=Paper=
{{Paper
|id=Vol-1651/12340003
|storemode=property
|title=Syllogistic Reasoning under the Weak Completion Semantics
|pdfUrl=https://ceur-ws.org/Vol-1651/12340003.pdf
|volume=Vol-1651
|authors=Ana Costa,Emmanuelle-Anna Dietz,Steffen Hölldobler,Marco Ragni
|dblpUrl=https://dblp.org/rec/conf/ijcai/CostaDHR16
}}
==Syllogistic Reasoning under the Weak Completion Semantics==
https://ceur-ws.org/Vol-1651/12340003.pdf
Syllogistic Reasoning under
the Weak Completion Semantics
Ana Costa1 , Emmanuelle-Anna Dietz1 , Steffen Hölldobler1 , and Marco Ragni2?
1
International Center for Computational Logic, TU Dresden, Germany,
ana.oli.costa@gmail.com, {dietz,sh}@iccl.tu-dresden.de
2
Marco Ragni, Center for Cognitive Science, Freiburg, Germany,
ragni@cognition.uni-freiburg.de
Abstract. In a recent meta-analysis, Khemlani & Johnson-Laird (2012)
showed that the conclusions drawn by human reasoners in psychological
experiments about syllogistic reasoning are not the conclusions predicted
by classical first-order logic. Moreover, current cognitive theories deviate
significantly from the empirical data. In this paper we show how hu-
man syllogistic reasoning can be modelled under the weak completion
semantics, a new cognitive theory.
1 Introduction
The way of how humans ought to reason correctly about syllogisms has already
been investigated by Aristotle. A syllogism consists of two quantified statements
using some of the four quantifiers all (A), no (E), some (I), and some are not (O)1
about sets of entities which we denote in the following by a, b, c. An example is:
Some a are b (IE1)
No b are c
and the task is to draw a conclusion. Implicitly, most experiments expect to draw
a logical consequence from these so-called premises, e.g., ‘some a are not c’ in
classical first-order logic (FOL). The four quantifiers and their formalization in
FOL are given in Table 1. The entities can appear in four different orders called
figures as shown in Table 2. Hence, a problem can be completely specified by
the quantifiers of the first and second premise and the figure. E.g., the example
discussed above is denoted by IE1.
Altogether, there are 64 syllogisms and, if formalized in FOL, we can compute
their logical consequence in classical logic. However, a meta-analysis [15] based
on six experiments has shown that humans do not only systematically deviate
from the predictions of FOL but from any other of at least 12 cognitive theories.
In the case of IE1, besides the above mentioned logical consequence, a significant
number of humans answered ‘no a are c’ which does not follow from IE1 in FOL.
?
The authors are mentioned in alphabetical order.
1
We are using the classical abbreviations.
Mood Natural Language FOL Short
affirmative universal (A) all a are b ∀X(a(X) → b(X)) Aab
affirmative existential (I) some a are b ∃X(a(X) ∧ b(X)) Iab
negative universal (E) no a are b ∀X(a(X) → ¬b(X)) Eab
negative existential (O) some a are not b ∃X(a(X) ∧ ¬b(X)) Oab
Table 1. The four syllogistic moods together with their logical formalization.
Figure 1 Figure 2 Figure 3 Figure 4
First Premise a-b b-a a-b b-a
Second Premise b-c c-b c-b b-c
Table 2. The four figures used in syllogistic reasoning.
In recent years, a new cognitive theory based on the weak completion se-
mantics (WCS) has been developed. It has its roots in the ideas first expressed
by Stenning and van Lambalgen [23], but is mathematically sound [11], and has
been successfully applied – among others – to the suppression task [6], the se-
lection task [7], the belief bias effect [18,19,2], to reasoning about conditionals
[3,5] and to spatial reasoning [4]. Hence, it was natural to ask whether WCS
is competitive in syllogistic reasoning and how it performs with respect to the
cognitive theories considered in [15].
In the following, three important cognitive approaches are presented and
compared to predictions made by WCS. Firstly, we show how to adequately
represent the syllogisms in logic programs and, secondly, we demonstrate how
to reason with respect to them. Afterwards we compare the results under the
WCS with the results of FOL, the syntactic rule based theory PSYCOP [22],
the Verbal Model Theory [21] and the Mental Model Theory [13].2 These two
model-based theories performed the best in the meta-analysis.
The predictions of the theories FOL, PSYCOP, Verbal, and Mental Models
for the syllogisms OA3, EA3, and AA4 and those of the participants, taken
from [15], are depicted in Table 3, where the participants were 156 high school
to university students. The significant percentage of participants means that the
number of participants who chose for the particular conclusion, was too high for
2
http://mentalmodels.princeton.edu/models/mreasoner/
Participants FOL PSYCOP Verbal Models Mental Models
OA4 Oca Oca Oca, Ica, Iac Ocs, NVC Oca, Oac, NVC
EA3 Eac Eac, Eca Eac, Eca NVC, Eca Eac, Eca
Oac, Oca Oac, Oca
AA4 Aac, NVC Iac, Ica Iac, Ica NVC, Aca Aca, Aac,
Iac, Ica
Table 3. The conclusions drawn by a significant percentage of participants are high-
lighted in gray and compared to the predictions of the theories FOL, PSYCOP, Verbal,
and Mental Models for the syllogisms OA4, EA3, and AA4. NVC stands for no valid
conclusion.
the conclusion to be chosen randomly.3 The interested reader is referred to [15]
for more details.
FOL and the other three cognitive theories make different predictions. In
particular, each theory provides at least one prediction which is correct with
respect to classical FOL and provides an additional prediction which is false with
respect to classical FOL. Currently, the best overall results are achieved by the
Verbal Models Theory which predicts 84% of the participants responses, closely
followed by the Mental Model Theory with 83%, whereas PSYCOP predicts 77%
of the participants responses.
2 Weak Completion Semantics
The general notation, which we will use in the paper, is based on [16,10].
2.1 Logic Programs
We assume the reader to be familiar with logic and logic programming, but
recall basic notions and notations. A (logic) program is a finite set of (program)
clauses of the form A ← >, A ← ⊥ or A ← B1 ∧ . . . ∧ Bn , n > 0, where A is
an atom, Bi , 1 ≤ i ≤ n, are literals and > and ⊥ denote truth and falsehood,
respectively. A is called head and >, ⊥ as well as B1 ∧ . . . ∧ Bn are called body
of the corresponding clause. Clauses of the form A ← > and A ← ⊥4 are called
3
The threshold for the percentage to be significant is determined as follows: Given
that there are nine different conclusion possibilities, the chance that a conclusion has
been chosen randomly is 1/9 = 11.1%. A binomial test shows that if a conclusion is
drawn in more than 16% of the cases by the participants it is unlikely that is has
been chosen by just random guesses. The statistical analysis is elaborately explained
in [15].
4
We consider weak completion semantics and, hence, a clause of the form A ← ⊥ is
turned into A ↔ ⊥ provided that this is the only clause where A is the head of.
F ¬F ∧>U ⊥ ∨>U ⊥ ←>U ⊥ ↔>U⊥
> ⊥ >>U ⊥ >>>> > >>> > >U⊥
⊥ > UU U ⊥ U>U U U U >> U U>U
U U ⊥⊥⊥⊥ ⊥>U ⊥ ⊥ ⊥U > ⊥ ⊥U>
Table 4. The truth tables for the connectives under L-logic. >, ⊥, and U denote true,
false, and unknown, respectively.
positive and negative facts, respectively. We restrict terms to be constants and
variables only, i.e., we consider data logic programs. Throughout this paper, P
denotes a program. We assume for each P that the alphabet consists precisely
of the symbols occurring in P and that non-propositional programs contain at
least one constant.
gP denotes the set of all ground instances of clauses occurring in P, where
a ground instance of clause C is obtained from C by replacing each variable
occurring in C by a term not containing any variables. A ground atom A is
defined in gP iff gP contains a clause whose head is A; otherwise A is said to be
undefined. def (S, P) = {A ← Body ∈ gP | A ∈ S ∨ ¬A ∈ S} is called definition
of S in P, where S is a set of ground literals. S is said to be consistent iff it does
not contain a pair of complementary literals. The set of atoms occurring in gP
is denoted as atoms(P).
2.2 Three-Valued Lukasiewicz Logic
We consider the three-valued Lukasiewicz logic (L-logic, [17]), for which the cor-
responding truth values are >, ⊥ and U, which mean true, false and unknown,
respectively. A three-valued interpretation I is a mapping from formulas to a set
of truth values {>, ⊥, U}. The truth value of a given formula under I is deter-
mined according to the truth tables in Table 4. We represent an interpretation
as a pair I = hI > , I ⊥ i of disjoint sets of atoms, where I > is the set of all atoms
that are mapped to > by I, and I ⊥ is the set of all atoms that are mapped to ⊥
by I. Atoms which do not occur in I > ∪ I ⊥ are mapped to U. Let I = hI > , I ⊥ i
and J = hJ > , J ⊥ i be two interpretations: I ⊆ J iff I > ⊆ J > and I ⊥ ⊆ J ⊥ .
I(F ) = > means that a formula F is mapped to true under I. M is a model
of P if it is an interpretation, which maps each clause occurring in gP to >. I
is the least model of P iff for any other model J of P it holds that I ⊆ J.
2.3 Least Models under the Weak Completion
For a given P, consider the following transformation:
1. For each ground atom A which is defined in P, replace all clauses of the form
A ← Body1 , . . . , A ← Bodym occurring in gP by A ← Body1 ∨ . . . ∨ Bodym .
2. Replace all occurrences of ← by ↔.
The obtained ground program is called weak completion of P or wcP.5
It has been shown in [12] that logic programs as well as their weak completions
admit a least model under L-logic. Moreover, the least L-model of wcP can be
obtained as the least fixed point of the following semantic operator, which is
due to Stenning and van Lambalgen [23]: Let I = hI > , I ⊥ i be an interpretation.
ΦP (I) = hJ > , J ⊥ i, where
J > = {A | A ← Body ∈ def (A, P) and Body is true under hI > , I ⊥ i}
J ⊥ = {A | def (A, P) 6= ∅ and
Body is false under hI > , I ⊥ i for all A ← Body ∈ def (A, P)}
Weak completion semantics (WCS) is the approach to consider weakly com-
pleted logic programs and to reason with respect to the least L-models of these
programs. We write P |=wcs F iff formula F holds in the least L-model of wcP.
In the remainder of this paper, MP denotes the least L-model of wcP.
The correspondence between weak completion semantics and well-founded
semantics [24] for tight programs, i.e. those without positive cycles, is shown
in [8].
2.4 Integrity Constraints
A set of integrity constraints IC consists of clauses of the form U ← Body,
where Body is a conjunction of literals and U denotes the unknown.6 Hence, an
interpretation maps an integrity constraint to > iff Body is either mapped to U
or ⊥. This understanding is similar to the definition of the integrity constraints
for the well-founded semantics in [20]. Given an interpretation I and a set of
integrity constraints IC, I satisfies IC iff all clauses in IC are true under I.
3 Reasoning Towards an Appropriate Logical Form
We will apply four principles in developing a logical form for the representation
of syllogisms.
3.1 Licenses for Inferences
Stenning and van Lambalgen [23] propose to formalize conditionals in human
reasoning not by inferences straight away, but rather by licenses for inferences.
For example, the conditional if p(X) then q(X) is represented by the program
{q(X) ← p(X) ∧ ¬ab(X), ab(X) ← ⊥},
which states that q(X) holds if p(X) and ¬ab(X) hold and ¬ab(X) holds, where
¬ab(X) means that nothing is abnormal for X with respect to this clause.
5
Note that undefined atoms are not identified with ⊥ as in the completion of P [1].
6
Formally, we need to extend the alphabet by this symbol.
3.2 Existential Import and Gricean Implicature
Humans do understand quantifiers differently due to a pragmatic understand-
ing of language. For instance, in natural language we normally do not quantify
over things that do not exist. Consequently, for all implies there exists. This
appears to be in line with human reasoning and has been called the Gricean
Implicature [9]. Several theories like the theory of mental models [14] or men-
tal logic [22] assume that the sets we quantify about are not empty. Likewise,
Stenning and van Lambalgen [23] have shown that humans require existential
import for a conditional to be true. Furthermore, as mentioned in [15], the quan-
tifier ‘some a are b’ often implies that ‘some a are not b’, which again can be
explained by assuming the Gricean Implicature: Someone would not state ‘some
a are b’ if that person knew that ‘all a are b’. As the person does not say ‘all a
are b’ but instead ‘some a are b’, we have to assume that ‘not all a are b’, which
in turn implies ‘some a are not b’.
3.3 Negation by Transformation
Logic programs do not allow negative literals as heads of clauses. In order to rep-
resent a negative conclusion ¬p(X) an auxiliary formula p0 (X) is used together
with a clause p(X) ← ¬p0 (X) and the integrity constraint U ← p(X) ∧ p0 (X).
This is a widely used technique in logic programming. Together with the prin-
ciple discussed in Section 3.1, the additional clause becomes p(X) ← ¬p0 (X) ∧
¬abnpp (X), and its weak completion is p(X) ↔ ¬p0 (X) ∧ ¬abnpp (X), stating
that p0 is the negation of p iff nothing abnormal is known with respect to that
clause. The integrity constraint states that an object cannot belong to both, p
and p0 . We call this principle negation by transformation.
3.4 Unknown Generalization
Humans seem to distinguish between ‘some y are z ’ and ‘some z are y’, as the
results reported in [15] show. However, if we would represent ‘some y are z ’ by
∃X(y(X) ∧ z(X)) then this is semantically equivalent to ∃X(z(X) ∧ y(X)) in
FOL because conjunction is commutative. Likewise, humans seem to distinguish
between ‘some y are z ’ and ‘all y are z ’, as we have already explained in Sec-
tion 3.2. Accordingly, if we only observe that an object o belongs to y and z then
we do not want to conclude both, ‘some y are z ’ and ‘all y are z ’.
In order to prevent such unwanted conclusions we introduce the following
principle: if we know that ‘some y are z ’ then there must not only be an object o1
which belongs to y and z (by Gricean implicature) but there must be another
object o2 which belongs to y and for which it is unknown whether it belongs
to z. We call this principle unknown generalization.
4 Representing the Syllogisms
Based on the principles presented in the previous section, we can now repre-
sent the syllogisms by logic programs. The programs will be specified using the
predicates y and z and depending on the figures shown in Table 2: yz must be
replaced by ab, ba, cb or bc.
4.1 All (A)
‘All y are z ’ is represented by the program PAyz which consists of the following
clauses:
z(X) ← y(X) ∧ ¬abyz (X)
abyz (X) ← ⊥
y(o) ← >
The first two clauses are obtained by applying the principle of using licenses for
inferences. The last clause follows by the principle of Gricean implicature, where
o is the object which is assumed to exist for y. The least L-model for wcPAyz is
h{y(o), z(o)}, {abyz (o)}i.
4.2 No (E)
‘No y are z ’ is represented by the program PEyz which consists of the following
clauses:
z 0 (X) ← y(X) ∧ ¬abynz (X)
abynz (X) ← ⊥
z(X) ← ¬z 0 (X) ∧ ¬abnzz (X)
y(o) ← >
abnzz (o) ← ⊥
In addition we need the integrity constraint
U ← z(X) ∧ z 0 (X).
The first two clauses are obtained by applying the principle of using licenses for
inferences, where z 0 is an auxiliary predicate symbol used to denote the negation
of z. This auxiliary predicate is formally related to z by the third clause applying
the principle of negation by transformation. In addition, this principle enforces
the integrity constraint. The fourth clause of PEyz follows by the principle of
Gricean implicature.
The last clause cannot be generalized to all X, because otherwise we al-
low conclusions by double negation. For example, consider the case where there
is some o0 for which y(o0 ) is false due to some clause being part of another
premise. In this case, the first clause will enforce the falsehood of z 0 (o0 ). Now,
if abnzz (X) ← ⊥ would hold, then abnzz (o0 ) would be false and, consequently,
by the third clause, z(o0 ) would be true. In other words, z(o0 ) follows by the
negation of y(o0 ), which in turn is responsible for the negation of z 0 (o0 ). Even
though this might be logically reasonable, the empirical results indicate that
participants do not infer conclusions based on double negation. Therefore, we
decided to restrict abnzz (o) ← ⊥ to the objects occurring in PEyz .
The least L-model for wcPEyz is
h{y(o), z 0 (o)}, {abynx (o), abnzz (o), z(o)}i.
4.3 Some (I)
‘Some y are z ’ is represented by the program PIyz which consists of the following
clauses:
z(X) ← y(X) ∧ ¬abyz (X)
abyz (o1 ) ← ⊥
y(o1 ) ← >
y(o2 ) ← >
The first two clauses are again obtained by the principle of using licenses for
inferences. However, the abnormality predicate is restricted to the object o1 ,
which is assumed to exist by the principle of Gricean implicature (see third
clause). The fourth clause is obtained by the principle of unknown generalization.
The least L-model of wcPIyz is
h{y(o1 ), y(o2 ), z(o1 )}, {abyz (o1 )}.
Note that nothing is stated in PIyz about abyz (o2 ). Accordingly, z(o2 ) stays
unknown in the least L-model.
4.4 Some Are Not (O)
‘Some y are not z ’ is represented by the program POyz which consists of the
following clauses:
z 0 (X) ← y(X) ∧ ¬abynz (X)
abynz (o1 ) ← ⊥
z(X) ← ¬z 0 (X) ∧ ¬abnzz (X)
y(o1 ) ← >
y(o2 ) ← >
abnzz (o1 ) ← ⊥
abnzz (o2 ) ← ⊥
In addition, we need the integrity constraint
U ← z(X) ∧ z 0 (X).
The first four clauses as well as the integrity constraints are derived as in the
program PEyz except that object o1 is used instead of o and abynz is restricted to
o1 like in PIyz . The fifth clause of POyz is obtained by the principle of unknown
generalization. The last two clauses are again not generalized to all objects for
the same reason as previously discussed in Section 5.2 for the representation of
E. The least L-model for wcPOyz is
h{y(o1 ), y(o2 ), z 0 (o1 )}, {abynz (o1 ), abnzz (o1 ), abnzz (o2 ), z(o1 )}i.
5 Entailment of Syllogisms
We can now combine the proposed representations with respect to the figures in
Table 2. In doing so, we replace yz by ab, ba, cb or bc. In addition, we may need to
rename objects such that different objects are referred to in the representations
of different syllogisms. Thereafter, we compute the least L-model of the obtained
programs and check which syllogisms hold in this model.
However, we have not yet defined when a syllogism holds given a program P.
These definitions will be developed in this section.
One should observe that [15] does not contain a formal definition for the
entailment of syllogisms. They use first-order theory as a normative theory, i.e.,
they test if the conclusions drawn by the participants are correct with respect to a
first-order representation of a syllogism. In the following an entailment regarding
the weak completion semantics is presented, where yz is to be replaced by ab,
ba, cb, bc, ac, or ca.
5.1 All (A)
P |= Ayz iff there exists an object o such that P |=wcs y(o) and for all objects o
we find that if P |=wcs y(o) then P |=wcs z(o) ∧ ¬aby z(o).
The existence of an object o belonging to y is due to the principle of Gricean
implicature. Moreover, all objects belonging to y must belong to z. The require-
ment that ¬abyz (o) is also entailed is a technical one which is based on the
principle of licences for inferences. We may omit these abnormalities to obtain
the following alternative definition.
P |=0 Ayz iff there exists an object o such that P |=wcs y(o) and for all
objects o we find that if P |=wcs y(o) then P |=wcs z(o).
5.2 No (E)
P |= Eyz iff there exists an object o such that P |=wcs y(o) and for all objects o
we find that if P |=wcs y(o) then P |=wcs z 0 (o) ∧ ¬z(o) ∧ ¬abynz (o) ∧ ¬abnzz (o).
The existence of an object o belonging to y is again due to the principle of
Gricean implicature. Moreover, all objects o belonging to y must not belong to
z and, hence, ¬z(o) must be entailed. The remaining entailed formulas are due
to the principles of negation by transformation and of licenses for inferences. We
may omit these formulas to obtain the following alternative definition.
P |=0 Eyz iff there exists an object o such that P |=wcs y(o) and for all
objects o we find that if P |=wcs y(o) then P |=wcs ¬z(o).
5.3 Some (I)
P |= Iyz iff there exists an object o1 such that P |=wcs y(o1 ) ∧ z(o1 ) ∧ ¬abyz (o1 )
and there exists an object o2 such that P |=wcs y(o2 ) and P 6|=wcs z(o2 ) ∧
¬abyz (o2 ).
The existence of an object o1 belonging to y is again due to the principle
of Gricean implicature. This object o1 must also belong to z. In addition, there
must be another object o2 belonging to y for which it is unknown whether it
belongs to z due to the principle of unknown generalization. The abnormality
predicates ¬abyz (o1 ) and ¬abyz (o2 ) are again due to the principle of licenses for
inferences. We may omit these abnormalities to obtain the following alternative
definition.
P |=0 Iyz iff there exists an object o1 such that P |=wcs y(o1 ) ∧ z(o1 ) and
there exists an object o2 such that P |=wcs y(o2 ) and P 6|=wcs z(o2 ).
5.4 Some Are Not (O)
P |= Oyz iff there exists an object o1 such that P |=wcs y(o1 ) ∧ z 0 (o1 ) ∧ ¬z(o1 ) ∧
¬abynz (o1 ) ∧ ¬abnzz (o1 ) and there exists an object o2 such that P |=wcs y(o2 )
and P 6|=wcs z 0 (o2 ) ∧ ¬z(o2 ) ∧ ¬aby nz(o2 ) ∧ ¬abnzz (o2 ).
This case combines E and I. First of all, there must be an object o1 which
belongs to y and does not belong to z. Moreover, there must be another object
o2 which belongs to y and for which it is unknown whether it belongs to z. If we
omit the abnormalities and the auxiliary predicate z 0 the following alternative
definition is obtained.
P |=0 Oyz iff there exists an object o1 such that P |=wcs y(o1 ) ∧ ¬z(o1 ) and
there exists an object o2 such that P |=wcs y(o2 ) and P 6|=wcs ¬z(o2 ).
5.5 No Valid Conclusion (NVC)
When we can not conclude any of the previous moods, then we derive from P
that no valid conclusion holds.
6 Predictions by the Weak Completion Semantics
Combining the syllogisms representation and entailment rules explained before
we accomplished an average of 85% accuracy in our predictions. In 9 cases we
have a perfect match with the answers given by the participants. In 29 cases the
match is 89% and in 21 cases the match is 78%. The five cases left, have a match
of 67%.
First, we will explain how the accuracy of the predictions is computed in
general. After that we will show the logic program representation of three syl-
logisms with different predictions accuracy, and compare their results with the
data from human experiments taken from [15]. In those experiments people were
asked to infer conclusions about the predicates a and c from a syllogism built
according to the figures in Table 2. Therefore, in our predictions we only consider
entailment of syllogisms between those two predicates, a and c.
6.1 Accuracy of Predictions
We have nine different answer possibilities for each of the 64 syllogisms:
Aac, Eac, Iac, Oac, Aca, Eca, Ica, Oca and NVC.
For every syllogism, we define a list of length 9 for the predictions of the weak
completion semantics, where the first element represents Aac, the second element
represents Eac, and so forth. When Aac is predicted under the weak completion
semantics for a given syllogism, then the value of the first element of this list
is a 1, otherwise it is a 0, and the same holds for the other eight elements in
the list (representing the other eight answer possibilities). Analogously, for every
syllogism we define a list of the participants’ conclusions of length 9 containing
either 1 or 0 for all nine answer possibilities, depending on whether the majority
concluded Aac, Eac, and so forth. For each syllogism we then simply compare
each element of both lists as follows, where i is the ith element of both lists:
(
1 if both lists have the same value for the ith element
comp(i) =
0 otherwise
P9
The matching percentage of this syllogism is then computed by i=1 comp(i)/9.
Note that the percentage of the match does not only take in account when the
weak completion semantics correctly predicts a conclusion, but also whenever
it correctly rejected a conclusion. The average percentage of accuracy is then
simply the average of the matching percentage of all 64 syllogisms.
6.2 OA4 - Perfect Match (1.00)
The syllogism OA4 is obtained by combining the last and the first mood in
Table 1 according to the figure 4 in Table 2. It can be read as:
First Premise Oba ‘Some b are not a’
Second Premise Abc ‘All b are c’
The program POA4 representing the two premises is obtained as the union
of the programs POba (obtained from POyz by replacing y and z by b and a,
respectively) and PAbc (obtained from PAyz by replacing y and z by b and c,
respectively). In addition, the constant o occurring in PAbc has been replaced
by o3 . POA4 consists of the following clauses:
a0 (X) ← b(X) ∧ ¬abbna (X)
abbna (o1 ) ← ⊥
a(X) ← ¬a0 (X) ∧ ¬abnaa (X)
b(o1 ) ← >
b(o2 ) ← >
abnaa (o1 ) ← ⊥
abnaa (o2 ) ← ⊥
c(X) ← b(X) ∧ ¬abbc (X)
abbc (X) ← ⊥
b(o3 ) ← >
The least L-model of POA4 is
h {b(o1 ), b(o2 ), b(o3 ), abca (o1 ), a0 (o1 ), c(o1 ), c(o2 ), c(o3 )},
{abbna (o1 ), abnaa (o1 ), abbc (o1 ), abbc (o2 ), abbc (o3 ), a(o1 )} i.
This model entails only the conclusion ‘some c are not a’. We have an object
that is in c and not in a (POA4 |=wcs c(o1 ) and POA4 |=wcs ¬a(o1 )) and an
object that is in c and it is not known if it is not in a (POA4 |=wcs c(o2 ) and
POA4 6|=wcs ¬a(o2 )). This prediction matches perfectly with the answers from
the participants.
6.3 EA3 - Worst Match (0.67)
We will discuss now the syllogism EA3, one of the syllogisms with the lowest
match (67%). EA3 can be read as:
First Premise Eab ‘No a are b’
Second Premise Acb ‘All c are b’
The program PEA3 representing the two premises is obtained as the union of the
programs PEab (obtained from PEyz by replacing y and z by a and b, respectively)
and PAcb (obtained from PEyz by replacing y and z by c and b, respectively).
In addition the constant o occurring in PEab and PAcb have been replaced by o1
and o2 , respectively. PEA3 consists of the following clauses:
b0 (X) ← a(X) ∧ ¬abanb (X)
abanb (X) ← ⊥
b(X) ← ¬b0 (X) ∧ ¬abnbb (X)
abnbb (o1 ) ← ⊥
a(o1 ) ← >
b(X) ← c(X) ∧ ¬abcb (X)
abcb (X) ← ⊥
c(o2 ) ← >
The least L-model of PEA3 is
h {a(o1 ), c(o2 ), b0 (o1 ), b(o2 )},
{abanb (o1 ), abnbb (o1 ), abanb (o2 ), abnbb (o2 ), abcb (o1 ), abcb (o2 )} i.
This model does not entail any conclusion between a and c, so our prediction is
NVC. Participants concluded both Eac and Eca.
Even though none of the entailed conclusions predicted by the weak completion
semantics match to the participants’ answers, a match of 67 % is computed. The
reason is that we also take into account the correct rejections, i.e. the conclusions
that are not entailed, as we have explained in Section 6.1
6.4 AA4 - Partial Match (0.78)
For AA4 we got an accuracy of 78% in our prediction. This syllogism combines
two premises in the first mood according to the figure 4. It can be read as:
First Premise Aba ‘All b are a’
Second Premise Abc ‘All b are c’
The program PAA4 representing the two premises is obtained as the union of
programs PAba (obtained from PAyz by replacing y and z by b and a, respectively)
and PAbc (obtained from PAyz by replacing y and z by b and c, respectively).
In addition the constant o occurring in PAba and PAbc have been replaced by o1
and o2 , respectively. PAA4 consists of the following clauses:
a(X) ← b(X) ∧ ¬abba (X)
b(o1 ) ← >
abba (X) ← ⊥
c(X) ← b(X) ∧ ¬abbc (X)
abbc (X) ← ⊥
b(o2 ) ← >
The least L-model of PAA4 is
h {b(o1 ), b(o2 ), a(o1 ), a(o2 ), c(o1 ), c(o2 )},
{abba (o1 ), abba (o2 ), abbc (o1 ), abbc (o2 )} i.
This model entails both ‘all a are c’ and ‘all c are a’. We have an object that is
in a, and for all objects that are in a (P |=wcs a(o1 ) and P |=wcs a(o2 )) it holds
that they are also in c (P |=wcs c(o1 ) and P |=wcs c(o2 )). Analogously this also
holds for ‘all c are a’. This prediction matches partially with the answers from
participants who concluded Aac and NVC.
7 Results
We discussed three examples, which we formalized under WCS. The results are
summarized and compared to FOL, PSYCOP, the Verbal, and the Mental Model
Theory in Table 5. The selected examples are typical in the sense that for some
syllogisms the conclusions drawn by the participants and WCS are identical, for
some syllogisms the conclusions drawn by the participants and WCS overlap,
and for some syllogisms the conclusions drawn by the participants and WCS are
disjoint. Moreover, WCS differs from the other cognitive theories.
The overall result with respect to the 64 syllogisms under WCS shows that we
can predict 85% of the participants responses. Compared to the other cognitive
theories, we achieve the best performance, which closely followed by the Verbal
Models Theory (84%) and the Mental Model Theory (83%). It seems natural to
compare these theories in more detail and see where their similarities are and
where they differ.
Participants FOL PSYCOP Verbal Models Mental Models WCS
OA4 Oca Oca Oca, Ica, Iac Ocs, NVC Oca, Oac, NVC Oca
EA3 Eac, Eac, Eca Eac, Eca NVC, Eca Eac, Eca NVC
Eca Oac, Oca Oac, Oca
AA4 Aac, Iac, Ica Iac, Ica NVC, Aca Aca, Aac, Aac,
NVC Iac, Ica Aca
Table 5. The conclusions drawn by a significant percentage of participants are high-
lighted in gray and compared to the predictions of the theories FOL, PSYCOP, Verbal,
and Mental Models as well as WCS for the syllogisms OA4, EA3, and AA4.
Another aspect that might be interesting to investigate is whether the combina-
tions of the moods influence how the participants perceive the syllogisms. For
instance, it seems that participants give different answers when they consider
syllogistic premises of the same mood, especially in the cases for AA and EE,
than when they consider syllogistic premises of two different moods. However,
this observation needs to be further examined, and can possibly give us more
insight about the human reasoning process.
References
1. K. L. Clark. Negation as failure. In H. Gallaire and J. Minker, editors, Logic and
Data Bases, volume 1, pages 293–322. Plenum Press, New York, NY, 1978.
2. E.-A. Dietz. A computational logic approach to syllogisms in human reasoning. In
Claudia Schon Ulrich Furbach, editor, Proceedings of the Workshop on Bridging
the Gap between Human and Automated Reasoning on the 25th International Con-
ference on Automated Deduction (CADE-25), pages 17–31. CEUR-WS.org, 2015.
3. E.-A. Dietz and S. Hölldobler. A new computational logic approach to reason with
conditionals. In F. Calimeri, G. Ianni, and M. Truszczynski, editors, Logic Pro-
gramming and Nonmonotonic Reasoning, 13th International Conference, LPNMR,
volume 9345 of Lecture Notes in Artificial Intelligence, pages 265–278. Springer,
2015.
4. E.-A. Dietz, S. Hölldobler, and R. Höps. A computational logic approach to human
spatial reasoning. In IEEE Symposium on Human-Like Intelligence (CIHLI), 2015.
5. E.-A. Dietz, S. Hölldobler, and L. M. Pereira. On conditionals. In G. Gottlob,
G. Sutcliffe, and A. Voronkov, editors, Global Conference on Artificial Intelligence,
Epic Series in Computing. EasyChair, 2015.
6. E.-A. Dietz, S. Hölldobler, and M. Ragni. A computational logic approach to the
suppression task. In N. Miyake, D. Peebles, and R. P. Cooper, editors, Proceedings
of the 34th Annual Conference of the Cognitive Science Society, pages 1500–1505,
Austin, TX, 2012.
7. E.-A. Dietz, S. Hölldobler, and M. Ragni. A computational logic approach to the
abstract and the social case of the selection task. In 11th International Symposium
on Logical Formalizations of Commonsense Reasoning, 2013.
8. E.-A. Dietz, S. Hölldobler, and C. Wernhard. Modeling the suppression task under
weak completion and well-founded semantics. Journal of Applied Non-Classsical
Logics, 24(1–2):61–85, 2014.
9. H Paul Grice. Logic and conversation. reprinted in studies in the way of words,
ed. hp grice.(1989), 1975.
10. S. Hölldober. Logik und Logikprogrammierung, volume 1: Grundlagen. Synchron
Publishers GmbH, Heidelberg, 2009.
11. S. Hölldobler and C. D. Kencana Ramli. Logic programs under three-valued
Lukasiewicz semantics. In P. M. Hill and D. S. Warren, editors, Logic Program-
ming, 25th International Conference, ICLP 2009, volume 5649 of Lecture Notes in
Computer Science, pages 464–478, Heidelberg, 2009. Springer.
12. S. Hölldobler and C. D. Kencana Ramli. Logics and networks for human reasoning.
In International Conference on Artificial Neural Networks, ICANN 2009, Part II,
pages 85–94, 2009.
13. P. N. Johnson-Laird. Mental models: towards a cognitive science of language, in-
ference, and consciousness. Harvard University Press, Cambridge, MA, 1983.
14. P. N Johnson-Laird. Mental models: Towards a cognitive science of language,
inference, and consciousness. Harvard University Press, 1983.
15. S. Khemlani and P. N. Johnson-Laird. Theories of the syllogism: A meta-analysis.
Psychological Bulletin, pages 427–457, 2012.
16. J. W. Lloyd. Foundations of Logic Programming. Springer-Verlag New York, Inc.,
New York, NY, USA, 1984.
17. Jan Lukasiewicz. O logice trójwartościowej. Ruch Filozoficzny, 5:169–171, 1920.
English translation: On three-valued logic. In: Lukasiewicz J. and Borkowski L.
(ed.). (1990). Selected Works, Amsterdam: North Holland, pp. 87–88.
18. L. M. Pereira, E.-A. Dietz, and S. Hölldobler. An abductive reasoning approach to
the belief-bias effect. In C. Baral, G. De Giacomo, and T. Eiter, editors, Principles
of Knowledge Representation and Reasoning: Proceedings of the 14th International
Conference, pages 653–656. AAAI Press, 2014.
19. L. M. Pereira, E.-A. Dietz, and S. Hölldobler. Contextual abductive reasoning
with side-effects. In I. Niemelä, editor, Theory and Practice of Logic Programming
(TPLP), volume 14, pages 633–648. Cambridge University Press, 2014.
20. Luı́s M. Pereira, Joaquim Nunes Aparı́cio, and José Júlio Alferes. Hypothetical
reasoning with well founded semantics. In B. Mayoh, editor, Scandinavian Con-
ference on Artificial Intelligence: Proc. of the SCAI’91, pages 289–300. IOS Press,
Amsterdam, 1991.
21. Thad A Polk and Allen Newell. Deduction as verbal reasoning. Psychological
Review, 102(3):533, 1995.
22. L. J. Rips. The psychology of proof: Deductive reasoning in human thinking. The
MIT Press, Cambridge, MA, 1994.
23. K. Stenning and M. van Lambalgen. Human Reasoning and Cognitive Science. A
Bradford Book. MIT Press, Cambridge, MA, 2008.
24. Allen Van Gelder, Kenneth A. Ross, and John S. Schlipf. The well-founded se-
mantics for general logic programs. Journal of the ACM, 38(3):619–649, 1991.