A Logical Characterization of Evaluable Knowledge Bases
Alexander Sakharov
Synstretch, Framingham, MA, U.S.A.
Keywords:
Evaluable Function/predicate, Reductio Ad Absurdum, 3-Valued Model, Substructural Sequent Calculus.
Abstract:
Evaluable knowledge bases are comprised of non-Horn rules with partial predicates and functions, some
of them are defined as recursive functions. This paper investigates logical foundations of the derivation of
literals from evaluable knowledge bases without reasoning by contradiction. The semantics of this inference
is specified by constrained 3-valued models. The derivation of literals without reasoning by contradiction is
characterized by means of sequent calculi with non-logical axioms expressing knowledge base rules and facts.
The logical rules of these calculi include only the negation rules, and cut is the only essential structural rule.
1 INTRODUCTION
A great deal of research has been devoted to logical
characterizations of emerging AI methods. Discov-
ering logical characterizations is important because
it brings clarity to the methods. Semantics establish
the legitimacy of method results. Calculi make
the results of these methods explainable. Besides,
the apparatus of mathematical logic can be used
to analyze AI methods and compare them. Both
first-order logic (FOL) and nonstandard logics are
used in these characterizations. Recent advances in
AI have given rise to interest in nonstandard logics
including description logics, argumentation logics,
default logic, etc. One advantage of nonstandard
characterizations is that they are usually simpler than
FOL.
The languages of logic programs and knowledge
bases (KB) are usually based on FOL (Russell and
Norvig, 2009). And FOL is normally used for
the characterization of relevant inference methods.
Nonetheless, inference for KBs and logic programs
differs significantly from inference in FOL. The
outcome of this inference and its intermediate steps
is literals as opposed to arbitrary FOL formulas. The
number of rules and especially the number of facts
involved in KB inference may be huge whereas FOL
axiom sets tend to be compact. The interpretation of
negation may be different from FOL, which entails
nonstandard semantics and calculi. KBs and logic
programs may include computable (aka evaluable)
functions and predicates (McCune, 2003).
Generally, KB facts are atoms or literals. Atoms
are expressions P(t
1
, ..., t
k
) where P is a predicate
and t
1
, ..., t
k
are terms. Literals are atoms or their
negations. Non-Horn rules are expressions A ⇐ A
1
∧
... ∧ A
k
, where A, A
1
, ..., A
k
are literals. FOL formulas
can be expressed by equisatisfiable conjunctions of
non-Horn clauses. In Horn rules, A and all A
i
are
atoms. In normal logic programs, A is an atom and
A
i
are literals.
The semantics of KB with Horn or non-Horn
rules/facts is given by classical 2-valued FOL models.
FOL calculi are used as the proof theories of Horn
and non-Horn KBs. FOL models and proof theories
are not adequate for normal logic programs, the re-
spective logical characterizations are nonmonotonic.
Availability of rules with both positive and negative
heads in non-Horn KBs helps avoid controversies
related to logical characterizations of normal logic
programs (Denecker et al., 2017). Half-finished non-
Horn KBs are not prone to erroneous derivations of
negative literals unlike nonmonotonic systems.
Evaluable functions and predicates may be partial,
and thus, their calculation may not yield a result.
In FOL, it is implicitly assumed that all functions
and predicates are total, they are not associated with
algorithms, and terms with constant arguments are
not mapped to constants unless we use equational
extensions of FOL. The computational nature of KBs
and logic programs is lost in this FOL treatment of
rules and facts. KBs with non-Horn rules/facts and
partial predicates and functions, some of which are
evaluable, will be called evaluable KBs.
The principle of Reductio Ad Absurdum (RAA)
states that if A is deduced from a hypothesis that is A’s
complement, then A is derivable. In author’s opinion,
Sakharov, A.
A Logical Characterization of Evaluable Knowledge Bases.
DOI: 10.5220/0010886700003116
In Proceedings of the 14th International Conference on Agents and Artificial Intelligence (ICAART 2022) - Volume 3, pages 681-688
ISBN: 978-989-758-547-0; ISSN: 2184-433X
Copyright
c
 2022 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
681

the legitimacy of RAA is questionable in the presence
of partial predicates or functions. It seems faulty to
apply RAA when the truth value of the hypothesis is
undefined because the hypothesis complement is also
undefined. For example, should P(a) be derivable
from fact R(a) and rules P(x) ⇐ R(x) ∧Q(x), P(x) ⇐
¬Q(x) if the computation of Q(a) does not terminate?
Derivations without RAA are often called direct.
The aim of this paper is to specify model and
proof theories for inference from evaluable KBs.
Inference without using RAA, i.e. without reasoning
by contradiction, seems more adequate for evaluable
KBs than FOL. There exists a variety of methods that
implement KB inference without reasoning by con-
tradiction (Sakharov, 2021). These methods include
an adaptation of highly efficient ordered resolution.
In section 3, the semantics of inference from
evaluable KBs is specified by constrained 3-valued
models. The third value represents undefined truth
values of partial predicates and atoms with undefined
arguments. In section 4, KB inference without RAA
is characterized by sequent calculi whose logical rules
are the negation rules and whose structural rules
exclude the contraction rule (Szabo, 1969). KB facts
and rules are non-logical axioms of these calculi.
2 PRELIMINARIES
We consider inference of sets of literals, which are
called goal lists, from evaluable KBs. We assume that
evaluable functions are defined by means that are ex-
ternal to KB rules. These functions could be defined
as recursive functions in a functional programming
language or as algorithms in a procedural program-
ming language. The same assumption applies to
evaluable predicates. The latter are boolean functions.
Combining logic and functional programming has
been extensively investigated (Hanus, 2013; Casas et al.,
2006; Rodr
´
ıguez-Hortal
´
a and S
´
anchez-Hern
´
andez, 2008).
It is possible to use other systems for speci-
fying evaluable functions/predicates. For example,
database query results can be interpreted as function
values or predicate truth values. Function/predicate
arguments are parameters of the queries. Some results
could be interpreted as undefined values. Evaluable
predicates could also be specified by a neural network
(Dong et al., 2019; Serafini and d’Avila Garcez,
2016), which is used to approximate the truth values
of atoms of these predicates with constant arguments.
Truth values are usually approximated by real num-
bers from interval [0, 1]. Numbers close to 1 are
equated with true, numbers close to 0 are equated with
false. Otherwise, the truth value is undefined.
Let us recall some definitions which will be used
later. A KB is called consistent if no atom is a fact
or is derivable from this KB, along with its negation
being derivable or a fact. A literal is called ground if
it does not contain variables. A substitution is a finite
set of mappings of variables to terms. The result of
applying a substitution to a formula or set of formulas
is called its instance.
Non-evaluable predicates in non-Horn KBs
should be considered partial by default. First, rule
sets are often incomplete by design. As a result, some
atoms with constant arguments are not derivable in
FOL or another formal system from KB rules and
facts, and their negations are not derivable either.
Second, some non-evaluable predicates are inherently
partial from the proof-theoretical perspective.
It is well-known that any partial recursive function
can be represented by Horn facts and rules (Voronkov,
1995). Consider function f whose domain is recur-
sively enumerable but not recursive (not decidable)
(Rogers, 1967) and whose range is a known finite
set. Let non-evaluable predicate P represent f , that is
P(x, y) is derivable if and only if f (x) = y. If P were
total, then the derivation of either P(a, b) or ¬P(a, b)
would succeed for any constants a, b provided that
the inference procedure is complete. This contradicts
our assumption about the domain of f , and thus, P is
partial.
Let −A denote the complement of A, i.e. it is the
negation of atom A, and the atom of negative literal
A. This notation will also be used for sets of literals.
As usual, it is assumed that premises of inference
rules have disjoint variables. For any KB rule A ⇐
A
1
∧ ... ∧ A
k
, the following implications −A
i
⇐ −A ∧
A
1
∧ ...A
i−1
∧ A
i+1
... ∧ A
n
are called contrapositives
to this rule (Stickel, 1992). KB rules entail their
contrapositives in classical FOL.
Both forward and backward chaining are based
on Generalized Modus Ponens (GMP) as the sole
inference rule (Russell and Norvig, 2009). If A
0
1
...A
0
k
are derived literals, A ⇐ A
1
∧ ... ∧ A
k
is a KB rule,
substitution θ is a unifier of literals A
0
i
and A
i
, i.e.
A
0
i
θ = A
i
θ, for i = 1...k, then Aθ is derivable. For
facts, k = 0. A forward chaining step derives Aθ
given that A
0
1
, ..., A
0
k
are derived literals. Given goal
list L = {...G...} and such substitution θ that Gθ = Aθ,
every step of backward chaining replaces goal G with
A
1
θ, ..., A
k
θ in L and also applies θ to the other goals
in L.
Forward and backward chaining have the same
inference power. Forward and backward chaining
are usually considered in the context of Horn KBs
but they are also applicable to non-Horn KBs. For
non-Horn KBs, chaining methods can be extended
ICAART 2022 - 14th International Conference on Agents and Artificial Intelligence
682

by adding contrapositives to KB rules. Forward
and backward chaining with both rules and contra-
positives is called extended chaining. As shown in
(Sakharov, 2021), extended chaining gives a proce-
dural characterization of KB inference without RAA.
Because of this, extended chaining will be used to
assess models and calculi for such inference.
3 CONSTRAINED 3-VALUED
MODELS
Determining whether a recursive function is partial or
total is undecidable. The problem of finding whether
a recursive function yields a definite value (i.e true
or false) for given arguments is also undecidable
(Rogers, 1967). Hence, it is not possible to effectively
separate ground atoms with definite truth values
from the rest. As it was first noted by Kleene
(Kleene, 1952), classical 2-valued FOL models are
not adequate in the presence of evaluable functions
and predicates. Models for evaluable KBs should
have at least three values: −1 (false), 1 (true), 0
(undefined). These values are assigned to all ground
atoms. The meaning of 0 is that the atom predicate or
a computable function occurring in the atom does not
yield a value. Fortunately, three values are sufficient
for models that are relevant to KB inference without
reasoning by contradiction.
Multi-valued models are usually defined by truth
tables (or functions) for logical connectives so that
the truth values of ground formulas can be calculated.
Let |A| denote the truth value of ground literal A.
The following equation defines the truth values for
negation: |¬A| = −|A|. This definition complies with
classical 2-valued FOL models for definite values
and gives a reasonable choice for the negation of
atoms whose truth values are undefined. This is how
negation truth values are defined in natural 3-valued
logics (Avron, 1991). Inference for multi-valued
logics is usually specified via nonstandard proof
theories (Gottwald, 2001; Takano, 1998; Malinowski,
2014).
No other formulas than literals are produced
during KB derivations. Because of this, legitimate
models for KB inference without RAA can be de-
fined by the above negation truth function and by
constraints on truth values in ground instances of
facts and rules as opposed to truth tables for other
connectives.
Definition 1. An assignment of truth values −1, 0, 1
to ground literals is a M
3
model if |¬A| = −|A| for
any ground atom A and the following constraints are
satisfied:
1. A is a ground fact instance: |A| = 1
2. A
0
⇐ A
1
∧ ... ∧ A
k
is a ground rule instance: If
|A
i
| = 1 for i = 1...k, then |A
0
| = 1.
3. A
0
⇐ A
1
∧ ... ∧ A
k
is a ground rule instance: If
|A
0
| = −1 and |A
i
| = 1 for i = 1... j − 1 and i =
j +1...k, then |A
j
| = −1.
Literal A is valid regarding M
3
if |A
0
| = 1 for all
its groundings A
0
in all M
3
models.
Theorem 1. Extended chaining is sound and com-
plete with respect to M
3
models for consistent KBs.
Proof. Soundness. Consider an arbitrary grounding
of an extended forward chaining derivation. If ground
literal B is the conclusion of GMP and ground literals
B
1
, ..., B
k
along with rule or contrapositive B
0
⇐ B
0
1
∧
... ∧ B
0
k
are the premises, then the model constraints
guarantee that |B| = 1 provided that |B
i
| = 1 for
i = 1...k. By induction on the depth of derivations,
extended forward chaining is sound with respect to
M
3
models.
Completeness. Suppose A is a ground literal,
|A| = 1 in all M
3
models, and A is not derivable by
chaining from KB facts and rules. Let us look at
model M in which |B| = 1 for every ground literal B
that is derivable by chaining from KB facts and rules,
|C| = −1 for every such ground literal C that −C is
derivable, and |D| = 0 for every other ground literal
D. Such model M exists for any consistent KB, and
|A| 6= 1 in M.
Constraint 1 holds for M because ground instances
of a facts are derivable. If constraint 2 is violated for
ground rule instance A
0
⇐ A
1
∧... ∧A
k
, then all A
i
for
i = 1...k are derivable by forward chaining. Hence,
A
0
is derivable from the latter by applying GMP.
If constraint 3 is violated for ground rule instance
A
0
⇐ A
1
∧ ...A
j
... ∧ A
k
, then all A
i
for i = 1... j − 1
and i = j + 1...k are derivable by forward chaining.
−A
0
is also derivable. Hence, −A
j
is derivable
from −A
0
, A
1
, ...A
j−1
, A
j+1
, ..., A
k
by applying GMP
to contrapositive −A
j
⇐ −A
0
∧A
1
∧...A
j−1
∧A
j+1
... ∧
A
n
. Therefore M is a M
3
model and the assumption
about A cannot be true.
Theorem 1 shows that M
3
models characterize
KB inference without RAA. M
3
models differ from
Kleene and Lukasiewicz 3-valued logics (Avron,
1991). |P| = 1 in all Kleene models for which rules
P ⇐ Q and P ⇐ ¬Q are true. |P| could be 0 in the
respective M
3
models. |¬R| = 1 in all Lukasiewicz
models for which fact ¬S and rules S ⇐ P ∧R, P ⇐ Q,
P ⇐ ¬Q are true. |¬R| could be 0 in the respective M
3
models.
A Logical Characterization of Evaluable Knowledge Bases
683

4 SEQUENT CALCULI
We rely on sequent calculi with non-logical axioms as
an instrument for logical characterization of inference
from evaluable KBs. A sequent is Γ ` Π where Γ is an
antecedent and Π is a succedent. Consider a variant
of Gentzen’s LK (Szabo, 1969) in which antecedents
and succedents are multisets of formulas instead of
sequences. This variant does not require the exchange
rule. LK has one logical axiom A ` A. Let us exclude
the contraction rule. The remaining structural rules
are cut and weakening:
Γ ` A, ∆ A, Π ` Σ
Γ, Π ` ∆, Σ
cut
Γ ` Π
A, Γ ` Π
LW
Γ ` Π
Γ ` A, Π
RW
KB inference and logic programming are con-
cerned about the derivation of literals, i.e. sequents
of the form ` A in the terminology of sequent calculi,
A is a literal. KB facts and rules can be treated
as non-logical axioms (Negri and Von Plato, 2001).
Sequents of the form ` A represent facts, and rules
are represented by sequents of the form A
1
, ..., A
n
`
A where A, A
1
, ..., A
n
are literals. Variables can be
replaced by terms in instances of these axioms.
Standard cut elimination techniques can be ap-
plied even in the presence of non-logical axioms
(Negri and Von Plato, 2001). Cut instances are moved
upward so that one premise of every cut is a non-
logical axiom. Therefore, the subformula property
(Negri and Von Plato, 2001) holds for formulas of the
forms A ∧ B, A ∨ B, A ⇒ B, ∀xA(x), ∃xA(x), and ¬C
where C is not an atom. Hence, the logical rules for
connectives ∧, ∨, ⇒ and for quantifiers are admissible
in derivations of literals and so are all formulas except
for literals. The only logical rules that are necessary
in these derivations are two negation rules:
Γ ` A, Π
¬A, Γ ` Π
L¬
A, Γ ` Π
Γ ` ¬A, Π
R¬
Definition 2. LK
−c
is the set of sequent calculi in
which formulas are literals, the structural rules are
cut, LW, RW , the logical rules are L¬, R¬ restricted
to atoms as the principal formulas, the logical axiom
is A ` A, and non-logical axioms represent KB rules
and facts.
Theorem 2. Any LK
−c
derivation of a literal can be
transformed into an extended chaining derivation and
vice versa provided that the KB is consistent.
Proof. Consider a backward chaining derivation. Ev-
ery step of this derivation involving a KB rule is
replaced with the cut rule whose premises are the
respective goal list and rule instance. Any contrapos-
itive instance −A
i
⇐ −A ∧ A
1
∧ ...A
i−1
∧ A
i+1
... ∧ A
n
is obtained from rule instance A ⇐ A
1
∧ ... ∧ A
k
by
applying two negation rules if A and A
i
are atoms. If
A is negative, then cut is applied to the rule instance
and A, −A `. If A
i
is negative, then cut is applied to the
rule instance and ` A
i
, −A
i
. Every step of the chaining
derivation involving a contrapositive is replaced with
the cut rule applied to the respective goal list and
contrapositive instance. Thus the entire chaining
derivation is transformed into a LK
−c
derivation.
Consider a LK
−c
derivation with the endsequent
` G. The applications of the weakening rules can
be moved below all other rules. We present relevant
permutations for the cases that the weakening formula
is principal in the following cut or negation rule. The
other cases are even simpler and left to the reader.
Γ ` A, Π
∆ ` Σ
A, ∆ ` Σ
Γ, ∆ ` Π, Σ →
∆ ` Σ
. . .
Γ, ∆ ` Π, Σ
Γ ` Π
Γ ` A, Π A, ∆ ` Σ
Γ, ∆ ` Π, Σ →
Γ ` Π
. . .
Γ, ∆ ` Π, Σ
Γ ` Π
Γ ` Π, A
Γ, ¬A ` Π →
Γ ` Π
Γ, ¬A ` Π
Γ ` Π
A, Γ ` Π
Γ ` ¬A, Π →
Γ ` Π
Γ ` ¬A, Π
Weakening cannot be the last rule in a derivation
of ` G because sequent ` is not derivable from
consistent KBs. Hence, weakening can be eliminated
from derivations of sequents like this. Now let
us move negation rules upward. The following
permutations accomplish this.
Γ ` A, ∆ A, Π ` B, Σ
Γ, Π ` B, ∆, Σ
¬B, Γ, Π ` ∆, Σ →
Γ ` A, ∆
A, Π ` B, Σ
A, ¬B, Π ` Σ
¬B, Γ, Π ` ∆, Σ
Γ ` A, ∆, B A, Π ` Σ
Γ, Π ` B, ∆, Σ
¬B, Γ, Π ` ∆, Σ →
Γ ` A, ∆, B
¬B, Γ ` A, ∆ A, Π ` Σ
¬B, Γ, Π ` ∆, Σ
ICAART 2022 - 14th International Conference on Agents and Artificial Intelligence
684

Γ ` A, ∆ A, Π, B ` Σ
Γ, Π, B ` ∆, Σ
Γ, Π ` ¬B, ∆, Σ →
Γ ` A, ∆
A, Π, B ` Σ
A, Π ` ¬B, Σ
Γ, Π ` ¬B, ∆, Σ
B, Γ ` A, ∆ A, Π ` Σ
B, Γ, Π ` ∆, Σ
Γ, Π ` ¬B, ∆, Σ →
A, Π ` Σ
Π ` ¬A, Σ
B, Γ ` A, ∆
¬A, B, Γ ` ∆
¬A, Γ ` ¬B, ∆
Γ, Π ` ¬B, ∆, Σ
By repeatedly applying these transformations, all
applications of the negation rules can be moved above
all applications of cut.
If the right premise of cut contains G occurring in
the endsequent and the left premise is the conclusion
of another cut, then the following transformation is
applied from top to bottom. If the left premise
contains this G and the right premise is a conclusion
of another cut, then this transformation is applied in
the opposite direction.
Γ ` A, ∆ A, Π ` B, Σ
Γ, Π ` B, ∆, Σ B, Φ ` Ψ
Γ, Π, Φ ` ∆, Σ, Ψ
l
Γ ` A, ∆
A, Π ` B, Σ B, Φ ` Ψ
A, Π, Φ ` Σ, Ψ
Γ, Π, Φ ` ∆, Σ, Ψ
Every such transformation reduces the total number
of instances of cut not containing G. By repeatedly
applying this transformation, we can receive such
derivation that every sequent not containing this G
is a KB fact instance, a KB rule instance, A ` A, or
the endsequent of a derivation comprised of negation
rules applied to the above.
Let us construct a backward chaining derivation
from the transformed sequent derivation. The endse-
quent of any derivation comprised of negation rules
with top sequent A ` A has one of these forms: A `
A; ¬A ` ¬A; ` A, ¬A; A, ¬A `. Applications of
cut to A ` A and ¬A ` ¬A can be removed from
any derivation. The endsequent of any derivation
comprised of negation rules with a KB fact instance
or a KB rule instance as the top sequent will be
called KB sequent. KB sequents are rearrangements
of fact/rule literals: a literal in the succedent becomes
its complement in the antecedent and vice versa.
If G occurs in the right premise of cut and the left
premise is a KB sequent, then (1) is interpreted as a
backward chaining step using the fact ` A, the source
rule of Γ, −∆ ` A, or a contrapositive for goal list
A, Π, −Σ. If G occurs in the left premise of cut and the
right premise is a KB sequent, then (2) is interpreted
as a backward chaining step using the fact ` −A, the
source rule of Π, −Σ ` −A or a contrapositive for goal
list −A, Γ, −∆. In both cases, if A or −A is a fact
or rule head, then a KB rule is used. Otherwise, a
contrapositive is used.
Γ ` A, ∆ A, Π ` G, Σ
Γ, Π ` G, ∆, Σ
(1)
Γ ` A, G, ∆ A, Π ` Σ
Γ, Π ` G, ∆, Σ
(2)
If G occurs in the right premise of cut, the left premise
could also be ` A, −A. If G occurs in the left premise
of cut, the right premise could also be A, −A `.
` A, −A A, Π ` G, Σ
Π ` −A, G, Σ
(3)
Π ` G, A, Σ A, −A `
−A, Π ` G, Σ
(4)
Let us look at the topmost premise of cut contain-
ing G. It could be one of these sequents: G ` G; ¬G `
¬G; ` G, ¬G; G, ¬G `. In this case, the other premise
of this cut specifies the rule of the first step of the
backward chaining derivation. The topmost premise
of cut containing G could also be a rearrangement
of literals of a KB rule. In this case, this rule or its
contrapositive is used in the first step of the backward
chaining derivation.
Consider the sequence of cut instances in the top-
down order. The premises that are not descendants
of the topmost sequent with G in cut instances of
the forms (1) and (2) specify the sequence of facts
or rules in all the following backward chaining steps.
The premises containing G specify goal lists. Cut
instances of the forms (3) and (4) simply rearrange
literals in sequents, they are skipped. The entire
sequence maps to an extended chaining derivation of
G.
This proof shows that weakening is admissible
in LK
−c
for consistent KBs. Theorem 2 guarantees
that LK
−c
constitute a proof theory of KB inference
without RAA. The constraints of M
3
can be reformu-
lated for non-logical axioms of LK
−c
instead of KB
rules and facts. The following corollary is basically a
combination of Theorem 1 and Theorem 2.
Corollary 1. Derivation of literals in LK
−c
is sound
and complete with respect to M
3
models for consis-
tent KBs.
A Logical Characterization of Evaluable Knowledge Bases
685

Clearly, Horn KBs can be characterized by LK
−c
without the negation rules (cf. simple logic (Besnard
and Hunter, 2018)). As shown earlier, logical
rules other than the negation rules are admissible
in derivations of literals in LK instances with non-
logical axioms expressing KB rules and facts. Conse-
quently, non-Horn KBs can be characterized by LK
−c
extended with the contraction rule. All these calculi
are more lucid than calculi for FOL in its entirety.
5 RELATED WORK
Logics with more limited capabilities than FOL
are relevant to KB inference. Description logics
(Russell and Norvig, 2009) are one example of that.
Logics with limited inference capabilities also play
an important role in argumentation. Simple logic
has one inference rule - Modus Ponens - and no
logical axioms (Besnard and Hunter, 2018). Simple
logic characterizes Horn KBs. Direct derivations are
defined in (Kakas et al., 2018) as natural deduction
done without using RAA. These direct derivations are
not limited to literals.
In addition to the FOL interpretation of KB rules,
the sets of rules with the same predicate in their
heads can also be treated as inductive definitions
(Moschovakis, 1974). Such rules in FO(FD) (Hou
et al., 2010) rely on the notation of FOL. The paper
(Hou et al., 2010) considers inference in FO(FD) for
finite domains only. Other formalizations of inductive
definitions as extensions of FOL are investigated in
(Brotherston and Simpson, 2010) and (Cohen and
Rowe, 2018). Inference in these systems is more
complicated than inference in FOL, it includes cyclic
proofs.
The semantics of normal logic programs are based
on the negation-as-failure paradigm (Denecker et al.,
2017). The most researched among them are stable
and well-founded semantics (Schlipf, 1995). The se-
mantics of normal logic programs are closely related
to the semantics of inductive definitions (Denecker
and Vennekens, 2014). In comparison to FOL,
inference procedures for stable and well-founded
semantics (Shen et al., 2002; Yamasaki and Kurose,
2001) are more complicated. They have not been
investigated as much as inference procedures for FOL
and its fragments. Also, 3-valued models have been
used for describing the semantics of logic programs
(H
¨
olldobler and Ramli, 2009; Naish, 2006).
Sequent calculi without contraction have been
extensively investigated (Grishin, 1981; Dyckhoff,
1992; Ono, 2010). Sometimes they are referred to
as affine logic. Direct predicate logic (Ketonen and
Weyhrauch, 1984) is LK without contraction, the cut
rule is admissible in this logic. LK without contrac-
tion is equivalent to its single-succedent version LJ
without contraction and with the axiom of double
negation (Ono, 2010). The latter is also known as
stable logic (Negri and Von Plato, 2001). Models
for LK without contraction are studied in (Ono, 2010;
Grishin, 1981).
The difference of LK
−c
is that its instances contain
non-logical axioms, and the cut rule is essential.
Usually, cut elimination is a central part of any
investigation of sequent calculi. Non-logical axioms
are an obstacle to the admissability of cut but cut
can be pushed up to them. Non-logical axioms
naturally represent KB rules and facts in sequent
calculi. Properties of sequent calculi with non-logical
axioms, including those in the form of so-called math-
ematical basic sequents, are investigated in (Negri
and Von Plato, 2001). Axioms corresponding to KB
rules/facts can be transformed to these sequents.
Sequent calculus derivations for Horn and heredi-
tary Harrop formulas are researched in (Miller et al.,
1991). A sequent calculus from (Harland et al.,
2000) combines forward and backward chaining in
linear logic. Forward and backward chaining are
characterized by means of the focusing calculi in
(Chaudhuri et al., 2008). The focusing calculi are
used for specifying nonstandard logics such as linear
intuitionistic logic. The inverse method can be
adapted to these calculi. The paper (Chaudhuri et al.,
2008) describes how to model forward and backward
chaining in the focused inverse method.
Studies of proof systems for Lukasiewicz logic
and substructural logics usually employ nonstandard
logical connectives (Je
ˇ
r
´
abek, 2010; Girard, 1987).
Our research is concerned about standard logical
connectives as implication, negation, and conjunction
occur in KB rules. Standard connectives are intuitive
and well understood by software developers. In au-
thor’s opinion, using nonstandard connectives in KBs
that are developed and debugged by non-logicians is
a recipe for problems, and those connectives should
be avoided in KB rules and facts.
6 CONCLUSION
Evaluable KBs retain the expressiveness of non-
Horn KBs and naturally integrate reasoning and
computation including neural computing. Reasoning
by contradiction is rather incompatible with evaluable
KBs. Non-Horn KBs are representable in a language
that is much simpler than the language of FOL.
ICAART 2022 - 14th International Conference on Agents and Artificial Intelligence
686

The constrained 3-valued models specifying the
semantics of evaluable KBs are quite straightforward
since the constraints are projections of KB facts and
rules. The simplicity of these models is due to
the fact that the truth values are defined for literals
only. Substructural sequent calculi LK
−c
representing
a proof theory for evaluable KBs are comprehensible
due to few logical rules and to the mapping of KB
rules/facts to non-logical axioms. LK
−c
fall into the
category of well-studied formal systems. It is fair to
say that sequent calculi without the contraction rule
can be viewed as proof theories of inference without
RAA.
Intuitionists criticized classical logic because the
law of excluded middle was abstracted from fi-
nite situations and extended without justification to
statements about infinite collections (Brouwer and
Heyting, 1975). The same argument is valid against
RAA in application to evaluable KBs. Intuitionistic
logic deals with this issue by redefining the semantics
of implication and negation as the definition of Kripke
models shows (Mints, 2000). KB inference without
RAA addresses this issue by interpreting predicates
as partial boolean functions while employing the
classical semantics of logical connectives as regards
to definite truth values.
REFERENCES
Avron, A. (1991). Natural 3-valued logics–characterization
and proof theory. The Journal of Symbolic Logic,
56(1):276–294.
Besnard, P. and Hunter, A. (2018). A review of argumen-
tation based on deductive arguments. Handbook of
Formal Argumentation, pages 437–484.
Brotherston, J. and Simpson, A. (2010). Sequent calculi for
induction and infinite descent. Journal of Logic and
Computation, 21(6):1177–1216.
Brouwer, L. and Heyting, A. (1975). L.E.J. Brouwer
Collected Works: Philosophy and foundations of
mathematics. 1. North-Holland Publishing Company.
Casas, A., Cabeza, D., and Hermenegildo, M. V. (2006). A
syntactic approach to combining functional notation,
lazy evaluation, and higher-order in LP systems. In
International Symposium on Functional and Logic
Programming, pages 146–162. Springer.
Chaudhuri, K., Pfenning, F., and Price, G. (2008). A logical
characterization of forward and backward chaining in
the inverse method. Journal of Automated Reasoning,
40(2-3):133–177.
Cohen, L. and Rowe, R. N. S. (2018). Uniform induc-
tive reasoning in transitive closure logic via infinite
descent. In 27th EACSL Annual Conference on
Computer Science Logic, volume 119, pages 17:1–
17:16. LIPICS.
Denecker, M., Truszczynski, M., and Vennekens, J. (2017).
About negation-as-failure and the informal semantics
of logic programming. Association for Logic Pro-
gramming.
Denecker, M. and Vennekens, J. (2014). The well-founded
semantics is the principle of inductive definition,
revisited. In 14th International Conference on the
Principles of Knowledge Representation and Reason-
ing. AAAI.
Dong, H., Mao, J., Lin, T., Wang, C., Li, L., and Zhou,
D. (2019). Neural logic machines. In International
Conference on Learning Representations.
Dyckhoff, R. (1992). Contraction-free sequent calculi for
intuitionistic logic. The Journal of Symbolic Logic,
57(3):795–807.
Girard, J.-Y. (1987). Linear logic. Theoretical computer
science, 50(1):1–101.
Gottwald, S. (2001). A treatise on many-valued logics.
Research Studies Press.
Grishin, V. N. (1981). Predicate and set-theoretic cal-
culi based on logic without contractions. Izvestiya
Rossiiskoi Akademii Nauk. Seriya Matematicheskaya,
45(1):47–68.
Hanus, M. (2013). Functional logic programming: From
theory to Curry. In Programming Logics, pages 123–
168. Springer.
Harland, J. A., Pym, D. J., and Winikoff, M. (2000).
Forward and backward chaining in linear logic. Elec-
tronic Notes in Theoretical Computer Science, 37:1–
16.
H
¨
olldobler, S. and Ramli, C. D. P. K. (2009). Logic
programs under three-valued Łukasiewicz semantics.
In International Conference on Logic Programming,
pages 464–478. Springer.
Hou, P., De Cat, B., and Denecker, M. (2010). FO (FD):
Extending classical logic with rule-based fixpoint def-
initions. Theory and Practice of Logic Programming,
10(4-6):581–596.
Je
ˇ
r
´
abek, E. (2010). Admissible rules of Łukasiewicz logic.
Journal of Logic and Computation, 20(2):425–447.
Kakas, A. C., Mancarella, P., and Toni, F. (2018). On
argumentation logic and propositional logic. Studia
Logica, 106(2):237–279.
Ketonen, J. and Weyhrauch, R. (1984). A decidable
fragment of predicate calculus. Theoretical Computer
Science, 32(3):297–307.
Kleene, S. C. (1952). Introduction to metamathematics. Van
Nostrand.
Malinowski, G. (2014). Kleene logic and inference.
Bulletin of the Section of Logic, 43(1/2):43–52.
McCune, W. (2003). Otter 3.3 reference manual and guide.
Technical report, Argonne National Lab.
Miller, D., Nadathur, G., Pfenning, F., and Scedrov, A.
(1991). Uniform proofs as a foundation for logic
programming. Annals of Pure and Applied logic,
51(1-2):125–157.
Mints, G. (2000). A Short Introduction to Intuitionistic
Logic. Kluwer Academic Publishers.
A Logical Characterization of Evaluable Knowledge Bases
687

Moschovakis, Y. N. (1974). Elementary induction on
abstract structures. North Holland Publiushing Co.
Naish, L. (2006). A three-valued semantics for logic pro-
grammers. Theory Pract. Log. Program., 6(5):509–
538.
Negri, S. and Von Plato, J. (2001). Structural proof theory.
Cambridge University Press.
Ono, H. (2010). Logics without the contraction rule and
residuated lattices. Australasian Journal of Logic.
Rodr
´
ıguez-Hortal
´
a, J. and S
´
anchez-Hern
´
andez, J. (2008).
Functions and lazy evaluation in Prolog. Electronic
Notes in Theoretical Computer Science, 206:153–174.
Rogers, H. (1967). Theory of recursive functions and
effective computability. McGraw-Hill Book Co.
Russell, S. and Norvig, P. (2009). Artificial Intelligence: A
Modern Approach. Prentice Hall Press, 3rd edition.
Sakharov, A. (2021). Inference methods for evaluable
knowledge bases. In Software Engineering Applica-
tion in Informatics, Lecture Notes in Networks and
Systems, pages 499–510. Springer.
Schlipf, J. S. (1995). The expressive powers of the
logic programming semantics. J. Comput. Syst. Sci.,
51(1):64–86.
Serafini, L. and d’Avila Garcez, A. S. (2016). Logic tensor
networks: Deep learning and logical reasoning from
data and knowledge. In 11th International Workshop
on Neural-Symbolic Learning and Reasoning, volume
1768. CEUR-WS.org.
Shen, Y.-D., Yuan, L.-Y., and You, J.-H. (2002). SLT-
resolution for the well-founded semantics. Journal of
Automated Reasoning, 28(1):53–97.
Stickel, M. E. (1992). A Prolog technology theorem
prover: a new exposition and implementation in
Prolog. Theoretical Computer Science, 104(1):109–
128.
Szabo, M. E., editor (1969). The collected papers of
Gerhard Gentzen. North-Holland.
Takano, M. (1998). Sequent calculi for three-valued logics.
Tsukuba Journal of Mathematics, 22(2):447–461.
Voronkov, A. (1995). On computability by logic programs.
Annals of Mathematics and Artificial Intelligence,
15(3-4):437–456.
Yamasaki, S. and Kurose, Y. (2001). A sound and
complete procedure for a general logic program in
non-floundering derivations with respect to the 3-
valued stable model semantics. Theoretical Computer
Science, 266(1-2):489–512.
ICAART 2022 - 14th International Conference on Agents and Artificial Intelligence
688