Conditional Logic Cb and Its Tableau System

Yuri Ozaki and Daisuke Bekki

Ochanomizu University,
Graduate School of Humanities and Sciences
{ozaki.yuri,bekki}@is.ocha.ac.jp

Abstract. Conditional logic is a kind of modal logic for analyzing the

truth conditions and inferences of conditional sentences in natural lan-
guage. However, it has been pointed out in the literature that empirical
problems plague all of the previously proposed conditional logics. More-
over, C1 and C2 are defined by imposing certain restrictions on their
Kripke frames, and there exist no corresponding proof systems.
In order to solve these problems, we propose a new system of con-
ditional logic, which we call Cb . Cb is an extension of C + through the
addition of new rules on accessibility, and it has a corresponding tableau
system. We show that Cb has empirical advantages over C1 and C2 as a
model of inference in natural language, and compare it with other proof
systems of conditional logic.

1 Introduction
1.1 Conditional Sentences in Natural Language and Classical Logic
The following inferences are valid in classical logic:
Antecedent strengthening: A B  (A C) B
Transitivity: A B, B C  A C
Contraposition: A B  B A
If we simply assume classical logic to explain the semantics of natural lan-
guage, the formulae above give rise to the following infelicitous arguments [6].
(1) If it does not rain tomorrow we will go to the cricket. Hence, if it does not
rain tomorrow and I am killed in a car accident tonight then we will go to
the cricket.
(2) If the other candidates pull out, John will get the job. If John gets the job,
the other candidates will be disappointed. Hence, if the other candidates pull
out, they will be disappointed.
(3) If we take the car then it wont break down en route. Hence, if the car does
break down en route, we didnt take it.
The reason for such infelicity is that obvious premises can be omitted in con-
ditionals. For example, the rst sentence in (1) introduces the condition it does
not rain and as we usually do not think about the possibility that we may be
killed in a car accident, we continue to assume that we will not to be killed in
a car accident as an obvious premise. However, the second sentence of (1) has

S. Pogodalla and J.-P. Prost (Eds.): LACL 2011, LNAI 6736, pp. 190204, 2011.

c Springer-Verlag Berlin Heidelberg 2011
 Conditional Logic Cb and Its Tableau System 191

the opposite meaning in that the premise that is omitted as obvious itself leads
to the invalid conclusion. In case (1), what we actually mean to say is:
(1) If it does not rain tomorrow and I am not killed in a car accident tonight,
then we will go to the cricket tomorrow.
Similar comments can be made about the arguments in (2) and (3). In case
(2), the actual wording should be if John gets the job and the other candidates
do not pull out, they will be disappointed. In case (3), the premise we will
take the car is omitted and the sentence actually states the opposite, we didnt
take it, making case (3) incomprehensible. Thus, we see that we cannot correctly
address conditionals of natural language with semantics based on classical logic.
Let us discuss case (1) in more detail. The correct sentence for (1) is if it does
not rain tomorrow then, other things being equal, we will go to the cricket. We
can call other things being equal ceteris paribus. Conditional sentences include
some notion of ceteris paribus, so GA in the conditional sentence if A and GA ,
then B can be referred to as a ceteris paribus clause, which depends on A.
Therefore, in the example sentence, if A is it does not rain tomorrow, then
GA includes the condition that we are not invaded by Martians. If A is ying
saucers arrive from Mars, it does not.
Thus, the notion of ceteris paribus is important in conditional sentences of
natural language. Accordingly, some logic systems based on the notion of ceteris
paribus have been proposed; however, they do not suciently represent truth
conditions in conditional sentences. This paper, rst, discusses a previously de-
veloped logic system for analyzing conditional sentences, and second, presents
an extension of it as a new logic system.

2 Modal Tableau
Modal Tableau is a tableau system for modal logic. In this section, we briey
describe the version of modal tableau that we adopt from [6]. In modal tableaux,
we put a natural number with each formula to designate a possible world in which
the formula is assumed to be true.
A B, i
 
A, i B, i

As the diagram above indicates, the tableau rules for truth functors are the same
as those for classical logic except for the numbers for possible worlds. Four new
rules are added for the modal operators.
A, i A, i A, i A, i
irj
A, i A, i irj
A, j A, j
In the rules above, r in irj represents the binary accessibility relation R
between two worlds i, j in a Kripke frame W, R,
of modal logic. W is a
192 Y. Ozaki and D. Bekki

non-empty set and is a function that assigns a truth value to each formula,
such that either w (p) = 1 or w (p) = 0. i and j are natural numbers, but j
must be new and must not occur at any node above in the same branch.
The two rules on the right are deduced by the following interpretation in
Kripke semantics for  and . For any world w W :
w (A) = 1 if, for all w W such that wRw , w (A) = 1;
w (A) = 0 otherwise.
w (A) = 1 if, for some w W such that wRw , w (A) = 1;
w (A) = 0 otherwise.
The two rules on the left can be explained by following proofs respectively.
In any world, w,
w (A) = 1 i w (A) = 0
i it is not the case that, for all w such that wRw , w (A) = 1
i for some w such that wRw , w (A) = 0
i for some w such that wRw , w (A) = 1
i w (A) = 1

In any world, w,
w (A) = 1 i w (A) = 0
i it is not the case that, for some w such that wRw,w (A) = 1
i for all w such that wRw , w (A) = 0
i for all w such that wRw , w (A) = 1
i w (A) = 1

The following is an example of a modal tableau for:  (A B) (A B)

((A B) (A B)), 0

(A B), 0
(A B), 0
 
A, 0 B, 0

A, 0 B, 0

0r1 0r1
A B, 1 A B, 1

A, 1 A, 1
B, 1 B, 1

A, 1 B, 1

 Conditional Logic Cb and Its Tableau System 193

As a proof of a theorem (rather than of deduction), the initial list has only
one formula, ((A B) (A B)), 0. The natural numbers for possible
worlds start at 0. When the rule (A B), 0 is applied, the new formula A B
is introduced for both branches and a new world number 1 is assigned to it. It
causes no problem that the same world number is used in both branches as they
do not interact with each other. When we judge whether a given branch closes
or not in modal tableau, we must compare contradictory formulae which are
assigned the same world number, as for A, 1 and A, 1 in the example above.

3 Conditional Logic
In section 1.1, we explained the concept of ceteris paribus in the context of
conditional sentences in natural language. The logic equipped with this concept
is called conditional logic [3,4,8], which is a kind of modal logic held to be useful
as a semantic framework of natural language.

3.1 Syntax of Conditional Logic

Let us write A > B for a conditional with a ceteris paribus condition: if A, then
B. The syntax of conditional logic is dened by the following BNF grammar,
where p is a propositional parameter:

F ::= p | F | F F | F F | F F | F F | F | F | F > F

3.2 Semantics of Conditional Logic

The Kripke frame of conditional logic consists of a quadruple: W, {RA : A
F }, R,
, where W is a non-empty set, is a function that assigns a truth value
to each pair comprising a world, w, and a propositional parameter, p, the same
as for modal logic. In world w, p is true (or false) is written as w (p) = 1
(or w (p) = 0). R is a binary relation on W which is reexive, symmetrical and
transitive. Each RA is a binary relation on W for any formula A. Intuitively,
w1 RA w2 ly means that w2 is the same as w1 except that A is true in w2 , which
represents the ceteris paribus condition.
In the settings of conditional logic,  and are treated as those of system
K [6].
w (A) = 1 if, for all w W such that wRw , w (A) = 1; and 0 otherwise.
For any world w W
w (A) = 1 if, for some w W such that wRw , w (A) = 1; and 0
otherwise.
By means of the notions introduced above, the semantics of A > B is dened
as follows.
w (A > B) = 1 i for all w such that wRA w , w (B) = 1 and w (B) = 0
otherwise.
194 Y. Ozaki and D. Bekki

Furthermore, in conditional logic, the following conception, fA (w), [A], is

added.
fA (w) = {x W : wRA x}
[A] = {w : w (A) = 1}
Here, fA (w) is the set of worlds accessible to w under RA . Also, R and fA (w)
are interdenable, since wRA w i w fA (w). [A] is a class of worlds where A
is true, {w : w (A) = 1}.
With this conception, the semantics of A > B can be simply dened as follows:
A > B is true in w fA (w) [B]

4 Previous Study
It would seem that conditional logic has a close connection with the phenomenon
of natural language, but there is as yet no logical system that can represent it. In
this section, we briey explain some logical systems that extend the conditional
logic set out in section 3.

4.1 C+
C + [1] is logic system which is an extension of conditional logic C by adding the
following conditions on its Kripke frame.
1. fA (w) [A]
2. If w [A], then w fA (w)
C + has a tableau system that corresponds to the Kripke frame above. The
following three rules are added to the tableau rules for C:
A > B, i (A > B), i

irA j
 
irA j
A, i A, i
B, j A, j
irA i
B, j
The dierence from a modal tableau is that each formula has its own relation
of accessible worlds like irA j.
As an example for a tableau proof in C + , we prove A, A > B C + B.
A, 0
A > B, 0
B, 0
 
A, 0 A, 0
0rA 0

B, 0

 Conditional Logic Cb and Its Tableau System 195

First, the rightmost rule of C + is applied, which yields 0rA 0. Then, the leftmost
rule can be applied to A > B, 0. This is closed by contradiction of B, 0 and
B, 0.
The tableau proof that follows is of p > q C + (p r) > q in C + . This tableau
proves that the inferences that result in the infelicitous semantics described in
section 1.1 are not valid when implementing the concept of ceteris paribus.

p > q, 0
((p r) > q), 0

0rpr 1
p r, 1
q, 1

p, 1
r, 1
 
p, 0 p, 0
  0rp 0
p, 1 p, 1
1rp 1 q, 0
 
q, 1 q, 1
1rq 1

Not all branches close completely, and for branches which do not do so, namely
p, 1 and q, 1, a similar formula already exists at the upper nodes. Thus, it seems
to apply the same rule innitely, indicating that this formula is indeed invalid.
Showing that the tree does not close does not mean that the formula is in-
valid, but indicates the possibility that it is invalid. One good way to prove the
invalidity of the formula is to draw a counter-model. Counter-models can be read
o from an open branch of a tableau in a natural way.
The counter-model of the formula above is as follows: w1 Rp w1 , w0 Rpr w1 and
w0 (p) = 1 and w1 (p) = w1 (q) = w1 (r) = 1. In regard to other formulae
for A, the accessibility relation RA is dened such that fA (w) = [A] for all w.
Thus, the interpretation can be depicted as follows, from which we can check
that the accessibility relation of worlds forms an innite loop. This means that
this tableau is never closed and p > q C + (p r) > q is proved.
Rp
Rpr 
w0 w1
p p, q, r
Similarly, cases (2) and (3) in the 1.1 are invalid in C + . Thus, by extending
to C + , we have solved the problem outlined in the 1.1 that classical logic is too
weak as semantics in natural language.
196 Y. Ozaki and D. Bekki

4.2 S

In section 3.2, we said that w1 RA w2 means that w2 is the same as w1 except

that A is true in w2 . Thus, we need somehow to consider the notion of similarity
between two worlds, in order to reect the intuition behind the ceteris paribus
condition. In C and C + , however, the relation RA by no means represents such
notion of similarity.
S is an extension of C + through the addition of the following three condi-
tions on a Kripke frame1 , which express a certain similarity between the two
arguments of RA .

3. If [A] = , then fA (w) =

4. If fA (w) [B] and fB (w) [A], then fA (w) = fB (w)
5. If fA (w) [B] = , then fAB (w) fA (w)

As a result, for example, the inference p > q, q > p  (p > r) (q > r) is

valid in S, but not in C + .
Proof. Suppose that the premise is true in world w, i.e., fp (w) [q] and fq (w)
[p]. Then, by applying condition 4, fp (w) = fq (w). Hence, fp (w) [r] i fq (w)
[r], i.e., (p > r)is true in w i (q > r) is true in w. i.e., (p > r) (q > r) is true
in w.
Because S does not have a corresponding tableau system, the proof in S can
rely on the semantic notions alone.

4.3 C1 , C 2

Although S is a stronger logic system than C + , it is still weak as semantics of

natural language. This prompted Stalnaker and Lewis to propose extensions2 of
S: C1 [8] and C2 [3,4]. Conditions 1 to 5 of S above are common conceptions of
C1 and C2 [7].
C1 and C2 also make a dierence through the addition of the following con-
ditions. C1 adds the following condition.

6. If w [A] and w fA (w), then w = w

C2 , on the other hand, adds the following instead of 6.

7. If x fA (w) and y fA (w), then x = y

Both condition 6 and condition 7 concern the relationship between two worlds,
but the dierence between them is that 7 entails 6, according to 2. So we can say
that C2 is stronger than C1 . We can sort these systems by increasing strength
as follows: C + < S < C1 < C2 .
1
The system S is defined as a common part of the conditional logics proposed by
Stalnaker [8] and Lewis [3,4]. Following the convention in [6], we call it S.
2
The names C1 and C2 are taken from [6].
 Conditional Logic Cb and Its Tableau System 197

However, C1 and C2 are not without problems [2]. For example, AB  A > B
is one of the formulae that is valid in C1 but not in S.
Suppose you go to a fake fortune-teller, who says that you will come into a
large sum of money. And suppose that, purely by accident, you do. The statement
If the fortune-teller says that you will come into a large sum of money, you will
still, however, would appear to be false.
Similarly,  (A > B) (A > B) is an example that it is valid in C2 but not
in S. However, both of the following conditionals would appear to be false: If
it will either rain tomorrow or it wont, then it will rain tomorrow and If it
will either rain tomorrow or it wont, then it wont rain tomorrow.
Thus, the two conceptions yield empirically wrong predictions, which are good
illustrations that each logic system is too strong to be semantics of natural lan-
guage. Given the above, although C1 and C2 are the strongest existing con-
ditional logic, they do not seem to be suitable as a logic system for natural
language.

5 Proposal: A New Conditional Logic Cb

In section 4, we pointed out the drawbacks with existing systems of conditional

logic. However, viewing them purely from the perspective of the semantics of
natural language, C + and S are too weak and C1 and C2 are too strong.
Moreover, S, C1 and C2 have no known tableau systems or any other proof sys-
tems. In developing a logic system for natural language, especially in the context
of natural language processing, whether it enables us to compute the entailment
relation between given sentences is an important feature, and is a feature that
has not been achieved in previous studies with a few notable exceptions, such as
[5].
Against such background, we propose here, as semantics of natural language,
a logic system we call Cb that properly extends C + and has a corresponding
tableau system.
One of the common problems with C + , S, C1 and C2 is when or appears
on the accessibility relations: when such truth functions occur in the antecedents
of formulae, we cannot apply any rules, and all inferences including these shall
be invalid.
The system Cb is obtained by adding the following new conditions on the
Kripke frame. It allows the nature that is inherited from C + to be preserved.

8. fB (w ) fAB (w)
w  fA (w)
9. fA (w) fAB (w)
fB (w) fAB (w)

Cb has a tableau system which is an extension of that of C + through the

addition of the following three rules:
198 Y. Ozaki and D. Bekki

irA j irA j irA j

jrB k
irAB j irBA j
irAB k
The additional rules refer only to an accessibility relation not formulae. The left
rule is for . For every irA j and jrB k on the branch, we can derive irAB k. The
right two rules are for . For every irA j on the branch, we can derive irAB j and
irBA j for any B. In section 7 and section 8, we prove that this tableau system is
sound and complete with respect to the Kripke semantics introduced above.
Note that any theorem of C + is a theorem of Cb . In the next section, we will
verify what inferences are valid in Cb .

6 Empirical Verification
The following tableau is an example of a proof in Cb . We note, in passing, that
all the following inferences are valid in Cb but not in C + , S, C1 or C2 .3
Ex.1 (p q) > r  p > (q > r)

(p q) > r, 0
(p > (q > r)), 0

0rp 1
p, 1
(q > r), 1

1rq 2
q, 2
r, 2

0rpq 2

r, 2

This is an example of applying the left rule proposed in section 5.

0rpq 2 is derived by applying the Cb rule to the second line 0rp 1 and the third
line 1rq 2. This formula creates a sentence like the following.
If the rain stops and the water temperature is more than 25 C, then we can
swim in the pool.
If the rain stops, then additionally, if the water temperature is more than
25 C, we can swim in the pool.
3
As discussed in section 4.1, we need to construct a counter-model for proving the
invalidity of a formula, but we omit this here due to the space limitations.
 Conditional Logic Cb and Its Tableau System 199

Ex.2 (p q) > r, p > q  p > r

(p q) > r, 0
p > q, 0
(p > r), 0

0rp 1
p, 1
r, 1

q, 1
 
q, 1 q, 1
1rq 1

0rpq 1

r, 1

This inference is little changed from Ex. 1, and similarly to Ex. 1, it applies
the left rule as in section 5 above.
This formula creates a sentences like the following:
If A and B come, C will also come. And if A comes then B comes.
If A comes then C comes.

Ex.3 (p q) > r  (p > r) (q > r)

(p q) > r, 0
((p > r) (q > r)), 0
 
(p > r), 0 (q > r), 0

0rp 1 0rq 2
p, 1 q, 2
r, 1 r, 2

0rpq 1 0rpq 2

r, 1 r, 2

This is an example of applying the right rules proposed in section 5 above.

In order to apply the rule to the antecedent of the formula, it is necessary
to account for the accessibility relation of p q. Then, after 0rp 1 and 0rq 2 are
200 Y. Ozaki and D. Bekki

produced, they derive 0rpq 1 and 0rpq 2, respectively. Hence, closing becomes
possible by applying the rule to the antecedent of the given formula.
This formula creates sentences like the following:
If it rains or snows, the game will be cancelled.
If it rains, the game will be cancelled and if it snows, the game will be
cancelled.
Ex.4 p > q, (p r) > q  p > r

p > q, 0
(p r) > q, 0
(p > r), 0

0rp 1
p, 1
r, 1

q, 1
 
r, 1 r, 1
1rr 1
r, 1
0rpr 1

q, 1

Although the above inference is valid in Cb , there is a problem in regard to

its empirical validity. For example, the conditional corresponding to this formula
seems to be false: If there are various drinks there, I will go. Even though there
are various drinks, if there is no pizza, I wont go.
?If there are various drinks, there will be pizza.

Ex.5 (p r) > (r q), p > r  p > q

(p r) > (r q), 0
p > r, 0
(p > q), 0

0rp 1
p, 1
q, 1

r, 1
 
r, 1 r, 1
1rr 1

 Conditional Logic Cb and Its Tableau System 201

0rpr 1

r q, 1

r, 1
q, 1

This inference is also valid in Cb , but we cannot nd a conditional to corre-

spond to it. In such instance, the system may be too strong. Therefore, in future
research, we need to nd a suitable limitation.

7 Soundness

Our proof of the soundness and completeness of Cb is based on the proof of

soundness and completeness of C + given in [6].

Definition 1 (Faithfulness). Let I = W, R,

be any Kripke interpretation,
and b be any branch of a tableau. Then, I is faithful to b i there is a map
g : N W such that:

1. For every node A, i on b, A is true at g(i) in I.

2. If irA j is on b, g(i)RA g(j) in I.

Lemma 2 (Soundness Lemma). Let b be any branch of a tableau and I =

w, R,
be any Kripke interpretation. If I is faithful to b and a tableau rule is
applied to it, then it produces at least one extension b such that I is faithful to
b .

Proof. Since C + is proved to be sound with respect to its semantics given in

section 4.1, we merely have to check the case for each rule of Cb .
The argument for the left tableau rule in section 5 is the following. Suppose
there are irp j and jrq k on a branch to which I is faithful and we apply the left
rule and obtain irpq k. According to the denition of faithfulness, g(i)rp g(j) and
g(j)rq g(k) are in I. Hence:
| w (g(i)Rp w wRq x)}
g(k) {x 
fq (g(i))
w  fp (w)
fpq (g(i)) (according to Condition 8 in section 5)
{x | g(i)Rpq x} (according to the denition of fA (w))
Therefore, g(i)Rpq g(k), which shows that I is faithful to this extension.
For the right rules, suppose that there is irp j on a branch to which I is
faithful and from applying the rule we obtain irpq j. According to the denition
of faithfulness, g(i)rp g(j) is in I. Hence:
202 Y. Ozaki and D. Bekki

g(j) {x W | g(i)Rp x}
fp (g(i)) (according to the denition of fA (w))
fpq (g(i)) (according to Condition 9 in section 5)
{x W | g(i)Rpq x}

Therefore, g(i)Rpq (j), which shows that I is faithful to this extension. The case
for irq j can be proved in the same way.

Theorem 3 (Soundness Theorem). The tableau system of Cb is sound with

respect to its semantics, i.e. for nite , if  A then |= A.

Proof. Suppose that  A. Then we have the interpretation I = W, R,

that
makes every formula in true and A false, in some world w. Let h : N W
be a map such that h(0) = w, which makes I faithful to the initial list. When
we apply a rule to the list, there is at least one extension to which I is faithful,
due to the Soundness Lemma. Thus, if the tableau is closed, there is at least one
branch b for which the interpretation I is faithful, and there is a formula B such
that both B and B are on b. This is impossible, however, because it means
(B) = 1 and (B) = 1. Therefore, the tableau must be open, i.e.  A.

8 Completeness

Definition 4 (Induced Interpretation). Let b be an open branch of a tableau.

The interpretation I = W, R,
induced by b is dened as follows:

W = {wi | i occurs in b}
For any formula A:
wi RA wj i irA j is on b, if A occurs as the antecedent of a conditional
or negated conditional at a node of b.
wi RA wj i wj (A) = 1 otherwise.

1 if A, i occurs on b
wi (A) = 0 if A, i occurs on b

1 or 0 otherwise

Lemma 5 (Completeness Lemma). Let b be any open complete branch of a

tableau. Let I = W, R,
be the interpretation induced by b. Then:

if A, i is on b, then wi (A) = 1
if A, i is on b, then wi (A) = 0

Proof. Since the syntax of Cb is the same as that of C and C + , the Completeness
Lemma can be proved in the same way [6].

Theorem 6 (Completeness Theorem). The tableau system of Cb is complete

with respect to its Kripke semantics: for nite , if  A then  A.
 Conditional Logic Cb and Its Tableau System 203

Proof. Suppose that  A. Given an open branch b of the tableau, the inter-
pretation induced by b makes all the formulae in true and A false in w0 , by
the Completeness Lemma.
Now we must check that the induced interpretation satises conditions 1 and 2
of C + , whose proof is given in [6], and conditions 8 and 9 of Cb given in section 5.
Suppose that b is a completed open branch. For any formula A, either A occurs
on b as an antecedent or not. In the former case, the result holds according to the
denition of RA . In the latter case, let us check the two conditions of Cb in turn.
For 8, let wx be any world. Suppose that there exists a world wy such
that wi RA wy and wy RB wx . Then, according to the denition of induced
interpretations, irA y and yrB x occur on b. Since b is completed, irAB x
is also on b. Again, according to the denition of induced interpretations,
wi RAB wx holds, as required.
For 9, let wx be any world such that wi RA wx . According to the denition
of induced interpretations, irA x occurs on b. Since b is completed, irAB x
and irBA x occurs on b for any formula B. So, according to the denition
of induced interpretations, wi RAB wx and wi RBA wx hold, as required.
Hence  A.

9 Conclusion and Future Work

In this paper, we have explored in detail the problems of conditional sentences
in natural language and proposed a new logic system Cb by extending existing
conditional logics.
There are two advantages to our logic Cb : rst, as a semantic theory of nat-
ural language, it is empirically more correct than preceding analyses; second, a
tableau proof is available for Cb , which is sound and complete with respect to
its Kripke semantics.
The following gure shows a comparison between S and Cb in regard to their
valid inferences.

Cb
valid invalid
S
p > (q r)  p > q p > q, q > r  p > r
valid

p > (p > q)  p > q p > q, (p > r)  (p r) > q

(p q) > r  (p > r) (q > r)
invalid

(p q) > r  p > (q > r) (p r) > (r q)  (p > q) (r > q)

(p q) > r, p > q  p > r
p > q, (p r) > q  p > r
(p r) > (r q), p > r  p > q
204 Y. Ozaki and D. Bekki

This gure indicates that inferences with conditional formulae that include
or in the antecedent are valid only in Cb , constituting what we believe to be
a substantial extension.
For future work, further examination of the empirical validity of Cb is required.
Cases such as Ex. 4 and Ex. 5 in section 6 may be problematic for the current
version of Cb . Moreover, we should think about how to treat inferences with
a conditional formula whose antecedent contains the negation symbol A. The
availability of an automatic proof and its implementation remain as topics for
future work.

Acknowledgments. We wish to thank Eric McCready, Alastair Butler, Kei

Yoshimoto and Kenichi Asai for their comments on an earlier version of this pa-
per. We would also thank the reviewers of this paper for their useful and valuable
comments. Daisuke Bekki is partially supported by Grant-in-Aid for Young Sci-
entists (A), 22680013, 2010-203, from the Ministry of Education, Science, Sports
and Culture, Japan.

References
1. Chellas, B.F.: Basic conditional logic. Journal of Philosophical Logic 4, 53133
(1980)
2. Harper, W.L., Stalnaker, R., Pearce, G.: Ifs. Reidel, Dordrecht (1981)
3. Lewis, D.K.: Counterfactuals. Blackwell, Oxford (1973)
4. Lewis, D.K.: Counterfactuals and comparative possibility. Journal of Philosophical
Logic 2 (1973)
5. Muskens, R.: A compositional discourse representation theory. In: The 9th Amster-
dam Colloquium, pp. 467486 (1993)
6. Priest, G.: An Introduction to Non-Classical Logic. Cambridge Uniersity Press,
Cambridge (2008)
7. Read, S.: Thinking About Logic: An Introduction to the Philosopy of Logic. Oxford
University Press, Oxford (1994)
8. Stalnaker, R.: A Theory of Conditionals, in Studies in Logical Theory. American
Philosophical Quanterly Monograph Series, vol. 2. Basil Blackwell, Oxford (1968)