cS already in Section
3 when we discussed the predictions of cir c W for sentences referring to the epistemic state of
some agent as in example (1).
27. Hence, s ∗ is ‘constructed’ from s by substituting for the valuation V in all worlds of R[w] where
the extension of P is at least as large as in w the valuation of V in w .
28. Let S = M, w ∈ cir cS (P(a), P). It follows that ∃v ∈ R[w] : M, v|=P(a). Take s  = M, v.
Obviously s ∼ =
s  and s ∼= s  . Hence, if s ∈ eps2S (A, P) then s  ∈ eps2S (A, P) as well. How-

ever, s |= ¬ P(a).
29. Notice that cir c2 does not make the right predictions for answers like ‘If the weather was fine
Peter was there’ to the question ‘Who was there?’. But according to many of our informants these
answers are not interpreted exhaustively anyway and, thus, cir c2 should not be applied. Hence,
they do not necessarily constitute counterexamples for this approach.
30. The first two conditions are totally independent of the choice of s: for all s we chose the
same knowledge-state of the speaker for s  , the state consisting exactly of the elements in
cir c2S (A, P)(modulo ≡0 ).

References
Atlas, J.D. and Levinson, S.C., 1981, “It-clefts, informativeness and logical form,” in Radical Prag-
matics, P. Cole, ed., New York: Academic Press.
Chierchia, G., “Scalar implicatures, polarity phenomena, and the syntax/pragmatics interface,”
Manuscript, University of Milan.
Gazdar, G., 1979, Pragmatics, London: Academic Press.
Green, M., 1995, “Quantity, volubility, and some varieties of discourse,” Linguistics and Philosophy
18: 83–112.
Grice, H.P., 1967, “Logic and Conversation,” typescript from the William James Lectures, Harvard
University. Published in P. Grice (1989), Studies in the Way of Worlds, Cambridge, MA: Harvard
University Press, pp. 22–40.
Groenendijk, J. and Stokhof, M., 1984, “Studies in the semantics of questions and the pragmatics of
answers,” Ph.D. Thesis, University of Amsterdam.
Halpern, J.Y. and Moses, Y., 1984, “Towards a theory of knowledge and ignorance,” pp. 165–193 in
Proceedings of 1984 Non-Monotonic Reasoning Workshop, American Association for Artificial
Intelligence, New Paltz, NY.
EXHAUSTIVE INTERPRETATION OF COMPLEX SENTENCES 519
Harnisch, R.M., 1976, “Logical form and implicature,” pp. 313–391 in An Integrated Theory of
Linguistic Ability, T.G. Bever et al., eds., New York: Crowell.
Heim, I., 1982, “The semantics of definite and indefinite noun phrases,” Ph.D. Dissertation, University
of Massachusetts, Amherst.
Hirschberg, J., 1985, “A theory of scalar implicature,” Ph.D. Thesis, UPenn.
Horn, L., 1972, “The semantics of logical operators in English,” Ph.D. Thesis, UCLA.
Horn, L., 1989, A Natural History of Negation, Chicago: University of Chicago Press.
Kamp, H., 1981, “A theory of truth and semantic representation,” pp. 277–322 in Formal Methods in
the Study of Language, Groenendijk et al., eds., Amsterdam.
Landman, F., 2000, Events and Plurality: The Jeruzalem Lectures, Dordrecht: Kluwer.
Leech, G.N., 1983, Principles of Pragmatics, London: Longman.
Levinson, S.C., 1983, Pragmatics, Cambridge: Cambridge University Press.
Levinson, S.C., 2000, Presumptive Meanings. The Theory of Generalized Conversational Implicature,
Cambridge, MA: MIT Press.
Matsumoto, Y., 1995, “The conversational condition on Horn scales,” Linguistics and Philosophy 18:
21–60.
McCarthy, J., 1980, “Circumscription – a form of non-monotonic reasoning,” Artificial Intelligence
13: 27–39.
Merin, A., 1994, Decision–Theoretic Pragmatics, Lecture Notes read in Course BA 5, Copenhagen
Business School, Denmark.
Sauerland, U., 2004, “Scalar implicatures of complex sentences,” Linguistics and Phiolosophy 27,
367–391.
Schulz, K., 2003, “You may read it now or later,” Master Thesis, University of Amsterdam.
Soames, S., 1982, “How presuppositions are inherited: A solution to the projection problem,”
Linguistic Inquiry 13: 483–545.
Spector, B., 2003, “Scalar implicatures: Exhaustivity and Gricean reasoning?,” in Proceedings of the
ESSLLI 2003 Student session, B. ten Cate, ed., Vienna.
van Benthem, J., 1989, “Semantic parallels in natural language and computation,” pp. 331–375 in
Logic Colloquium ’87, H. D. Ebbinghaus et al., eds., Amsterdam: Elsevier Science Publishers.
van der Hoek, W., Jaspers, J., and Thijsse, E., 1999, “Persistence and minimality in epistemic logic,”
Annals of Mathematics and Artificial Intelligence 27: 25–47.
van der Hoek, W., Jaspers J., and Thijsse, E., 2000, “A general approach to multiagent minimal
knowledge,” pp. 254–268 in Proceedings JELIA 2000, M. Ojeda-Aciego et al., eds., Heidelberg:
Springer-Verlag.
van Rooij, R. and Schulz, K., “Pragmatic meaning and non-monotonic reasoning: The case of ex-
haustive interpretation,” submitted, University of Amsterdam.
von Stechow, A. and Zimmermann, T.E., 1984, “Term answers and contextual change,” Linguistics
22: 3–40.
Wainer, J., 1991, “Uses of nonmonotonic logic in natural language understanding: Generalized im-
plicatures,” Ph.D. Thesis, Pennsylvania State University.
Zimmermann, T.E., 2000, “Free choice disjunction and epistemic possibility,” Natural Language
Semantics 8: 255–290.