Linguist and Philos (2009) 32:207–244

DOI 10.1007/s10988-009-9059-0

RESEARCH ARTICLE

Counterfactuals, correlatives, and disjunction

Luis Alonso-Ovalle

Published online: 8 July 2009

Ó Springer Science+Business Media B.V. 2009

Abstract The natural interpretation of counterfactuals with disjunctive anteced-

ents involves selecting from each of the disjuncts the worlds that come closest to
the world of evaluation. It has been long noticed that capturing this interpretation
poses a problem for a minimal change semantics for counterfactuals, because
selecting the closest worlds from each disjunct requires accessing the denotation
of the disjuncts from the denotation of the disjunctive antecedent, which the
standard boolean analysis of or does not allow (Creary and Hill, Philosophy of
Science 43:341–344, 1975; Nute, Journal of Philosophy 72:773–778, 1975; Fine,
Mind 84(335):451–458, 1975; Ellis et al. Journal of Philosophical Logic 6:335–
357, 1977). This paper argues that the failure to capture the natural interpretation
of disjunctive counterfactuals provides no reason to abandon a minimal change
semantics. It shows that the natural interpretation of disjunctive counterfactuals is
expected once we refine our assumptions about the semantics of or and the logical
form of conditionals, and (i) we assume that disjunctions introduce propositional
alternatives in the semantic derivation, in line with independently motivated
proposals about the semantics of or (Aloni, 2003a; Simons, Natural Language
Semantics 13:271–316, 2005; Alonso-Ovalle, Disjunction in Alternative Seman-
tics. PhD thesis, 2006); and (ii) we treat conditionals as correlative constructions,
as advocated in von Fintel (1994), Izvorski (Proceedings of NELS 26, 1996),
Bhatt and Pancheva (2006), and Schlenker (2004).

Keywords Disjunction
Minimal change semantics for conditionals

Alternative semantics

L. Alonso-Ovalle (&)
University of Massachusetts Boston,
100 Morrissey Boulevard,
Boston, MA 02125, USA
e-mail: [email protected]

123
208 L. Alonso-Ovalle

1 The problem

Consider the following scenario: the summer is over and you and I are visiting a
farm. The owner of the farm is complaining about last summer’s weather. To give
us an example of its devastating effects, he points to the site where he used to grow
huge pumpkins: there is a bunch of immature pumpkins and many ruined pumpkin
plants. He then utters the counterfactual in (1):

(1) If we had had good weather this summer or the sun had grown cold, we would
have had a bumper crop.
(A variation on an example in Nute 1975.)

We conclude, right then, that there is something strange about this farmer. We have
a strong intuition that the counterfactual in (1) is false: if we had had good weather
this summer, he would have had a good crop, but we know for a fact—and we
assume that the farmer does too—that if the sun had grown cold, the pumpkins,
much as everything else, would have been ruined.
The problem that this article deals with is that a standard minimal change
semantics for counterfactuals (Stalnaker 1968; Stalnaker and Thomason 1970;
Lewis 1973) fails to capture this intuition: if the usual boolean semantics for or is
assumed, a standard minimal change semantics for counterfactuals predicts the
counterfactual in (1) to be true.
To see why this is so, we need to adopt a simple minimal change semantics for
counterfactuals.
In a minimal change semantics, evaluating whether a counterfactual is true
involves checking whether the consequent is true in the world(s) in which the
antecedent is true that are as close as possible to the actual world. We will assume
here that would counterfactuals are true in the actual world if and only if the
consequent is true in all the worlds where the antecedent is true that differ as little as
possible from the actual world.1
To state these truth-conditions formally, we will take the interpretation of
counterfactuals to be relative to a relation of comparative similarity defined for the
set of accessible worlds W (which, we will assume, is the set of all possible worlds).
Following Lewis (1973, p. 48), we will take any admissible relation of comparative
similarity w to be a weak ordering of W, with the world w alone at the bottom of
the ordering (w is more similar to w than any other world w0 ).2 We will also make
what Lewis calls ‘the Limit Assumption’: for any world w and set of worlds W we
assume that there is always at least one world w0 in W that come closest to w. Under

1
For the differences between the minimal change semantics presented in Lewis (1973) and Stalnaker
(1968), see Nute (1984), which provides an overview of the different flavors minimal change semantics
come in.
2
A weak ordering is a relation that is transitive (for any worlds w0 ; w00 ; w000 , whenever w0 w w00 and
w00 w w000 , then w0 w w000 ) and strongly connected (for any worlds w0 and w00 , either w0 w w00 or
w00 w w0 ). Unlike in a strong ordering, ties are permitted (two different elements can stand in the relation
to each other). ‘Being as old as’, ‘being at least as far from Boston as’ are naturally interpreted as weak
orderings. In our metalanguage,‘w0 w w00 ’ is meant to convey that w0 is at least as similar to w as w00 is.

123
Counterfactuals, correlatives, and disjunction 209

these assumptions, the semantics of would counterfactuals can be formalized by

means of a class selection function f that picks up for any world of evaluation w,
any admissible relation of comparative similarity w , and any proposition p, the
worlds where p is true that come closest to w.

(2) For any proposition p, worlds w, w0 , and any relation of relative similarity ,
fw ðpÞðw0 Þ $ ½pðw0 Þ & 8w00 ½pðw00 Þ ! w0 w w00 

We can now state the truth-conditions of would-counterfactuals as follows: a would

counterfactual is true in a world w (with respect to an admissible ordering) if and
only if all worlds picked up by the class selection function (all the closest worlds to
w in which the antecedent is true) are worlds in which the consequent is true.

(3) ½½If /; then would w ðwÞ , 8w0 ½ fw ð½½/Þðw0 Þ ! ½½wðw0 Þ

Let us see now why, contrary to our intuitions, this semantics predicts the
counterfactual in (1) (repeated in (4) below) to be true in the situation we started the
discussion with.

(4) If we had had good weather this summer or the sun had grown cold, we would
have had a bumper crop.
(A variation on an example in Nute 1975.)

The semantics in (3) predicts the counterfactual in (4) to be true in the actual
world with respect to an admissible ordering if and only if the worlds in the
proposition expressed by the antecedent of the conditional in (4) that come the
closest to the actual world in the ordering are all worlds where we have a bumper
crop. What is the proposition expressed by the antecedent of the conditional in (4)?
The disjunction in the antecedent of the counterfactual in (4) operates over the
propositions in (5a) and (5b).

(5) a. ½½we had had good weather this summer ¼ kw. good-weatherw
b. ½½the sun had grown cold ¼ kw. grow-coldw (s)

Under the standard boolean semantics of or, the proposition expressed by the
if-clause is the union of the set of worlds where the summer weather was good and
the set of worlds where the sun grows cold3 :

(6) ½½If we had had good weather this summer or the sun had grown cold
¼ kw. good-weatherw _ grow-coldw (s)

3
In the metalanguage, I make use of a two-sorted language (Gallin 1975). World arguments are sub-
scripted. For the purposes of illustration, I take good-weather to be a predicate of worlds. The contri-
bution of tense and mood is ignored. I will omit the superscript indicating that the interpretation function
is relative to a relative similarity ordering when the ordering bears no effect on the interpretation.

123
210 L. Alonso-Ovalle

According to the semantics in (3), then, the counterfactual in (1) is predicted to be

true in the actual world with respect to an admissible ordering of worlds if and only
if the closest worlds where the proposition in (6) is true are all worlds where we
have a bumper crop.

(7) ½½ð4Þ ðw0 Þ , 8w0 ½fw0 ð½½ð6ÞÞðw0 Þ ! ½½we have a bumper cropðw0 Þ

The problem is that the truth-conditions in (7) are too weak. In the scenario we
started the discussion with, the counterfactual in (4) is evaluated with respect to an
intuitive notion of relative similarity according to which the possible worlds where
the sun grows cold are more remote from the actual world than the possible worlds
where we have a good summer. (More actual facts have to be false in a world where
the sun grows cold than in a world where we have good summer weather.)
This similarity relation is represented in Fig.1, where each circle represents a set
of worlds that are equally close to the actual world, the dotted line surrounds the
worlds where the sun gets cold, and the solid line the worlds where there is good
weather this summer. With respect to this relation of comparative similarity, none of
the worlds in which the proposition in (6) is true where the sun grows cold can
count closer to the actual world than the worlds where we have a good summer.
The selection function, therefore, only returns worlds where we have a good summer.
Since in all the closest worlds where we have a good summer it is true that we
have a bumper crop, the counterfactual in (4) is predicted to be true, contrary to our
intuition.
When coupled with the standard boolean semantics for or, a minimal change
semantics for counterfactuals does not capture by itself the natural interpretation of
would counterfactuals with disjunctive antecedents: would disjunctive counterfac-
tuals are naturally interpreted as claiming that the consequent is true in the closest
worlds in each of the disjuncts; under a minimal change semantics, however they
are predicted to claim that their consequent is true in the closest worlds in the union
of the disjuncts—at least if the standard boolean semantics for or is assumed.

Fig. 1 The problem of disjunctive antecedents

123
Counterfactuals, correlatives, and disjunction 211

The problem was first pointed out in the philosophical literature in the mid-
seventies, and taken as an argument against Lewis’ minimal change semantics for
counterfactuals (Lewis 1973).4 There have been many reactions to it since then
(see Nute 1984; Nute and Cross 2001 for an overview). For some researchers, the
problem justifies abandoning a minimal change semantics for counterfactuals
altogether: Ellis et al. (1977), for instance, propose abandoning a possible world
semantics for subjunctive conditionals, Warmbr
od (1981) advocates adopting a
context-dependent downward monotonic semantics, and, most recently, Herburger
and Mauck (2007) propose developing an event-based semantics. There might be
reasons to abandon a minimal change semantics for counterfactuals, but the goal
of this article is to show, as van Rooij (2006) does, that the failure to capture the
natural interpretation of disjunctive counterfactuals need not be one.
This article argues that the natural interpretation of disjunctive counterfactuals is
in fact expected, even within a minimal change semantics, once we refine our
assumptions about the semantics of or and the logical form of conditionals, and (i)
we assume that disjunctions introduce propositional alternatives in the semantic
derivation, in line with independently motivated recent proposals about the
semantics of or (Aloni 2003a,b; Simons 2005; Alonso-Ovalle 2006); and (ii) we
treat conditionals as correlative constructions (as advocated in von Fintel 1994,
Izvorski 1996, Bhatt and Pancheva 2006, and Schlenker 2004).
The article is organized as follows: Sect. 2 presents the analysis, it shows that once
counterfactuals are interpreted as correlatives, and a Hamblin-style semantics for or is
adopted, the natural interpretation of would disjunctive counterfactuals is expected;
Sect. 3 shows that there are reasons to believe that the natural interpretation of dis-
junctive counterfactuals is not due to a downward entailing inference; Sect. 4 discusses
some problems for the derivation of their interpretation as a conversational impli-
cature, Sect. 5 discusses further two assumptions of the analysis, and Sect. 6 concludes
the discussion by introducing some issues for further research.

2 The analysis

The analysis of counterfactuals with disjunctive antecedents that we will entertain

makes two novel assumptions: the first has to do with the semantics of disjunction,
and the second with the logical form of conditionals. In the illustration of the
problem in Sect. 1, we have taken for granted the standard boolean semantics for or.
In line with recent work on the semantics of natural language disjunction (Aloni
2003a; Simons 2005; Alonso-Ovalle 2006), we will assume, instead, that or
introduces into the semantic derivation a set of propositional alternatives. The
proposal will be cast in Sect. 2.1 within a Hamblin-style alternative semantics. We
will then adopt in Sect. 2.2 a compositional analysis of conditionals that assumes
that they are correlative constructions. The natural interpretation of disjunctive
counterfactuals is shown to follow from these two assumptions.

4
See, among others, Creary and Hill (1975), Nute (1975), Fine (1975), and Ellis et al. (1977).

123
212 L. Alonso-Ovalle

2.1 Disjunctive antecedents in an alternative semantics

We start by laying out the Hamblin-style alternative semantics in which the analysis
will be cast.5
In a Hamblin semantics, expressions of type s are mapped to sets of objects in Ds .
Most lexical items denote singletons containing their standard denotations: the
proper name in (8a) is mapped to a singleton containing an individual, and the verbs
in (8b–8c) are mapped to a singleton containing a property.

(8) a. ½½Sandy ¼ fsg

b. ½½sleep ¼ fkx:kw.sleepw ðxÞg
c. ½½see ¼ fky:kx:kw:seew ðx; yÞg

We will only be concerned for the most part with the way expressions combine by
functional application. In an alternative semantics, functional application is defined
pointwise, as in (9) below: to combine a pair of expressions denoting a set of objects
of type hr; si and a set of objects of type r, every object of type hr; si applies to
every object of type r, and the outputs are collected in a set.

(9) The Hamblin Rule

If ½½a  Dhr;si and ½½b  Dr ; then
½½aðbÞ ¼ f c 2 Ds j 9a 2 ½½a9b 2 ½½bðc ¼ aðbÞÞ g (Hamblin 1973)

Within this framework, it is natural to assume that or introduces into the semantic
derivation the denotation of the disjuncts as alternatives6 :

ð10Þ

With these assumptions in mind, let us consider again the disjunctive coun-
terfactual in (1), repeated in (11) below:

5
Charles Leonard Hamblin developed an alternative semantics in his analysis of questions (Hamblin
1973). A Hamblin semantics has been invoked in the analysis of focus (Rooth 1985; Rooth 1992), and
indeterminate pronouns (Ramchand 1997; Hagstrom 1998; Kratzer and Shimoyama 2002; Alonso-Ovalle
and Menéndez-Benito 2003)
6
In what follows, we will represent the internal structure of disjunctions at LF as flat. It is inmaterial for
the present analysis whether this assumption is justified, but the reader is referred to Munn (1993) and den
Dikken (2003), where the internal structure of disjunctive constituents is assumed not to be flat.
Although the rule in (10) associates or with set union, the role of or is different from the role of or
under its standard boolean analysis. Notice that, in the rule in (10), B and C denote sets of semantics
objects. Or, according to this rule, simply collects those semantics objects into a set, rather than mapping
those objects into their boolean join.

123
Counterfactuals, correlatives, and disjunction 213

(11) If we had had good weather this summer or the sun had grown cold, we would
have had a bumper crop.
(A variation on an example in Nute 1975.)

We will assume, as we did when illustrating the problem in Sect. 1, that the dis-
junction within the antecedent operates over two propositions. The relevant inter-
pretable structure of the disjunction within the if-clause is the one in (12) below:

ð12Þ

In the discussion of the example in Sect. 1, we took for granted the textbook
semantics for or, under which IP1 denotes a proposition (the set of worlds in which
at least one of the disjuncts is true). We will now drop that assumption. In the
Hamblin-style semantics that we are assuming, IP1 denotes a set of propositions. Or
operates over the denotation of IP2 and IP3 . IP2 denotes the singleton containing the
proposition that we have good weather this summer, and IP3 the singleton con-
taining the proposition that the sun grows cold. IP1 denotes the union of these two
sets: the set containing the proposition that we have good weather this summer and
the proposition that the sun grows cold.

(13) a. ½½we had had good weather this summer ¼ fkw:good-weatherw g

b. ½½the sun had grown cold ¼ fkw:grow-coldw (s)g


kw:good-weatherw ;
c. ½½IP1  ¼
kw:grows-coldw ðsÞ
The interpretation of disjunctive counterfactuals illustrated in Sect. 1 involves
checking whether the consequent holds in the closest worlds in each disjunct. For
this interpretation to be captured in a standard minimal change semantics, the
selection function needs to apply to each of the propositions that or operates over.
Under the standard semantics for or, the selection function can only see the prop-
osition that the whole disjunction denotes. That causes the problem. Under the
present setup, however, the denotation of the disjuncts can be easily retrieved from
the denotation of the whole disjunction. In principle, the semantic composition of
conditionals can now access each disjunct on its own to select the worlds that come
closest to the world of evaluation. How does the set of propositions in the ante-
cedent contribute to the semantic composition of conditionals? To see how the
semantics can select the closest worlds in each of the disjuncts we need to say
something about the logical form of conditionals. We will see in the next section
that a natural answer emerges once we assume that conditionals are correlative
constructions.

123
214 L. Alonso-Ovalle

2.2 Conditionals as correlatives

In line with much syntactic and semantic work on conditionals (von Fintel 1994;
Izvorski 1996; Bhatt and Pancheva 2006; Schlenker 2004), we will assume that
conditionals are correlative constructions.
The analysis that I present next builds on work on the semantics of correlatives
by Veneeta Dayal (Srivastav 1991a,b; Dayal 1995,1996). There are two main
components to it: first, the consequent of a conditional is analyzed as denoting a
property of propositions, much as the main clause of a correlative denotes a property
of individuals (this is possible once then is analyzed as a propositional anaphor);
and, second, if-clauses are analyzed as universal quantifiers ranging over proposi-
tions, much as antecedents of correlatives universally quantify over individuals.

2.2.1 Then as a resumptive pronoun

In correlatives, a relative clause adjoined to the matrix clause provides an anaphoric

pronoun inside the main clause with an antecedent.

ð14Þ

The construction is illustrated in (15) with a few examples from Hindi:

(15) a. [jo laRkii khaRii hai ]i voi lambii hai

which girl standing be-present she tall be-present
‘Which girl is standing, that one is tall.’ (Dayal 1996, p. 188).

b. [jo laRkiyaaN khaRii haiN ]i vei lambii haiN

which girls standing be-present they tall be-present
‘Which girls standing are, they are tall.’ (Dayal 1996, p. 192).

c. [jo do laRkiyaaN khaRii haiN ]i vei lambii

which two girls standing be-present they tall
haiN
be-present
‘Which two girls are standing, they are tall.’ (Dayal 1996, p. 192)

In conditionals, then has been analyzed as a resumptive pronoun that picks up the
denotation of the if-clause as its antecedent, much as in other types of correlatives a
pronoun ranging over individuals picks up the denotation of the relative clause that
serves as its antecedent, as illustrated in (16) below (Iatridou 1991a,b, 1994; von
Fintel 1994; Hegarty 1996). We will follow this analysis.

123
Counterfactuals, correlatives, and disjunction 215

ð16Þ
conditionals correlatives

As other natural language quantifiers do, modals range over a contextually

supplied domain. We will capture this contextual dependency by assuming that they
take as an argument a pronoun ranging over propositions (von Fintel 1994). Then, I
want to assume, is one such pronoun, which is in complementary distribution with a
covert counterpart.7 Its interpretation, like the interpretation of other pronouns, is
provided by the variable assignment. At LF, then bears an index. In the type of
alternative semantics that I am assuming, then denotes a singleton containing the
proposition that the variable assignment maps its index to8 :

(17) ½½thenh7;hs;tii g ¼ fgðh7; hs; tiiÞg

What is the semantic import of the anaphoric link between the if-clause and the
pronoun then in the main clause? Dayal (1996) assumes that the anaphoric relation
between the relative and the pronoun in the main clause is a case of variable
binding. The antecedent of a correlative is a generalized quantifier, which takes as
an argument the property that results from abstracting over the pronoun in the main
clause, as illustrated below with a plural correlative.9

(18) jo laRkiyaaN khaRii haiN ve lambii haiN

which girls standing be-PR they tall be-PR
‘Which girls are standing, they are tall.’ Dayal (1996, p. 192)

7
This is a simplification. Schlenker (2004) argues that then is really doubling an implicit argument.
8
A variable assignment is assumed to be a function from pairs of natural numbers and type specifications
to entities of the right type. In what follows, I will use a slightly different notation: instead of writing
‘thenh7;hs;tii ’, I will write ‘then7hs;ti .’
9
In the illustration in (19), I simplify a bit for expository reasons. Srivastav (1991a, p. 668) assumes that
the domain of individuals is closed under sum formation and treats the antecedent of a plural correlative
as a universal quantifier whose domain of quantification is the supremum of the set of girls who are
standing, as in (i) below. If the predicate abstract with which this quantifier combines is distributive, as in
the example in (19), the resulting truth conditions are equivalent.
(i)
Phe;ti :8x½ðx ¼ x:ð girlðxÞ & standðxÞÞ ! P ðxÞ (see Srivastav 1991a, p. 668)

123
216 L. Alonso-Ovalle

ð19Þ

Once then is analyzed as a propositional anaphor, we can analyze the con-

sequent of a conditional as denoting a property of propositions, much as the con-
sequent of a correlative denotes a property of individuals.
Consider again, for instance, the counterfactual that we opened this article with,
repeated in (20) below:

(20) If we had had good weather this summer or the sun had grown cold, (then)
we would have had a bumper crop.
(A variation on an example in Nute 1975.)
We will assume that the interpretable structure of its consequent is as in (21).
ð21Þ

Would is assumed to be a function that takes two propositions p and q as argu-

ments and returns (the singleton) containing the proposition that is true in a world w
if and only if the closest worlds to w where p is true are all worlds where q is true.

(22) ½½would;g ¼ fkphs;ti :kqhs;ti :kw:8w0 ½fw ðpÞðw0 Þ ! qðw0 Þg

With respect to any admissible similarity relation , the LF in (21) denotes the
singleton containing the proposition that is true in a world w if and only if the
worlds in the proposition that then takes as its antecedent that come closest to w in
the relevant ordering are all worlds where we have a bumper crop.
8 2 39
< fw ðgð5hs;ti ÞÞðw0 Þ =
(23) ½½(21);g ¼ kw:8w0 4 ! 5
: ;
have-a-bumper-cropw0 ðweÞ

By abstracting over then in (21), we end up with (a set containing) a function from
propositions to propositions that maps any proposition p into the proposition that is
true in a world w if and only if the p-worlds that come closest to w are all worlds
where we have a bumper crop.10

10
For ease of exposition, I assume that the lambda abstraction is represented at LF by means of an index,
as in Heim and Kratzer (1998). From now on, I will use the expression ‘the p-worlds’ to refer to the
worlds where a certain proposition p is true.

123
Counterfactuals, correlatives, and disjunction 217

ð24Þ

2.2.2 If-clauses as quantifiers over propositions

What is the denotation of the if-clause? The antecedent of a correlative denotes, under
Dayal’s analysis, a generalized quantifier: a property of properties of individuals. The
relative in the example in (19) denotes a property of properties of individuals that holds
of any property P if and only if P holds of every individual which is a girl and is
standing. Under the Hamblin-style analysis of disjunction that we are assuming, we
can treat the if-clause in parallel to Dayal’s analysis as denoting a property of prop-
erties of propositions which holds of any property of propositions Phhs;ti;hs;tii if and only
if Phhs;ti;hs;tii holds of every proposition in the set of propositional alternatives denoted
by the antecedent (a singleton, in the case of non disjunctive antecedents; a set that
contains all the atomic propositional disjuncts in the case of disjunctive antecedents.)
The if-clause in (25) denotes, under this analysis, a property of properties of propo-
sitions that holds of any property of propositions Phhs;ti;hs;tii in a world w if and only if
Phhs;ti;hs;tii holds in w of the proposition that we have good weather this summer and of
the proposition that the sun grows cold.

ð25Þ

The denotation of the whole conditional can be now calculated by applying the
denotation of the if-clause to the denotation of the consequent.

123
218 L. Alonso-Ovalle

(26) ½½ð20Þ;g ¼ ½½ð25Þ;g ð½½ð24Þ;g Þ

Under the present analysis, the sentence in (20) denotes, for any admissible
ordering, the singleton containing the proposition that is true in a world w if and
only if all the closest worlds to w in which we have good weather are worlds where
we have a bumper crop, and all the closest worlds to w in which the sun grows cold
are worlds where we have a bumper crop.11 With respect to an ordering that makes
every world where the sun grows cold less similar to the actual world than any
world where we had good weather this summer, the sentence in (20) expresses a
proposition that is false in the actual world, because none of the closest worlds to the
actual world where the sun grows cold are worlds where we have a good crop. The
intuition reported in Sect. 1 is then captured.12

2.2.3 The universal force

The analysis of disjunctive counterfactuals that we have entertained has two main
components: (i) the assumption that or introduces a set of alternatives into the
semantic derivation allows for the selection function to have access to each of the
disjuncts; and (ii) the assumption that conditionals, much like other correlatives,
convey universal quantification over the type of entities described by the antecedent
captures the intuition that a disjunctive counterfactual of the form of (27a) is
equivalent to the conjunction of counterfactuals of the type of (27b) and (27c).

(27) a. If / or w, then would n.

b. If /, then would n.
c. If w, then would n.

Before concluding this section, I would like to address the source of this second
meaning component.
We have treated if-clauses as universal quantifiers ranging over propositions.
Dayal (1996) treats the antecedent of correlatives as definite descriptions: the
antecedent of the plural correlative in (28) below denotes the maximal sum of girls
that are standing. Since the main clause is associated with a distributive property,
the whole correlative is predicted to be true if and only if every girl is tall.

(28) a. jo laRkiyaaN khaRii haiN ve lambii haiN

which girls standing be-PR they Tall be-PR
‘Which girls are standing, they are tall.’ (Dayal 1996, p. 192)

11
Notice that, in the present analysis, the non-monotonicity of questions is due to the semantics of the
modal in the consequent.
12
In the analysis that we have just presented, then is a bound variable pronouns ranging over the
propositions in the domain set up by the if-clause. An anonymous reviewer points out that the alternatives
introduced by a disjunction in a modal context can also be suitable antecedents for then in intersentential
anaphora cases such as (i) below (example due to the reviewer):
(i) Let us suppose that John Kerry or Hillary Clinton became president. Then Bill Clinton would / might
become ambassador to the United Nations.

123
Counterfactuals, correlatives, and disjunction 219

b:

If the predicate abstract that if-clauses combine with is always distributive (if
it is true of a sum of propositions if and only if it is true of all the atomic parts of the
sum), we get, in the end, the same truth-conditions that we got by assuming a
universal quantifier over propositions. Should we assume that if-clauses denote
sums of propositions, then? To finish, I would like to consider a potential argument
against doing so and conclude, tentatively, that it is not a knock-down one.
Consider what happens when we embed a would counterfactual under what is
presumably a wide scope negation:

(29) It is plain false that Hitler would have been pleased if Spain had joined Germany
or the U.S.
(Kratzer, p.c., a variation on an example in Nute 1980 (p. 157))

If the if-clause is a universal quantifier over propositions, the sentence in (29) is

predicted to be true if and only if it is false that both counterfactuals below are true:

(30) a. Hitler would have been pleased if Spain had joined Germany.
b. Hitler would have been pleased if Spain had joined the U.S.

This is, of course, compatible with one of them being true. The possibility of
continuing (29) as in (31) shows that this is the case.

(31) . . .There is enough evidence showing that he might have objected to Spain
joining the U.S. If she had joined Germany, he would have been pleased,
of course. (Kratzer, p.c.)

If if-clauses denoted sums of propositions, and the predicate abstracts associated

with the consequents were distributive, the disjunctive counterfactual in (29) could
also in principle be true if the predicate abstract is not true of all the propositional
alternatives introduced by or (but just of one of them.) However, plural definite
descriptions are known to interact with negation in a peculiar way: the sentence in
(32) conveys that Sandy saw none of the cats, not just that Sandy didn’t see every
cat.

(32) Sandy didn’t see the cats.

123
220 L. Alonso-Ovalle

To capture this, a ‘homogeneity’ presupposition is usually invoked (Loebner 1998;

Schwarzschild 1994). Beck (2001) formulates homogeneity as follows (where P is a
predicate of atomic individuals,  P a pluralized distributive predicate, and A a
plurality):

(33) PðAÞ ¼ 1 iff 8x½x 2 A ! PðxÞ

PðAÞ ¼ 0 iff 8x½x 2 A ! :PðxÞ
(undefined otherwise)

One could argue that disjunctive counterfactuals do not seem to behave like
plural definite descriptions, after all, by pointing out that the sentence in (29) does
not convey an homogeneity presupposition. Yet, for the argument to go through, one
would have to show, as an anonymous reviewer points out, that the homogeneity
presupposition of definite descriptions is present even with the type of wide scope
negation in (29). This is not that clear, since the sentence in (34) does not seem to
convey that Sandy saw none of the cats.

(34) It is plain false that Sandy saw the cats.

I will assume in the rest of the paper that if-clauses contribute the universal
quantification over propositions, and remain agnostic as to whether this force is best
captured by means of a universal propositional quantifier or a sum operator.

3 Downward entailingness

To capture the intuition reported in Sect. 1, the analysis presented in Sect. 2 moves
beyond the standard boolean semantics of or. This move allows us not to give up on
a minimal change semantics for counterfactuals. However, it is easy to see that the
intuition reported in Sect. 1 is in fact expected under the standard boolean analysis
of or if a material conditional analysis for conditionals is assumed. For suppose, for
the sake of the argument, that would counterfactuals were to be analyzed as strict
conditionals (material conditionals under the scope of a necessity operator (Lewis
1973, p. 4)), as in (35) below, where the domain of quantification of would is fixed
with the help of a variable C ranging over functions of type hs; hs; tii:
2 3
ð½½/ðw0 Þ & gðCÞðwÞðw0 ÞÞ
(35) ½½If /; would ðCÞ wg ¼ kw:8w0 4 ! 5
0
½½wðw Þ

If would counterfactuals were strict conditionals, the counterfactual in (36), for

instance, would express the proposition p that is true in a world w if and only if all
accessible worlds from w (say, all worlds w0 in which the laws of nature of w hold)
in which kangaroos have no tails are worlds where they topple over.

(36) If kangaroos had no tails, they would topple over (Lewis 1973, p. 1).

123
Counterfactuals, correlatives, and disjunction 221

Consider now what the truth-conditions of a would disjunctive counterfactual would

be like under the standard boolean analysis of or. With respect to the assignment of
a domain of accessible worlds A, a disjunctive counterfactual would be true in a
world w if and only if all worlds that are in at least one of the disjuncts and that are
also in A are worlds where the consequent is true—if and only if the proposition
expressed by the consequent is a subset of the set of worlds that belong to the union
of the propositions expressed by the disjuncts and to A.

(37) ½½If / or w; would ðCÞ ng ¼

220 1 3 3
½½/ðw0 Þ
kw:8w0 4 4 @ _ A& gðCÞðwÞðw0 Þ 5 ! ½½nðw0 Þ 5
½½wðw0 Þ

Under these truth-conditions, if a disjunctive counterfactual of the form in (37) is

true in a world w with respect to a domain of accessible worlds A, both counter-
factuals of the form in (38) must be true in w—assuming that they are evaluated
with respect to the same set of accessible worlds, that is—because, in virtue of the
transitivity of the subsethood relation, if a set X is a subset of a set Y, any subset of
X must also be a subset of Y.13

(38) a. If / then would n.

b. If w then would n.

Under the strict conditional analysis, substituting a certain proposition p in the

antecedent with any of the subsets of p preserves truth: the antecedent of coun-
terfactuals is a downward entailing environment. The interpretation of would
counterfactuals illustrated in Sect. 1 can be then captured as a downward entailing
inference. But then, wouldn’t the problem of capturing the natural interpretation of
disjunctive counterfactuals be solved by adopting a strict conditional analysis? This
section argues that it would not be.
The discussion is organized as follows. We will first consider the status of the
inference reported in Sect. 1 in the close-to-downward entailing semantics for
counterfactuals presented in von Fintel (1999). We will see that the inference is
predicted to be as context dependent as other monotonic inferences, for which well
known counterexamples exist. We will then look at the interpretation of might
counterfactuals. If might counterfactuals are the duals of would counterfactuals, they
are not downward entailing. However, a might counterfactual of the form of (39a)

13
This is independent, of course, of whether the set of accessible worlds A contains both / and w worlds.
If there are no /-worlds in the set of accessible worlds A, the strict conditional analysis makes a
counterfactual of the type in (38a) trivially true.

123
222 L. Alonso-Ovalle

can be naturally understood as equivalent to the conjunction of counterfactuals of

the type in (39b) and (39c).

(39) a. If / or w, then might n.

b. If /, then might n.
c. If w, then might n.

We will conclude by showing that the inference from (39a) to both (39b) and (39c),
which is not licensed under a monotone semantics, is expected under the analysis
presented in Sect. 2.

3.1 Strawson downward entailingness

One important advantage of the strict conditional analysis over the minimal change
semantics analysis is that, under the strict conditional analysis, the fact that negative
polarity items are licensed in the antecedent of counterfactuals, as illustrated in (40),
does not come out as a surprise: under the strict conditional analysis, the antecedent
of counterfactuals is a downward entailing environment, and, according to the
received view, negative polarity items are licensed in downward entailing envi-
ronments (Ladusaw 1980).

(40) a. If you had left any earlier, you would have missed the plane.
(von Fintel 1999, p. 33)
b. If you had ever heard my album, you would know that I could never
consider the music business.
(www.brainyquote.com/quotes/quotes/d/dwaynehick217577.html)

But before solving the problem presented in Sect. 1 by endorsing a strict conditional
analysis of counterfactuals, we have to address the classic counterexamples to the
monotonicity of counterfactuals that Lewis (1973) brought into play to justify his
minimal change semantics. The inference from (41a) to (41b), for example, (an
instance of the pattern known as ‘Strengthening the Antecedent’) is not valid,
although it should be, under the strict conditional analysis—if all the accessible
worlds where kangaroos have no tails are worlds where they topple over, all worlds
where kangaroos have no tails but use crutches (a subset of the worlds where
kangaroos have no tails) must be worlds where they topple over.

(41) a. If kangaroos had no tails, they would topple over.

(Lewis 1973, p. 1)
b. If kangaroos had no tails but used crutches, they would topple over.
(Lewis 1973, p. 9)

Or take the argument that has (42b) and (42a) as premises and (42c) as a conclusion
(an instance of the pattern known as ‘Hypothetical Syllogism’): intuitively, it is not
valid; and, yet, under the strict conditional analysis, it should be, given the transi-
tivity of subsethood.

123
Counterfactuals, correlatives, and disjunction 223

(42) a. If Hoover had been born in Russia, he would have been a Communist.
b. If Hoover had been a Communist, he would have been a traitor.
c. If Hoover had been born in Russia, he would have been a traitor.
(Lewis (1973, p. 33) attributed to Stalnaker (1968))

Likewise for the pattern illustrated in (43) (known as ‘Contraposition’): suppose the
set of accessible worlds where Goethe does not die in 1832 is a subset of the worlds
where he is dead now; it must then follow that the set of accessible worlds where he
is alive by now is a subset of the set of worlds where he dies in 1832. Yet the
inference from (43a) to (43b) does not seem to be valid.

(43) a. If Goethe had survived the year 1832, he would nevertheless be dead
by now. (Kratzer 1979, p. 128)
b. If Goethe were alive now, he would have died in 1832.

Lewis’ counterexamples to the monotonicity of would counterfactuals pose a

problem to the strategy of deriving the interpretation of would counterfactuals with
disjunctive antecedents as a downward monotone inference. If the intuition reported
in Sect. 1 is to be captured by adopting a downward entailing semantics for would
counterfactuals, Lewis’ counterexamples must be accounted for.
Kai von Fintel (2001) presents an analysis of counterfactuals that addresses
Lewis’ counterexamples. He has convincingly argued that counterfactuals are close
to downward entailing. They are not downward entailing in the strict sense (thus
accounting for Lewis’ counterexamples) but they show limited downward entail-
ingness—what he dubbed ‘Strawson downward entailingness’—and it is this
property, he argues, that licenses negative polarity items. The question, then, is
whether the assumption that counterfactuals are Strawson downward monotonic
solves the problem illustrated in Sect. 1. We will conclude that it does not.14 To see
why, we need to bring into the discussion the analysis presented in von Fintel
(2001).
Under von Fintel’s analysis, counterfactuals are evaluated with respect to a
contextually fixed accessibility function f , which changes as discourse evolves. The
accessibility function, which he calls the ‘modal horizon’, assigns to any world of
evaluation w a set of worlds that come closest to w (with respect to an admissible
ordering of relative similarity). Counterfactuals carry the presupposition that the
modal horizon assigns to the world of evaluation worlds where the antecedent is true.
Accommodating that presupposition is what makes the modal horizon evolve. In the
initial context the modal horizon assigns to any world w the singleton that contains w.
If, by the time a counterfactual is asserted, the modal horizon f does not assign to the
world of evaluation w any worlds where the antecedent is true—as typically happens

14
Kai von Fintel himself discussed the problem in some unpublished notes (von Fintel 1997) and
reached the same conclusion.

123
224 L. Alonso-Ovalle

with respect to an initial context—those worlds that are at least as close to w as the
closest antecedent worlds are added to the worlds that f assigns to w.15

(44) Where f is an accessibility function and  a relation of relative similarity,

f jIf /; then would wj ¼ kw:fðwÞ [ fw0 j 8w00 2 ½½/f; : w0 w w00 g
(von Fintel 2001)

The proposition expressed by the conditional is then computed with respect to the
updated modal horizon. With respect to a modal horizon f (and ordering ) that
assigns to any world w some worlds where its antecedent is true, a would coun-
terfactual expresses the proposition that is true in w if and only if all worlds in f ðwÞ
where the antecedent is true are worlds where the consequent is true.

(45) ½½If /; then would w f; ðwÞ ,

2 3
½½/ f; ðw0 Þ
8w0 2 f jIf /; then would wj ðwÞ4 ! 5
f jIf /;then wj ; 0
½½w ðw Þ
(von Fintel 2001)

Most of the monotonic inferences are invalid in this system. Strengthening the
Antecedent is. Consider as illustration the inference from (46a) to (46b):

(46) a. If kangaroos had no tails, they would topple over. (Lewis 1973, p. 9)
b. If kangaroos had no tails but used crutches, they would topple over.
(Lewis 1973, p. 9)

Is the inference from (46a) to (46b) valid? In classic logic, shifting the context is
considered a fallacy: when assessing the validity of arguments the context should
not shift. We are now assuming that conterfactuals can shift the context. We then
need to take this fact into account when assessing the validity of arguments. To
assess the validity of the inference from (46a) to (46b) we will consider the fol-
lowing dynamic notion of entailment16 :

(47) Dynamic entailment

/1 ; . . . /n dynamic w iff for all contexts c, ½½/1 c \ . . . ½½/n cj/1 j...j/n1 j
 ½½wcj/1 j...j/n j (von Fintel 2001, p. 24)

15
Notation: ‘f jIf /; then would wj ’ is the modal horizon that results from accommodating the pre-
supposition that f assigns to any world w the closest worlds to w (with respect to ) where / is true.
Kai von Fintel notes in his paper that the context change potential that I am reporting here has no
provision for embedded conditionals. He offers a more complex one that does (von Fintel 2001, p. 21).
The context change potential that we are using here will do for our purposes of illustrating that this
analysis of counterfactuals does not solve the problem of disjunctive counterfactuals.
16
‘cj/j’ names the result of updating c with /.

123
Counterfactuals, correlatives, and disjunction 225

Strengthening the Antecedent is not dynamically valid. Take an arbitrary context c

and an arbitrary world w. Assume the counterfactual in (46a) is true with respect to c
in w. For that to be the case, the modal horizon in that context must assign to w
worlds where kangaroos have no tails. The counterfactual in (46b) can be undefined
in w if f ðwÞ contains no worlds where kangaroos have no tails but use crutches.
Let us now consider the case of counterfactuals with disjunctive antecedents:

(48) If we had had good weather this summer or the sun had grown cold, we
would have had a bumper crop. (Nute 1975)

The inference from (48) to (49a) and (49b) below is not dynamically valid either:
the counterfactual in (48) could be true in a world w with respect to a modal horizon
f and yet (49b) could be undefined for w. That would be the case if, for instance,
none of the closest worlds to w where the antecedent of (48) is true are worlds where
the sun grows cold.

(49) a. If we had had good weather this summer, we would have had a bumper
crop.
b. If the sun had grown cold, we would have had a bumper crop.

For the inference to go through, there must be worlds in the modal horizon of the
type described by each disjunct. To enforce this condition, we need, again, to
assume that the interpretive system has access to each of the propositions that or
operates over. The problem persists.
There is, however, a weaker notion of validity that can be formulated in the
system. We can check whether the propositions expressed by the premises of an
argument with respect to a context that satisfies the presuppositions of both the
premises and the conclusion entail the conclusion. This is the notion of entailment
that is claimed to be the one that NPIs are sensitive to (von Fintel 1999).

(50) Strawson entailment

/1 ; . . . /n Strawson w iff for all contexts c such that c ¼ cj/1 j . . . j/n jjwj,
½½/1 c \ . . . ½½/n cj/1 j...j/n1 j  ½½wcj/1 j...j/n j (von Fintel 2001, p. 26)

Strengthening the Antecedent is Strawson-valid. Take any arbitrary context c

whose modal horizon f is already such that (46a) and (46b) will not expand it
anymore. For any world w, f ðwÞ will include worlds where kangaroos have no tails
but use crutches. Assume that (46a) is true in a world w with respect to f . All the
worlds in f ðwÞ where kangaroos have no tails are worlds where they topple over.
Since f ðwÞ includes worlds where kangaroos do not have tails, but use crutches, the
counterfactual in (46b) must be true in w with respect to f .
Consider now the inference from (48) to (49b). Take a context c whose modal
horizon f is such that (48) and (49b) will not expand it. Such a modal horizon will
already include worlds where the sun grows cold. Assume that the proposition
expressed by (48) with respect to f is true in a world w. All worlds in f ðwÞ where the

123
226 L. Alonso-Ovalle

antecedent of (48) is true will be worlds where we have a bumper crop. That means
that all worlds in f ðwÞ where we have good weather this summer are worlds where
we have a good crop and all worlds in f ðwÞ where the sun grows cold are worlds
where we have a bumper crop. The counterfactual in (49b) must be true in w with
respect to f . We can reason likewise to show that the inference from (48) to (49b) is
Strawson-valid.
The analysis of counterfactuals presented in von Fintel (2001) treats the inference
from (48) to both (49a) and (49b) on a par with Strengthening the Antecedent: neither
is dynamically valid, but both are Strawson-valid. Yet there seems to be a difference
between these two inference patterns: the pattern in the interpretation of disjunctive
counterfactuals we are trying to capture seems to be reliable and stable, it does not
depend on any contextual shift; Strengthening the Antecedent is not.17 But even if the
inference illustrated in Sect. 1 were shown to be as shifty as Strengthening the
Antecedent, there are reasons to believe that the inference does not have to do with
the monotonicity of counterfactuals, since the problem illustrated in Sect. 1 has a
parallel in the case of might counterfactuals, which, if they are the duals of would
counterfactuals, are not downward entailing. We turn to this issue next.

3.2 Might counterfactuals

3.2.1 As duals of would counterfactuals

Consider the following scenario. Suppose that we are watching a magic show. The
magician mysteriously bends a fork with the power of his mind. We are shocked. To
remedy the shock, I utter the counterfactual in (51) after the show:

(51) If you had had a good magic book or you had been a newborn baby, you
might have bent that fork too.

What would be your reaction? I think you would disagree with me: perhaps you
think one could learn to bend forks from magic books (you certainly can), but if you
were a newborn baby, you know you would not have bent that fork.
The problem, again, is that a minimal change semantics for might counterfactuals
predicts the counterfactual in (51) to be true if the standard boolean semantics for or
is assumed.
To see why, we will assume, with Lewis (1973), that might counterfactuals are
the duals of would counterfactuals: a might counterfactual is true in a world w (with
respect to an admissible ordering) if and only if the proposition expressed by the
consequent is compatible with the worlds in the proposition expressed by the
antecedent that come the closest to w in the relevant ordering (but see Sect. 3.2.2
below).

(52) ½½If /; then might w ¼ kw:9w0 ½ fw ð½½/Þðw0 Þ & ½½wðw0 Þ

17
For the context dependence of Strengthening the Antecendent, see the discussion in Kratzer (1979).

123
Counterfactuals, correlatives, and disjunction 227

According to the standard boolean analysis of or, the antecedent of the counter-
factual in (51) denotes the union of the set of possible worlds where you have a good
magic book and the set of worlds where you are a newborn baby.

(53) ½½If you had had a magic book or you had been a newborn baby
¼ kw:9x½ðbookw ðxÞ & havew ðyou; xÞÞ _ childw ðyouÞ

The counterfactual in (51) is then predicted to be true in the actual world (with
respect to an admissible ordering) if and only if the consequent is true in at least one
of the worlds where (53) is true that come closest to the actual world in the relevant
ordering.
These truth-conditions are, again, too weak. Take the picture in Fig. 2 as illus-
tration. The picture in Fig. 2 represents a relative similarity relation according to
which the worlds where you are a newborn baby are more remote than the worlds
where you have a good book on magic. The counterfactual in (51) is likely to be
evaluated with respect to this type of relative similarity relation (more actual facts
have to be false in a world where you are a newborn baby than in the worlds where
you have a good book on magic), but that means that the closest worlds where the
antecedent of (51) is true are all worlds where you have a magic book, and, since,
presumably, there are worlds among the closest worlds where you have a good
magic book in which you do in fact bend the fork, the counterfactual is predicted
with respect to this similarity relation, contrary to our intuitions.
The intuition that we are reporting shows that a might counterfactual of the the
type in (54a) can be naturally understood as conveying that both counterfactuals of
the type in (54b) and (54c) are true.

(54) a. If / or w, then might n.

b. If /, then might n.
c. If w, then might n.

Fig. 2 The problem of disjunctive antecedents: might counterfactuals

123
228 L. Alonso-Ovalle

The difference with respect to the case of would counterfactuals is that the strict
conditional analysis fails to predict the validity of the inference from (54a) to (54b)
and (54c). For suppose we adopt a strict conditional analysis of might counterfac-
tuals. If they are the duals of would counterfactuals, they would claim that the
proposition expressed by the consequent is compatible with the set of accessible
worlds in the proposition expressed by the antecedent:

(55) ½½If /; might ðCÞ ng ¼ kw:9w0 ½½½/ðw0 Þ & gðCÞðwÞðw0 Þ & ½½nðw0 Þ

Consider now a disjunctive counterfactual, like the one in (51), repeated in (56)
below.

(56) If you had had a good magic book or you had been a newborn baby, you
might have bent that fork too.

Under the strict conditional analysis, the counterfactual in (56) conveys that the
proposition that you bend that fork is compatible with the accessible worlds that are
either worlds where you have a good magic book or worlds where you are a
newborn baby, as captured below.

(57) ½½ð56Þg ¼
20 1 3
9x½bookw0 ðxÞ & havew0 ðyou; xÞ & gðCÞðwÞðw0 Þ
kw:9w0 4 @ _ A & 5
babyw0 ðyouÞ bendw0 ðyou; fÞ

But then, according to the truth-conditions in (63), for the counterfactual in (56) to be
true, the proposition expressed by the consequent need not be compatible with each
disjunct: the truth conditions in (63) will be satisfied in a world w in case none of the
accessible worlds where you are a newborn baby are worlds where you bend that fork.
The inference from (54a) to (54b) and (54c) is not Strawson-valid either, of
course. Take an arbitrary context c and an arbitrary world w. Assume that the
counterfactual in (56) is true with respect to c in w. For that to be the case, the
proposition expressed by its antecedent must be compatible with the accessible
worlds in the modal horizon f ðwÞ. We now assume that the propositions expressed
by the antecedents of both (58a) and (58b) are compatible with f ðwÞ. That means
that the modal horizon f ðwÞ contains worlds where you have a good magic book,
and worlds where you are a newborn baby. Now suppose that the proposition that
you bent that fork is compatible with the closest worlds where you have a good
magic book, but that it is incompatible with the closest worlds where you are a
newborn baby. The counterfactuals in (56) and (58a) will be both true in w with
respect to context c, but the counterfactual in (58b) will be false.

(58) a. If you had had a good magic book, you might have bent that fork.
b. If you had been a newborn baby, you might have bent that fork.

Assuming that might counterfactuals are the duals of would counterfactuals

doesn’t help capturing their natural interpretation.

123
Counterfactuals, correlatives, and disjunction 229

One could argue, however, that the assumption that might counterfactuals are the
duals of would counterfactuals should not be taken for granted. There is a debate in
the literature on minimal change semantics for conditionals between Lewis (1973)
and Stalnaker (1984) that focuses on the duality of would and might counterfactuals.
Lewis sticks to the assumption that would and might counterfactuals are duals of each
other. Stalnaker doesn’t. Stalnaker’s semantics for would conditionals makes use of a
selection function that picks up for any world of evaluation w, the closest world to w
in which the antecedent is true. A would conditional says that the closest world where
the antecedent is true is one where the consequent is. Since would counterfactuals are
not universal quantifiers, they do not have duals. In Stalnaker’s system might
counterfactuals are epistemically qualified versions of would counterfactuals (Stal-
naker 1984, p. 144).18 To conclude, I want to show that adopting Stalnaker’s analysis
does not solve the problem of capturing the natural interpretation of disjunctive
antecedent, because the problem arises with conditionals containing other possibility
modals for which a Stalnaker-type analysis is not plausible.

3.2.2 Stalnaker on might counterfactuals

Under Stalnaker’s analysis, the example in (56), repeated in (59), receives the LF in
(60):

(59) If you had had a good magic book or you had been a newborn baby, you
might have bent that fork too.

ð60Þ

18
The main motivation for Stalnaker’s analysis is that it is hard to follow the denial of a would
counterfactual with a might counterfactual, as the following example illustrates:
(i) a. Would President Carter have appointed a woman to the Supreme Court last year if a
vacancy had occurred?
b. # No, certainly not, although he might have appointed a woman.
(Stalnaker 1984, p. 144)
This is unexpected under Lewis’ analysis: it could very well be that not all closest worlds of the
antecedent type are worlds of the consequent type, while some of them are.

123
230 L. Alonso-Ovalle

If might counterfactuals are analyzed as would counterfactuals embedded under

epistemic might, and would counterfactuals are analyzed as strict conditionals, the
sentence in (59) will entail both (61a) and (61b). Let us see why.

(61) a. If you had had a good book on magic, you might have bent that fork.
b. If you had been a newborn baby, you might have bent that fork.

Might denotes the function from propositions to propositions that maps any
proposition p into the proposition that is true in any world w if and only if p is
consistent with the set of worlds epistemically accessible from w.

(62) Where Ew is the set worlds epistemically accessible from w,

½½might ¼ kp:kw:9w0 ½w0 2 Ew & pðw0 Þ

The LF in (60) denotes the proposition that is true in any world w if and only if there
is at least one world epistemically accessible from w in which the embedded would
counterfactual is true. Under the strict conditional analysis, the counterfactual
embedded under might in (60) denotes the proposition p that is true in a world w if
and only if all accessible worlds in the union of the set of worlds where you have a
good magic book and the set of worlds where you are a newborn baby are worlds
where you bend that fork, as illustrated below.
20 13
9x½bookw0 ðxÞ & havew0 ðyou; xÞ
6@ _ babyw0 ðyouÞ A7
(63) ½½g ¼ kw:8w0 6 4 0
7
5
& gðCÞðwÞðw Þ
! bendw ðyou; fÞ
0

Suppose now that the proposition denoted by the LF in (60) is true in a world w, but
that the proposition expressed by (61a) is false (under Stalnaker’s analysis, the
proposition expressed by (61a) is the proposition p that is true in a world w if and
only if there is at least one epistemically accessible world where the proposition
expressed by (64) is true.)

(64) If you had had a good magic book, you would have bent that fork.

There must then be at least one world w0 epistemically accessible from w such that the
set of worlds accessible from w0 where you have a good book on magic is a subset of
the worlds where you bend that fork. Call that world w1 . If the sentence in (61a) is
false in w, there should not be any world w0 epistemically accessible from w such that
the set of accessible worlds from w0 where you have a good magic book is a subset of
the worlds where you bend that fork. That contradicts the conclusion that w1 is one
such world. Reasoning likewise, we conclude that (60) also entails (61b). Stalnaker’s
analysis of might counterfactuals could then capture the desired entailments.
The problematic entailments do not only arise with might counterfactuals,
though: they also arise with other types of conditionals with possibility modals.
Take, for instance deontic may:

123
Counterfactuals, correlatives, and disjunction 231

(65) Mom, to Dad: ‘‘If Sandy does her homework or yours, she may eat this ice
cream.’’

It seems natural to conclude from (65) that both sentences below are true:

(66) a. If Sandy does her homework, she may eat this ice cream.
b. If Sandy does your homework, she may eat this ice cream.

In fact, the discourse in (67) sounds contradictory:

(67) # If Sandy does her homework or her sister’s, she may eat this ice cream,
but if she does her homework, she may not eat it.

If we were to adopt Stalnaker’s strategy to account for the interpretation of dis-

junctive counterfactuals, we would also have to adopt it to account for the inter-
pretation of disjunctive deontic conditionals. It is far from clear how Stalnaker’s
strategy might be applied to disjunctive deontic conditionals, though: embedding
the corresponding universal must conditional would not help, since the conditional
in (66b) is definitely not equivalent to the one in (68b).19

(68) a. If Sandy does her homework, she may eat this ice cream.
b. It might be the case that if Sandy does her homework, she must
eat this ice cream.

3.3 Conclusion

Appealing to a strict conditional analysis of counterfactuals while keeping the

standard boolean semantics for or does not help much, then, because a strict con-
ditional analysis does not capture by itself the natural interpretation of might
counterfactuals.
The analysis that we presented in Sect. 2, however, naturally extends to the case
of disjunctive might counterfactuals: it treats them in completely parallel fashion
to would counterfactuals. Take, as illustration, the might counterfactual in (59),
repeated in (69) below.

(69) If you had had a good magic book or you had been a newborn baby,
you might have bent that fork too.

19
An anonymous reviewer points out that we could get the desired entailments if we embed the con-
ditional under a restricted universal modal:

(i)
w8w0 ½ðw0 2 ½A _ B & gðCÞðwÞðw0 ÞÞ ! 9w00 ½w00 is deontically accessible from w0 & w00 2 D

To get the desired entailments, we would need to make sure that C selects both A and B worlds, of course.
This move, as the reviewer points out, would mean giving up on a unified analysis of epistemic and
deontic modals. The analysis I presented in the previous sections is consistent with the type of unified
analysis of conditionals pursued over the years by Angelika Kratzer (Kratzer 1991, a. o.), see discussion
of (72) below.

123
232 L. Alonso-Ovalle

In the consequent of the conditional in (69), might combines with then and with the
proposition that you have bent that fork to yield a (singleton containing the)
proposition that is true in a world w if and only if the set of worlds in the denotation
of the antecedent of then that come closest to w is compatible with the proposition
that you have bent that fork.
ð70Þ

Abstracting over the denotation of then, we get a singleton containing a function

from propositions to propositions that maps any proposition p into the proposition
that is true in a world w if and only if the set of p-worlds that come closest to w is
compatible with the set of worlds where you bend that fork.
ð71Þ

Under this analysis, the if-clause denotes a set containing a function of type
hhhs; ti; hs; tii; hs; tii that maps the function in the set in (71) to the proposition that
is true in any world w if and only if the set of worlds where you have a good magic
book that come closest to w is compatible with the proposition that you bend that
fork, and the set of worlds where you are a newborn child that come the closest to w
is also compatible with the proposition that you bend that fork. The analysis predicts
the sentence in (20) to be false in the context presented in Sect. 2, because, under the
relevant ordering, none of the closest worlds to the actual world where you are a
newborn child are worlds where you bend the fork.
The reader can probably see how the analysis can extend to other types of
disjunctive conditionals. Take, for instance, the deontic conditional in (65), repeated
in (72) below:

(72) Mom, to Dad: ‘‘If Sandy does her homework or yours, she may eat this ice
cream.’’

123
Counterfactuals, correlatives, and disjunction 233

The analysis presented in Sect. 2 would claim that the consequent of this conditional
expresses the property of propositions P that is true of any proposition p in a world
w if and only if in some of the closest worlds to a deontic ideal in w in which p is
true are worlds in which she eats ice cream. The whole conditional would claim that
in the world of evaluation w, the property P is true of the proposition that Sandy
does her homework and of the proposition that she does her sister’s homework.

4 An implicature?

To conclude, I would like to address a different potential derivation of the natural

interpretation of disjunctive counterfactuals. The idea is to derive the interpretation
of disjunctive counterfactuals, while keeping the standard boolean analysis of or, by
resorting to a conversational implicature. The following quote illustrates the spirit of
the proposal20 :

‘‘There is certainly evidence for SDA. From the statement

D: If there had been rain or frost, the game would have been called off.
one naturally infers both of these:
Dr : If it had rained, the game would have been called off.
Df : If there had been frost, the game would have been called off.
What validates those inferences if SDA is not valid? [. . .]
The explanation is Gricean. D would be a sensible, decent, verbally eco-
nomical thing to say only [emphasis added] for someone who did think that Dr
and Df are both true. Consider a person who asserts D because he is confident
of Dr , he regards the closest Freeze-worlds as remote, and does not believe Df .
What this person asserts is true if he is right about Dr ; but asserting it on this
basis is bad behaviour. It is of the same general kind—though perhaps not so
bad in degree—as your saying ‘If there had been rain or 90 per cent of the
world’s Buddhist priests had converted to Catholicism overnight, the game
would have been called off.’ The second disjunct is pointless in this case.
There would be a point in including it only if it too had some bearing on the
Consequent.’’
(Bennett 2003, pp. 168–170)

How should this type of argument be spelled out? Consider again, as illustration,
the disjunctive might counterfactual that we discussed in Sect. 3.2, together with the
simpler counterfactuals in (74a) and (74b).
(73) If you had had a good magic book or you had been a newborn child, you
might have bent that fork.
20
In the quote below, ‘SDA’ stands for the ‘simplification of disjunctive antecedents’ inference pattern,
the name used in the philosophical literature to refer to the interpretation of disjunctive counterfactuals
illustrated in Sect. 2.

123
234 L. Alonso-Ovalle

(74) a. If you had had a good magic book, you might have bent that fork.
b. If you had been a newborn child, you might have bent that fork.

The starting point for this line of reasoning is the observation that it does not seem
cooperative for a speaker to utter the disjunctive counterfactual in (73) if, for any world
w compatible with what the speaker believes, the closest worlds to w where you have a
good magic book are closer to w than the closest worlds where you are a newborn child.
Why? If for any world w compatible with what the speaker believes, the closest worlds
to w where you have a magic book are closer to w than the closest worlds where you are
a newborn baby, the disjunctive counterfactual in (73) will be true in a world w
compatible with what the speaker believes if and only if the simpler counterfactual in
(74b) is. But the disjunctive counterfactual in (73) is a more complex (longer)
expression than the counterfactual in (74a). Some kind of economy principle should
rule out uttering (74b) instead of the simpler (74b) in this situation. What kind of
economy principle can we appeal to? Grice’s maxim of manner seems to be a possi-
bility. Let’s see what it takes to appeal to the maxim of manner.
The sentences in (73) and (74a) are not logically equivalent, so the manner
reasoning cannot rely on comparing two expressions that share the same meaning,
but maybe it is enough to assume that what is being compared here are two sen-
tences—one of which is more complex than the other—that are truth-conditionally
equivalent throughout the speaker’s belief state. Let us then assume tentatively the
following principle:

(75) Suppose the sentence S contains a proper subset of the lexical items (sub-
trees . . .) of the sentence S0 . If S and S0 have identical semantic interpretation
in the speaker’s belief state, then S0 cannot be uttered felicitously.

By assuming that the speaker is obeying the principle in (75), the hearer could
reason as follows:

(76) The speaker is obeying the principle in (75). Thus, if the speaker uses a complex
form S0 , then no simpler sentence S has identical semantic interpretation in the
speaker’s belief state.

The principle in (76) does not justify concluding from an utterance of (73) that
both (74a) and (74b) are true. What the hearer can conclude from (76) is that there
must be at least one world compatible with what the speaker believes where (73) is
true, but (74a) isn’t; and that there must be a world compatible with what the
speaker believes where (73) is true, but (74b) isn’t. That cannot be the case if in all
worlds w compatible with what the speaker believes, the closest worlds to w where
the antecedent of (73) is true are all worlds where you have a good magic book
(because in that case the sentence in (74a) would have to be true in all worlds w
compatible with what the speaker believes), but we cannot yet conclude that for any
world w compatible with what the speaker believes, both (74a) and (74b) must be
true in w: for let us suppose that the worlds where you have a good magic book and
the worlds where you are a newborn child are equally close to any world w

123
Counterfactuals, correlatives, and disjunction 235

compatible with what the speaker believes; and let us also suppose that there are
only two types of worlds compatible with what the speaker believes: worlds w in
which the closest worlds where you have a magic book are compatible with your
bending that fork, but in which you don’t bend that fork in any of the closest worlds
where you are a newborn child (as illustrated in Fig. 3); and worlds w where the
closest worlds where you are a newborn child are compatible with your bending that
fork, but in which none of the closest worlds where you have a good magic book are
(as illustrated in Fig. 4). In all worlds w compatible with what the speaker believes,
the sentence in (73) is true, but neither (74a) or (74b) is true in all those worlds. The
discourse below is not predicted to be deviant.

(77) # If you had a good magic book or you were a newborn child, you might
have bent that fork; but, for all I know, it is possible that if you have a good
magic book you might not have bent that fork.
Appealing to the principle in (76) alone does not seem to be enough.
What other principle could be ruling out uttering the disjunctive conditional in
the scenarios where it quantifies over worlds that are only in one of the two
disjuncts? The reasoning that Bennett entertains is slightly different from the one
illustrated above. What Bennett seems to be assuming is that the hearer can

Fig. 3 If you had had a good magic book, you might have bent that fork

Fig. 4 If you had been a newborn child, you might have bent that fork

123
236 L. Alonso-Ovalle

conclude from an utterance of (73) that the speaker is not in a belief state in all
whose worlds (74a) is true, but (74b) is false. Why is that so? The idea seems to be
that when (73) is true in any world w compatible with what the speaker believes if
and only if (74a) is, the second disjunct seems to have no role whatsoever. But even
if the selection function were to pick up some worlds where the second disjunct is
true, what makes sure that both (74a) and (74b) are true? The counterfactual in (73)
can be true in a world w, whose closest counterparts include worlds where you have
a magic book and worlds where you are a newborn child, in case none of the worlds
where you are a newborn baby are worlds where you bend that fork.
If the natural interpretation of disjunctive counterfactuals is indeed an impli-
cature, a different pragmatic principle should be at work.21

5 On the universal force and the visibility of the disjuncts

According to the analysis presented in Sect. 2, the interpretation of disjunctive

counterfactuals involves universal quantification over antecedents: a counterfactual
of the form ‘‘If A or B, then C’’ conveys that for all propositions p in the set
containing the propositions expressed by A and B, it is true that if p, then C.
The analysis has two components: (i) adopting a Hamblin semantics for or allows
for the semantic composition of counterfactuals to make reference to each disjunct
on its own, and (ii) assuming that they are correlatives justifies their universal force.
Before concluding, I would like to address two issues concerning these two
components: in the next subsection I discuss the assumption that source of the
universal force is external to the consequent, and, in the last subsection, I discuss the
assumption that the disjuncts are always visible to the interpretation component.

21
In van Rooij (2007) one finds the sketch of an elegant pragmatic account of the natural interpretation
of disjunctive counterfactuals. The account defines an ordering of the possible belief states of the speaker
in which (i) below is true. The belief states in which (i) is true are ranked with respect to how many of the
alternatives in (ii) are true in them.
(i) If A or B, would C.
(ii) a. If A, would C.
b. If B, would C.
Assuming that the speaker is competent about both (iia) and (iib) (that for either of those alternatives, it
holds that they are true either in all worlds compatible with what the speaker believes or in none of them),
there are two minimal states in the ordering: one in which (iia) is true, but (iib) is false; and another in
which (iib) is true and (iia) is false. An optimal interpretation for (i) is defined on the basis of these
minimal belief states. The interpretation maps a sentence like (i) into the set of belief states s in which all
the alternatives that are true in a minimal belief state are true. An optimal interpretation of (i) is then one
in which the speaker believes (ii) and also (iii).
It would be interesting to see what the system predicts for the interpretation of disjunctive counter-
factuals embedded under negation (see discussion of example (29)). Are the alternatives to a sentence like
(iii) the ones in (iv)? In view of the discussion, should we define the optimal belief state for (iii) as one in
which both alternatives in (iv) are true?
(iii) It’s false that if A or B, would C.
(iv) a. It’s false that if A, would C.
b. It’s false that if B, would C.

123
Counterfactuals, correlatives, and disjunction 237

5.1 Is the universal force external to the consequent?

In the analyses presented in Alonso-Ovalle (2004) and van Rooij (2006), the disjuncts
are made visible to the interpretation function via the variable assignment, by
assuming that disjunctions impose a condition on a variable, as first proposed in Rooth
and Partee (1982), following the Heimian analysis of indefinites (Heim 1982). The
disjunctive counterfactual of the sentence in (1), repeated in (78a) below, would be
analyzed, according to this analysis, as in (78b), where p is a free variable of type hs; ti.

(78) a. If we had had good weather this summer or the sun had grown cold,
we would have had a bumper crop.
(A variation on an example in Nute 1975.)
b. p & p ¼ kw:good-summerw _ p ¼ kw:coldw ðsÞ

The effect of quantifying over the disjuncts is derived from the universal force of
would counterfactuals. These analyses account for the natural interpretation of
would counterfactuals by assuming that they quantify over pairs hw; gi of worlds w
and variable assignments g such that g is a variable assignment that maps the
variable introduced by or to one of its possible values (determined, as in (78b), by
the disjuncts), and w is a world (i) in which the proposition expressed by the
antecedent under g is true, and (ii) that gets as close to the world of evaluation as
any other world w0 in which the proposition expressed by the antecedent under g is
true. Would counterfactuals claim that all those pairs are pairs that satisfy the
consequent. The universal force is built into the semantics of would counterfactuals.
This type of analysis, however, does not cover the case of might counterfactuals.
If might counterfactuals are duals of would counterfactuals, they would say that at
least one of the pairs hw; gi contributed by the antecedent satisfy the consequent, not
that all do. To cover the case of might counterfactuals, we need to resort to an
external force of universal quantification, as the analysis presented in Sect. 2 does.
Take for instance the might counterpart of the example in (78a).

(79) If we had had good weather this summer or the sun had grown cold, we might
have had a bumper crop.

According to the analysis presented in Sect. 2, this counterfactual claims that we

might have had a good crop if we had had good weather this summer, and also if the
sun had grown cold. To capture this interpretation, we need to move beyond
quantifying over pairs of worlds and assignment and mimic the effects of quanti-
fying over the disjuncts in the antecedent. One possibility, suggested in van Rooij
2006 (footnote 26), is this: the semantic interpretation can collect in one set the pairs
hw; gi such that g maps p to the proposition that we have a good summer and w is
one of the closest worlds where we have a good summer, and in another set the pairs
hw; gi such that g maps p to the proposition that the sun grows cold. We can then say
that the antecedent of the counterfactual contributes a set containing these two sets,
and that both would and might counterfactuals claim that the consequent follows
from any of those sets. We would end up with an analysis similar to the one

123
238 L. Alonso-Ovalle

proposed in Sect. 2, but in which the source of the universal quantification is left
unexplained. The analysis presented in Sect. 2 justifies the source of the universal
quantification by assuming that counterfactuals are correlatives.

5.2 On the visibility of the disjuncts

To conclude, I would like to mention that there is a known recipe to construct

counterexamples to the interpretation predicted by the analysis presented in Sect. 2:
make up a disjunctive counterfactual of the type that we have been looking at (the
type where one of the disjuncts is more remote than the other), and be sure that the
consequent denotes the proposition expressed by one of the disjuncts. Here’s a
famous case:

(80) If the U.S. devoted more than half of its budget to defense or to education,
it would devote more than half of its budget to defense.
(Nute 1984)

According to the analysis presented in Sect. 2, this counterfactual denotes (a

singleton containing) the proposition p that is true in a world w if and only if the
closest worlds to w where the U.S. spends more than half of its budget in defense are
all worlds where the U.S. spends more than half of its budget in defense, and the
closest worlds to w where the U.S. spends more than half of its budget in education
are all worlds where it spends more than half of its budget to defense. The sentence
seems to be intuitively true, but this proposition is a contradiction.
The standard theory of or fares better here. Under the standard analysis of or, the
antecedent expresses the proposition that at least one of the disjuncts is true (the
union of the set of worlds where the U.S. devotes more than half of its budget to
defense and the set of worlds where it devotes more than half of its budget to
education). Under the plausible relative similarity relation that ranks the worlds
where the U.S. spends more in defense than in education as closer to the actual
world than the worlds where the U.S. spends more in education than in defense, the
selection function will only pick up worlds where the U.S. devotes more than half of
its budget to defense. The counterfactual is then predicted to be true.
The analysis presented in Sect. 2 can still capture the interpretation that we want
by letting an Existential Closure operation range over the set of propositional
alternatives introduced by or, as illustrated below, in which case the proposition
expressed by the antecedent would be the same as in the standard analysis of or.22
22
Similarly, an anonymous reviewer offers the following context:
(81) John is playing a game of luck with white, black, and red balls. We know (but he doesn’t)
that none of the white balls has a winning number, but that taken together 30% of the black
and red balls have a winning number (but we don’t know the proportion of winning numbers
among the black balls, nor among the read balls.) John picked a ball, and lost.
The reviewer points out that, in this context, (i) can be read as (ii), as if Existential Closure was
triggered within the scope of the if-clause.
(i) If John had picked a black ball or a red ball, he would have had 30% chances of winning.
(ii) If John had picked a ball that wasn’t white, he would have had 30% chances of winning.

123
Counterfactuals, correlatives, and disjunction 239

ð82Þ

a. ½½ ;g ¼ fkfhhs;ti;hs;tii :kw:8p½ p 2 ½½9P;g ! fðpÞðwÞg

 
;g 0
kw:spend-w 12 ðus; edÞ; 0
b. ½½9P ¼ kw :9p 2 & pðw Þ
kw:spend-w 12 ðus; dfÞ


kw:spend-w 12 ðus; edÞ;
c. ½½
;g ¼
kw:spend-w 12 ðus; dfÞ

We are left with an important question, though: why is Existential Closure triggered
in this example, but not in the ones that we discussed before? If disjunctive
counterfactuals were ambiguous and their LFs could optionally include an Exis-
tential Closure operator under the scope of if, the counterfactuals that we discussed
in the previous sections should have the reading predicted by the standard analysis
of or, and, yet, they don’t seem to.
If we want to capture the interpretation of examples like (80) by resorting to an
operation of Existential Closure, we seem to be forced to conclude that the operation
is a last resort strategy to avoid interpreting examples like (80) as contradictions.
Maybe one could reason as follows: the analysis predicts that the example in (80)
can only be true if the proposition that the U.S. devoted more than half of its budget
to education were the impossible proposition (the proposition that is true in no
world). Here’s why: the analysis predicts the sentence to be true if and only if the
closest worlds where the U.S. spends more than half of its budget in education are
all worlds where it spends more than half of its budget in defense and the closest
worlds where the U.S. spends more than half of its budget in defense are all worlds
where it spends more than half of its budget in education. As long as there are
worlds where the U.S. spends more than half of its budget in education, these truth-
conditions will not be satisfied in any world, because, given our assumptions about
the selection function, if there are worlds where the U.S. spends more than half of its
budget in education, when applied to the proposition that the U.S. spends more than
half of its budget in education, the selection function will return no world where the
U.S. spends more than half of its budget in defense. If there were no possible worlds
where the U.S. spends more than half of its budget to education, the selection
function will return the empty set, and since the empty set is a subset of any set, it
will be a subset of the set of worlds where the U.S. devotes more than half of its
budget to defense. The sentence could then be true. Now, since the proposition that
the U.S. devotes more than half of its budget to education is not the impossible

123
240 L. Alonso-Ovalle

proposition, the hearer knows that the sentence in (80) is a contradiction. For (80) to
be contingent, Existential Closure should be triggered.
However, I am not fully convinced that this is all there is to be said about the
pattern that (80) illustrates. Consider for instance the following example, with
exactly the same characteristics:

(83) If I earned at most $30,000 or more than a billion, I would surely earn at
most $30,000.

The example sounds contradictory to me, unless it is forced to be interpreted as the

following more verbose examples:

(84) a. If I were to earn at most $30,000 or more than a billion, I would earn at
most $30,000.
b. If I might earn at most $30,000 or more than a billion, I would earn at
most $30,000.
c. (Even) if it were possible that I earned at most $30,000 and it were also
possible that I earned more than a billion, I would nevertheless earn at
most $30,000.

Similarly, I think Nute’s example accepts the following paraphrases:

(85) a. If the U.S. were to devote more than half of its budget to defense or
education, it would devote more than half of its budget to defense.
b. If it were the case that the U.S. might devote more than half of its budget
to defense or to education, it would devote more than half of its budget to
defense.
c. (Even) if it were possible that the U.S. devoted more than half of its
budget to defense and it were possible that the U.S. devoted more than
half of its budget to education, the U.S. would nevertheless devote more
than half of its budget to defense.

These paraphrases reveal some implicit modality. The disjunctions in the ante-
cedent could be under the scope of a modal. There is then much more to say about
these examples. We need to know where the implicit modality comes from, and we
need to know how the propositional alternatives introduced by disjunction interact
with modals. These two questions go beyond the scope of this article.

6 To conclude

We have seen that the failure to capture the natural interpretation of disjunctive
counterfactuals provides no reason to abandon a minimal change semantics for
counterfactuals.
The analysis of disjunctive counterfactuals that we have presented, however,
does not come for free. To be sure that the interpretation has access to each disjunct
on its own, we have moved beyond the textbook analysis and embraced a Hamblin

123
Counterfactuals, correlatives, and disjunction 241

semantics for disjunction. Isn’t this a costly move? I would like to let David Lewis
answer the question.
In a short reply to his critics (Lewis 1977), Lewis suggested in passing moving
beyond the standard analysis of or to capture the natural interpretation of disjunctive
counterfactuals, and justifies the move as follows:
‘‘Isn’t it badly ad hoc to solve a problem in counterfactual logic by compli-
cating our treatment of ‘or’? When we have a simple, familiar, unified
treatment (marred only by the irrelevant question of exclusivity) wouldn’t it be
more sensible to cherish it? I reply that if I considered our present problem in
isolation, I would share these misgivings. But parallel problems arise from
other constructions, so our nice uncomplicated treatment of ‘or’ is done for in
any case. Consider:

(4) I can lick any man in the house, or drink the lot of you under the table.
(5) It is legal for you to report this as taxable income or for me to claim you
as a dependent.
(6) Holmes now knows whether the butler did it or the gardener did.
Take the standard treatment of ‘or’. Try wide or narrow scope; try inclusive or
exclusive. (4–6) will prove as bad as (1).
(1) If either Oswald had not fired or Kennedy had been in a bullet-proof car,
then Kennedy would be alive today.’’
(Lewis 1977, pp. 360–361)

Thirty odd years after it was written, Lewis’ answer sounds premonitory. A
number of recent works have resorted to an alternative semantics to solve several
long-standing problems that have to do with the interpretation of disjunction:
Alonso-Ovalle (2008), for instance, shows the advantages of adopting an Alterna-
tive Semantics to capture the exclusive interpretation of disjunctions with more than
two disjuncts, and Aloni (2003a,b), Simons (2005) and Alonso-Ovalle (2006)
pursue the hypothesis that disjunctions denote sets of propositional alternatives to
account for its behavior in modal contexts, like the ones that Lewis considers.23
Capturing the interpretation of disjunctive counterfactuals is not the only reason to
embrace an Alternative Semantics for disjunctions, then.
To conclude, I would like to point out two issues for further research. The first one
is the analysis of the behavior of indefinites and Free Choice items in the antecedent
of counterfactual conditionals. A number of recent works (see Kratzer and Shi-
moyama 2002; Alonso-Ovalle and Menéndez-Benito 2003; Menéndez-Benito 2005)
have offered reasons to analyze in an alternative semantics both existential (German
irgendein, Spanish algún) and universal (Spanish cualquiera) Free Choice items.
Alonso-Ovalle and Menéndez-Benito (in preparation) explain the scope behavior of

23
See Alonso-Ovalle (2006) for extensive discussion of the literature. In the analysis that we have
presented, and in these more recent works, it is assumed that disjunctions do not have quantificational
force of its own. This is an assumption that was first made in Rooth and Partee (1982).

123
242 L. Alonso-Ovalle

Spanish algún in conditionals by adopting the analysis of conditionals presented

above. It would be interesting to investigate whether the analysis presented here can
also say something about the interpretation of other Free Choice items in the ante-
cedent of counterfactuals.
The second issue for further research is the analysis of and. Under the set-up that
we have presented, or does not have any existential force of its own. Its only role is
to introduce a set of propositional alternatives into the semantic derivation. An
external Existential Closure operator is responsible for the existential force tradi-
tionally associated with or. Are there are reasons to believe that the universal force
of and is also external.
Kratzer (1977) points out that examples like (86) are ambiguous between giving
a single conjoined recommendation (that students engage in doing two things) and
giving a pair of recommendations (recommending students to practice striding and
also recommending them to practice flying.) It remains to be seen whether the fact
that each term of the conjunction can have an equal status in the recommendation is
connected to the phenomenon discussed in this paper.24

(86) Te Miti recommended that students practice striding and flying.

(Kratzer 1977)

Acknowledgments For insightful comments and suggestions, I would like to express my gratitude to
Kai von Fintel, Lyn Frazier, Elena Herburger, Manfred Krifka, Barbara Partee, Christopher Potts, Robert
van Rooij, and, especially, Angelika Kratzer. I would also like to thank the audiences at the 26th West
Coast Conference on Formal Linguistics, the 33rd Annual Meeting of the Berkeley Linguistics Society,
the 2005 LSA summer institute, the 79th LSA Annual Meeting, Sinn und Bedeutung 9, and the 34th
Meeting of the North East Linguistic Society, where parts of this project were presented at different
stages. Many thanks to two anonymous Linguistics and Philosophy reviewers for their detailed criticism
of the paper. All mistakes are my own.

References

Aloni, M. (2003a). Free choice in modal contexts. In M. Weisgerber (Ed.), Proceedings of the conference
‘‘SuB7—Sinn und Bedeutung’’. Arbeitspapier Nr. 114 (pp. 28–37). Konstanz.
Aloni, M. (2003b). On choice-offering imperatives. In P. Dekker & R. van Rooy (Eds.), Proceedings of
the Fourteenth Amsterdam Colloquium, Amsterdam.
Alonso-Ovalle, L. (2004). Simplification of disjunctive antecedents. In K. Moulton & M. Wolf (Eds.),
Proceedings of the North East Linguistic Society. Amherst, MA, pp. 1–15.
Alonso-Ovalle, L. (2006). Disjunction in alternative semantics. Ph.D. thesis, University of Massachusetts,
Amherst, MA.
Alonso-Ovalle, L. (2008). Innocent exclusion in an alternative semantics. Natural Language Semantics,
16, 115–128. doi:10.1007/s11050-008-9027-1.
Alonso-Ovalle, L., & Menéndez-Benito, P. (2003). Some epistemic indefinites. In M. Kadowaki &
S. Kawahara (Eds.), Proceedings of the North East Linguistic Society (Vol. 33, pp. 1–12). Amherst.
MA: GLSA.
Alonso-Ovalle, L., & Menéndez-Benito, P. (in preparation). Exceptional scope and specificity: The case
of Spanish. University of Massachusetts Boston, University of Goettingen, ms.
Beck, S. (2001). Reciprocals and definites. Natural Language Semantics, 9, 69–138.
Bennett, J. (2003). A philosophical guide to conditionals. Oxford: Oxford University Press.

24
Thanks to Angelika Kratzer for pointing out this potential connection.

123
Counterfactuals, correlatives, and disjunction 243

Bhatt, R., & Pancheva, R. (2006). Conditionals. In The Blackwell companion to syntax (Vol. I, pp. 638–
687). Oxford: Blackwell.
Creary, L., & Hill, C. (1975). Review of counterfactuals. Philosophy of Science, 43, 341–344.
Dayal, V. (1995). Quantification in correlatives. In E. Batch, et al. (Eds.), Quantification in natural
languages (pp. 179–205). Dordrecht: Kluwer.
Dayal, V. (1996). Locality in Wh-quantification. Questions and relative clauses in Hindi. Dordrecht:
Kluwer.
den Dikken, M. (2006). Either-floar and the syntax of co-or-dination. Natural Language and Linguistic
Theory, 3(24), 689–749.
Ellis, B., Jackson, F., & Pargetter, R. (1977). An objection to possible-world semantics for counterfactual
logics. Journal of Philosophical Logic, 6, 355–357.
Fine, K. (1975). Critical notice: Counterfactuals by David K. Lewis. Mind, 84(335), 451–458.
von Fintel, K. (1994). Restrictions on quantifier domains. Ph.D. thesis, University of Massachusetts,
Amherst, MA.
von Fintel, K. (1997). Notes on disjunctive antecedents in conditionals. Cambridge, MA: Ms., MIT.
von Fintel, K. (1999). NPI licensing, Strawson-entailment, and context-dependency. Journal of Seman-
tics, 16(2), 97–148.
von Fintel, K. (2001). Counterfactuals in a dynamic context. In M. Kenstowicz (Ed.), Ken Hale. A life in
language (pp. 123–153). Cambridge, MA: The MIT Press.
Gallin, D. (1975). Intensional and higher-order modal logic. Amsterdam: Noth-Holland.
Hagstrom, P. (1998). Decomposing questions. Ph.D. thesis, MIT, Cambridge, MA.
Hamblin, C. L. (1973). Questions in Montague English. Foundations of Language, 10, 41–53.
Hegarty, M. (1996). The role of categorization in the contribution of conditional then: Comments on
Iatridou. Natural Language Semantics, 4(1), 111–119.
Heim, I. (1982). The semantics of definite and indefinite noun phrases. Ph.D. thesis, University of
Massachusetts, Amherst, MA.
Heim, I., & Kratzer, A. (1998). Semantics in generative grammar. Malden, MA: Blackwell.
Herburger, E., & Mauck, S. (2007). (Disjunctive) Conditionals. Ms. Georgetown University.
Iatridou, S. (1991a). If then, then what? NELS, 22, 211–225.
Iatridou, S. (1991b). Topics in conditionals. Ph.D. thesis, MIT, Cambridge, MA.
Iatridou, S. (1994). On the contribution of conditional then. Natural Language Semantics, 2(3), 171–199.
Izvorski, R. (1996). The syntax and semantics of correlative proforms. In K. Kusumoto (Ed.),
Proceedings of NELS, 26 (pp. 189–203). Amherst, MA: GLSA.
Kratzer, A. (1977). What ‘must’ and ‘can’ must and can mean. Linguistics and Philosophy, 1(3), 337–355.
Kratzer, A. (1979). Conditional necessity and possibility. In R. Bäuerle, U. Egli, & A. von Stechow
(Eds.), Semantics from different points of view (pp. 117–147). Berlin: Springer-Verlag.
Kratzer, A. (1991). Conditionals. In A. von Stechow & D. Wunderlich (Eds.), Semantics: An interna-
tional handbook of contemporary research (pp. 651–656). Berlin: de Gruyter.
Kratzer, A., & Shimoyama, J. (2002). Indeterminate phrases: The view from Japanese. In Y. Otsu (Ed.),
The Proceedings of the third Tokyo conference on psycholinguistics (pp. 1–25). Tokyo: Hituzi
Syobo.
Ladusaw, W. (1980). Polarity sensitivity as inherent scope relations. New York: Garland.
Lewis, D. (1973). Counterfactuals. Oxford: Blackwell.
Lewis, D. (1977). Possible-world semantics for counterfactual logics: A rejoinder. Journal of Philo-
sophical Logic, 6, 359–363.
Loebner, S. (1998). Polarity in natural language: Predication, quantification and negation in particular and
characterizing sentences. Linguistics and Philosophy, 23, 213–308.
Menéndez-Benito, P. (2005). The grammar of choice. Ph.D. thesis, University of Massachusetts,
Amherst, MA.
Munn, A. (1993). Topics in the syntax and semantics of coordinate structures. Ph.D. thesis,
The University of Maryland, College Park.
Nute, D. (1975). Counterfactuals and the similarity of worlds. Journal of Philosophy, 72, 773–778.
Nute, D. (1980). Conversational scorekeeping and conditionals. Journal of Philosophical Logic, 9,
153–166.
Nute, D. (1984). Conditional logic. In D. Gabbay & F. Guenthner (Eds.), Handbook of philosophical logic
(Vol. II, pp. 387–439). Dordrecht: Reidel.
Nute, D., & Cross, C. B. (2001). Conditional logic. In D. M. Gabbay & F. Guenthner (Eds.), Handbook of
philosophical logic (2nd ed., Vol. 4, pp. 1–98). Dordrecht: Kluwer.
Ramchand, G. C. (1997). Questions, polarity, and alternative semantics. In K. Kusumoto (Ed.), Pro-
ceedings of NELS (Vol. 27, pp. 383–396). Amherst, MA.

123
244 L. Alonso-Ovalle

Rooth, M. (1985). Association with focus. Ph.D. thesis, University of Massachusetts, Amherst, MA.
Rooth, M. (1992). A theory of focus interpretation. Natural Language Semantics, 1(1), 75–116.
Rooth, M., & Partee, B. (1982). Conjunction, type ambiguity, and wide scope ‘‘or’’. In D. Flickinger,
M. Macken, & N. Wiegand (Eds.), Proceedings of the first west coast conference on formal lin-
guistics (pp. 353–362). Stanford Linguistics Association.
Schlenker, P. (2004). Conditionals as definite descriptions (a referential analysis). Research on language
and computation, 2(3), 417–462.
Schwarzschild, R. (1994). Plurals, presuppositions, and the sources of distributivity. Natural Language
Semantics, 2, 201–249.
Simons, M. (2005). Dividing things up: The semantics of ‘‘or’’ and the modal/ ‘‘or’’ interaction. Natural
Language Semantics, 13, 271–316.
Srivastav, V. (1991a). The syntax and semantics of correlatives. Natural Language and Linguistic Theory,
9, 637–686.
Srivastav, V. (1991b). Wh-dependencies in Hindi and the theory of grammar. Ph.D. thesis, Cornell
University, Ithaca, NY.
Stalnaker, R. (1984). Inquiry. Cambridge, MA: The MIT Press.
Stalnaker, R. C. (1968). A theory of conditionals. In N. Rescher (Ed.), Studies in logical theory (pp. 98–113).
Oxford: Blackwell.
Stalnaker, R. C., & Thomason, R. (1970). A semantic analysis of conditional logic. Theoria, 36(1), 23–42.
van Rooij, R. (2006). Free choice counterfactual donkeys. Journal of Semantics, 23(4), 383–402.
van Rooij, R., & Schulz, K. (2007). Extending grice. Presentation at Journées Sémantique & Modélisation
2007, CNRS & Université Paris 8 Vincennes/St-Denis.
Warmbr
od, K. (1981). Counterfactuals and substitution of equivalent antecedents. Journal of Philo-
sophical Logic, 10, 267–289.

123