A Formal Approach To Problem Solving

A Formal Approach to Problem Solving.⋆
Diderik Batens
Centre for Logic and Philosophy of Science
Universiteit Gent, Belgium
Diderik.Batens@UGent.be
1 Aim of this paper
For more than forty years, problem solving is recognized as central for the under-
standing of reasoning in the sciences. In this paper a formal approach to problem
solving is spelled out.
The essential aim is to devise a formal procedure that can function as an
explication of a problem-solving process—Sects. 2–5. Touchy topics in this con-
nection are (i) the relation to the relevant philosophy of science literature and
(ii) the role of erotetic logic in the approach. For (i) I shall rely on the approach
developed by Thomas Nickles ([19], [20], [21], [22]), on the model developed by
myself ([1], [2] and [3]) and on the work of Joke Meheus ([14], [15], [16], [17],
[18]). A particular difficulty concerns the function of constraints in a formal ap-
proach to problem solving. The relation can be shown to be sensible but cannot
be discussed here. As for (ii), previous work, for example [10], suggests that the
required erotetic logic differs in central respects from the ‘standard approach’
([24], [25], [26]). However, it can be shown that all derivations of (sets of) ques-
tions in terms of instruction DP (Sect. 2) are derivable (possibly in several steps)
according to the standard approach.
A problem-solving process is clearly different from a proof—for one thing,
problem-solving processes may be unsuccessful. As “proof” is used to denote
both the explicanda in natural language and their explicata in a formal language,
I shall use “problem-solving process”—psp for short—to denote the explicata as
well as the explicanda.
Two features seem essential to psps. First, some of their parts may be justified
at a stage of the process, but may turn out dead ends at a later stage. The
formal approach to this feature is familiar from the dynamic proofs of adaptive
logics—see for example [4] or [5]. Next, realistic psps require ‘prospective’ steps:
if A1, . . . , An are true, the solution of problem P is B. Such considerations enable
one to proceed in a goal-directed way: to tackle the problem P by first solving
the sequence of problems “Is A1 true?”, . . . , “Is An true?”. The formal approach
to this feature is provided by the prospective proofs studied in [11], [6] and other
⋆
This abstract is based on technical report [7] on which I received interesting com-
ments from Kristof De Clercq, Andrzej Wiśniewski and others. The research was
supported by the Fund for Scientific Research – Flanders, by the Royal Flemish
Academy of Belgium for Science and the Arts, and indirectly by the Flemish Minis-
ter responsible for Science and Technology (contract BIL01/80).

papers. Such prospective steps are sometimes deemed “proof heuristics”, for
example by Hintikka in [13], and thus expelled from the realm of logic proper. In
my view this is a serious mistake, but I cannot argue for this view in the present
context.
Given the generality of the notion of problem solving, a psp should be able to
incorporate most logics, and especially most adaptive logics developed so far. I
can only provide some hints in this respect for the reader unfamiliar with adap-
tive logics. There is no positive test1
for many actual reasoning processes—see
[5] for a host of examples. Hence, those reasoning processes cannot be explicated
by proofs of the traditional sort—(of which any initial fragment is itself a proof).
Adaptive logics circumvent this limitation by having dynamic proofs in which
a formula (a line in an annotated proof) may be introduced at some stage of a
proof but may be marked (and hence considered as not derived) at a later stage.
By doing so, dynamic proofs enable one to explicate reasoning processes that
range from interpreting a set of premises as consistently as possible to deriving
inductive generalizations or inductive predictions.
To keep this paper within bounds, I shall restrict psps to a special case:
solving the problem whether A or ∼A is derivable by CL (Classical Logic) from
(a consistent subset of) the premises (Sects. 2–3), possibly extended with the
answers to questions that may be directly answered by standard means (Sect. 4).
This allows, for example, for the explication of predictive problems from data
and theories—a problem falling beyond the scope of the present paper but easily
solved in view of the sections on background knowledge from [9]. The way in
which the restriction may be removed is suggested in Sect. 5.
My approach is computational in that it defines a problem-solving procedure,
which is intended to explicate (a specific form of) human reasoning.2
I shall not
present a completely deterministic version of it. To do so would take away the
attention from the basic mechanisms. Moreover, the efficiency of the possible
specifications of the procedure has not been sufficiently studied to obtain a final
judgement on the matter.
2 The Elements of a Psp
A stage of a psp is a sequence of lines; a psp is a chain of stages. The first
stage consists of a single line. Where Si and Si+1 are subsequent stages, Si+1 is
obtained by adding one line to Si and possibly changing the marks—see below—
of the lines of Si+1. The instructions determine which lines may be added to Si
in view of Si, viz. in view of the formulas that occur in unmarked lines of Si.
The instructions obviously depend on the underlying logic or logics.3
1
See [12] on decidability, positive test, and other computational matters.
2
Hence results on logic programming are at best indirectly relevant.
3
To save some space, I describe the rules as part of the instructions. A more general
and more systematic outlook considers the rules as given and sees the instructions
as specifying whether a rule may or should be applied in view of the stage of the
psp.

Two kinds of lines will occur in a psp: problem lines and declarative lines.
In the narrow sense, a problem is a non-empty set of questions. For the wider
sense I refer to the aforementioned philosophy of science literature. In this paper,
the main problem will be a single yes–no question ?{M, ∼M}, and all derived
problems will be sets of yes–no questions. A and ∼A will be called the direct
answers of the question ?{A, ∼A}.
Γ will always be the premise set. The formula of a declarative line may be
a member of Γ or a formula derived from members of Γ. It may also have the
form
[B1, . . . , Bn] A
which indicates that Γ ∪ {B1, . . . , Bn} ⊢ A. {B1, . . . , Bn} will be called the
condition of A. If A is the formula of a declarative line, this formula is said to
be conditionless—the condition is then also said to be empty.
Given a problem P, the problem solver has to chose a direct answer of a
question of the problem as a target (from P), which is noted on a target line.4
A target is a formula the problem solver tries to establish. There is no objection
against introducing several targets at once. A target A will be written as
[A] A
which is logically redundant but guides the procedure, as we shall see.
As the psp proceeds, some (declarative lines as well as question) lines may
become useless for solving the main problem. Such lines will be marked.
I now turn to the formal machinery. First some preparatory definitions. Let
us distinguish α-formulas from β-formulas (varying on a theme from [23]) and
assign to each formula two other formulas according to the following table:
α α1 α2 β β1 β2
A ∧ B A B ∼(A ∧ B) ∼A ∼B
A ≡ B A ⊃ B B ⊃ A ∼(A ≡ B) ∼(A ⊃ B) ∼(B ⊃ A)
∼(A ∨ B) ∼A ∼B A ∨ B A B
∼(A ⊃ B) A ∼B A ⊃ B ∼A B
∼∼A A A
The positive part relation is defined recursively by the following three clauses:
1. pp(A, A).
2. pp(A, α) if pp(A, α1) or pp(A, α2).
3. pp(A, β) if pp(A, β1) or pp(A, β2).
The instructions These rely on the prospective rules from [11]. Γ is the premise
set. A psp starts with an application of Main, which introduces the main problem
?{M, ∼M}. In all instructions, k denotes a suitable line number.
4
The formal machinery may be defined without target lines, but the latter greatly
clarify the goal-directed character of psps.

Main Start a psp with the line:
1 {?{M, ∼M}} Main
Target If P is the problem of an unmarked problem line, and A is a direct
answer of a member of P, then one may add:
k [A] A Target
Prem If A is an unmarked target,5
B ∈ Γ, and pp(A, B), then one may add:
k B Prem
Let ∗A denote the ‘complement’ of A, viz. B if A has the form ∼B and ∼A
otherwise. The formula analysing rules of CL (see [11]) may be summarized as
follows:6
[∆] α
[∆] α1 [∆] α2
[∆] β
[∆, ∗β2] β1 [∆, ∗β1] β2
The general form of the rules is [∆] A / [∆ ∪ ∆′
] B. The following instruction
handles their application:
FAR If C is an unmarked target, [∆] A is the formula of an unmarked line i,
[∆] A / [∆ ∪ ∆′
] B is a formula analysing rule, and pp(C, B), then one
may add:
k [∆ ∪ ∆′
] B i; R
in which R is the name of the formula analysing rule.
The condition analysing rules of CL are summarized by:
[∆ ∪ {α}] A
[∆ ∪ {α1, α2}] A
[∆ ∪ {β}] A
[∆ ∪ {β1}] A [∆ ∪ {β2}] A
They all have the form [∆ ∪ {B}] A / [∆ ∪ ∆′
] A. The following instruction
refers to them:
CAR If A is an unmarked target, [∆ ∪ {B}] A is the formula of an unmarked
line i, and [∆ ∪ {B}] A / [∆ ∪ ∆′
] A is a condition analysing rule, then
one may add:
k [∆ ∪ ∆′
] A i; R
in which R is the name of the condition analysing rule.
5
Phrased more precisely: A is the formula of an unmarked target line.
6
The left rule states that both [∆] α1 and [∆] α2 may be derived (separately) from
[∆] α (notational abuse here and in the text).

The instructions EM (excluded middle) and EM0 allow one to eliminate
certain problems without answering them. Line i (either line i or line j) will be
R-marked after an application of EM0 (EM)—see below for marking.
EM0 If [∆ ∪ {∗A}] A is the formula of a line i that is neither R-marked nor
I-marked, then one may add:
k [∆] A i; EM0
EM If A is an unmarked target, [∆ ∪ {B}] A and [∆′
∪ {∼B}] A are the
respective formulas of the unmarked or only D-marked lines i and j,
and ∆ ⊆ ∆′
or ∆′
⊆ ∆, then one may add:
k [∆ ∪ ∆′
] A i, j; EM
Transitivity is essential for eliminating solved questions from an (at least
implicit) problem as well as for summarizing the remaining problems (and paths)
in a problem-solving process. If ∆′
⊆ ∆, line i is R-marked after the application
of Trans.
Trans If A is an unmarked target, and [∆∪{B}] A and [∆′
] B are the respective
formulas of the at most S-marked7
lines i and j, then one may add:
k [∆ ∪ ∆′
] A i, j; Trans
The last instruction handles derived problems:
DP If A is an unmarked target from problem line i and [B1, . . . , Bn] A is the
formula of an unmarked line j, then one may add:
k {?{B1, ∼B1}, . . . , ?{Bn, ∼Bn}} i, j; DP
In view of the intended applications (deriving predictions, explanations, etc.)
the procedural system has no instruction for applying EFQ—the paraconsistent
procedural variant logic of CL is studied in [8].
Marking It is necessary to introduce several kinds of marks, which have distinct
effects on the procedure. Each kind is governed by a definition, which applies to
stages of the psp—marks may come and go with each new stage. The fist kind
of mark indicates that a line is redundant.
Definition 1. An at most S-marked declarative line i that has [∆] A as its for-
mula is R-marked at a stage iff, at that stage, [Θ] A is the formula of a line for
some Θ ⊂ ∆.8
Definition 2. An unmarked problem line i is R-marked at a stage iff, at that
stage, a direct answer A of a question of line i is the formula of a line.
7
A line is at most S-marked iff it is not R-marked, not I-marked and not D-marked.
8
A formula A is identified with [∅] A.

The next marking definitions require some terminology. Remember that a
target is a target from a problem line (and from its problem). If A is a target,
any line in which [∆] A is derived for some ∆ 6= {A} will be called a resolution
line (for target A). The line called j in instruction DP will be said to generate
the problem line introduced by DP. A is a direct target from [∆] B iff [∆] B is
the formula of the resolution line that generates problem P, A is a target from P,
and A ∈ ∆—remark that some targets from P are not members of ∆. A target
sequence is a sequence h[∆1
] A1
, . . . , [∆n
] An
i in which every Ai+1
(1 ≤ i < n) is
a direct target from [∆i
] Ai
. A target sequence h[∆1
] A1
, . . . , [∆n
] An
i is grounded
iff A1
is not a direct target from any unmarked [Θ] B derived in the psp. A set
is flatly inconsistent iff it contains A as well as ∼A for some A. I-marked lines
are inoperative: acting on the line is not useful for solving any extant problem.
Definition 3. An at most S-marked target line that has [A] A as its formula is
I-marked at a stage iff every problem line from which A is a target is marked at
that stage.
Definition 4. An at most S-marked resolution line of which [∆1
] A1
is the for-
mula for some ∆1
6= ∅ is I-marked at a stage iff, at that stage, for every grounded
target sequence h[∆n
] An
, . . . , [∆1
] A1
i,
(i) some target [Ai
] Ai
(1 ≤ i ≤ n) is marked, or
(ii) {An
, . . . , A1
} ∩ ∆1
6= ∅, or
(iii) ∆1
∪ . . . ∪ ∆n
∪ Γ◦
s is flatly inconsistent.
Definition 5. An unmarked problem line is I-marked iff no unmarked resolution
line generates it.
Dead end marks indicate that no further action can be taken in view of a
line. A is a dead end iff A is (in the present propositional context) a literal and
A is not a positive part of a premise. If A is not a literal, then CAR leads from
the unmarked [∆ ∪ {A}] B to one or more [∆ ∪ ∆′
] B. The latter is called a
CAR-descendant of [∆ ∪ {A}] B.
Definition 6. An at most S-marked resolution line with formula [∆] A is D-
marked at a stage iff some B ∈ ∆ is a dead end or, at that stage, all CAR-
descendants of [∆] A occur in the psp and are D-marked.
Definition 7. An at most S-marked target line with formula [A] A is D-marked
at a stage iff A is a dead end or no further action can be taken in view of target
A.
Whether an action that can be taken in view of target A is obvious in view
of the instructions. If no further action can be taken in view of target A, all
instructions that could be applied in view of that target were taken and resulted
in lines that are R-marked, I-marked or D-marked.
The proofs from [11] may be easily adapted to show that, for all consistent
Γ, if Γ ⊢ A, then the procedure applied to Γ and {?{A, ∼A}} results in the

answer A, and if Γ 0 A, then the procedure applied to Γ and {?{A, ∼A}} stops
without the main problem being answered or results in the answer ∼A.
The procedure may be sped up (in a way that derives from a straightforward
insight in the psp) by S-marks. Let Γ◦
s be the union of the set of premises and
the set of conditionless formulas that occur at stage s of the psp.
Definition 8. A R-unmarked resolution line in which [∆1
] A1
is derived is S-
marked iff
(i) ∆1
∩ Γ◦
s 6= ∅, or
(ii) for some target sequence h[∆n
] An
, . . . , [∆1
] A1
i, {An
}∪∆1
is flatly incon-
sistent whereas ∆1
is not flatly inconsistent, or
(iii) ∆1 ⊂ ∆n
∪ . . . ∪ ∆2
for some target sequence h[∆n
] An
, . . . , [∆1
] A1
i.
3 An example
I can only offer an extremely simple example, but the reader can easily explore
other examples. Let ?{p ∨ q, ∼(p ∨ q)} be the main problem and let {∼s, ∼u ∨
r, (r ∧ t) ∨ s, (q ∨ u) ⊃ (∼t ∨ q), t ⊃ u} be the premise set. The superscripts of
the marks name the stage at which the line is (thus) marked—this convention
saves rewriting.
1 {?{p ∨ q, ∼(p ∨ q)}} Main
2 [∼(p ∨ q)] ∼(p ∨ q) Target D3
3 [∼p, ∼q)] ∼(p ∨ q) 2; C∼∨E D3
As ∼p is not a positive part of any premise, line 3 is D-marked, and hence so is
line 2.
4 [p ∨ q] p ∨ q Target
5 [p] p ∨ q 4; C∨E D5
6 [q] p ∨ q 4; C∨E
7 {?{q, ∼q}} 4, 6; DP
8 [q] q Target
9 (q ∨ u) ⊃ (∼t ∨ q) Prem
10 [q ∨ u] ∼t ∨ q 9; ⊃E
11 [q] ∼t ∨ q 10; C∨E
12 [q, t] q 11; ∨E I12
13 [u] ∼t ∨ q 10; C∨E
14 [u, t] q 13; ∨E
15 {?{u, ∼u}, ?{t, ∼t}} 8, 14; DP
16 [t] t Target
17 (r ∧ t) ∨ s Prem
18 [∼s] r ∧ t 17; ∨E
19 [∼s] t 18; ∧E S22
20 {?{s, ∼s}} 16, 19; DP R22
21 [∼s] ∼s Target R22
22 ∼s Prem

The S-mark of line 19 indicates that Trans can be applied. I repeat part of the
psp:
14 [u, t] q 13; ∨E S23
15 {?{u, ∼u}, ?{t, ∼t}} 8, 14; DP R23
16 [t] t Target R23
17 (r ∧ t) ∨ s Prem
18 [∼s] r ∧ t 17; ∨E
19 [∼s] t 18; ∧E S22
R23
20 {?{s, ∼s}} 16, 19; DP R22
21 [∼s] ∼s Target R22
22 ∼s Prem
23 t 19, 22; Trans
The S-mark of line 14 again indicates how to proceed. Once line 24 is added to
the psp, line 14 is R-marked (“R24
” is added to it on the present convention).
24 [u] q 14, 23; Trans S29
R30
25 {?{u, ∼u}} 8, 24; DP R29
26 [u] u Target R29
27 t ⊃ u Prem
28 [t] u 27; ⊃E S28
R29
29 u 23, 28; Trans
30 q 24, 29; Trans
Lines 14 and 15 are R-marked at stage 24. The S-mark of line 24 triggers Trans
(line 30). Several marks are added at stage 30, most importantly, line 6 is S-
marked, which leads to
31 p ∨ q 6, 30; Trans
whence the problem is solved (and line 1 is R-marked).
This is not the shortest way to derive an answer to ?{p ∨ q, ∼(p ∨ q)} from
the premise set. However, (most of) the heuristics is pushed into the psp. More-
over, each step of the psp is sensible in view of the previous stage—remark, for
example, ∼u ∨ r, [∼s] r, etc. are not the formula of a line in any psp for this
main problem and premises.
4 Answerable Questions
A straightforward extension of the described procedure introduces, next to the
main problem and the premises, a set A of questions that can be directly an-
swered by standard means, for example questions that are answered (obviously
at a cost) by observational or experimental means. The procedure should decide
when such a question may be ‘launched’, and hence when its answer surfaces in
the psp as a new premise. This simple extension drastically enhances the realistic

character of psps and is a first step away from psps that reduce to proof search
devices for CL.
Whether a question ?{A, ∼A} ∈ A is launched seems to depend on prag-
matic considerations. If the expense of directly answering the main problem,
?{M, ∼M}, is near to nil, to do so may be preferred over engaging in logical
deduction. If the expense is extremely high, one may renounce from seeking a
direct answer, even where this entails that the main problem is left unsolved.
However, such considerations do also affect logical deduction, and hence will be
neglected in the present context.
It should not be required that A be an unmarked target in order for ?{A, ∼A} ∈
A to be launched. Thus a target p may justify that ?{p∧q, ∼(p∧q)} is launched.
If the answer is p∧q, the target is reached. A target p may also justify launching
?{p ∨ q, ∼(p ∨ q)}. The answer p ∨ q leads to [∼q] p, from which p is obtained
if ∼q is reached by deduction or by launching other members of A. Where p
is the target and ?{p ∧ q, ∼(p ∧ q)} is launched, the answer ∼(p ∧ q) is useless
for reaching p, but may be useful for reaching ∼p, which forcibly is a possible
target. All this suggest that a member of A may be launched iff an unmarked
target B is a positive part of A or of ∼A.
A launched question is answered ‘outside of’ the psp, but its answer enters
the psp. It does so with the status of a new premise. As any instruction handling
this would be awkward from a logical point of view, it is advisable to redefine A
as a set of couples: their first element is a question, their second element a direct
answer to this question—in other words, each couple has the form h?{A, ∼A}, Ai
or the form h?{A, ∼A}, ∼Ai.9
That the second element is unknown to the prob-
lem solver until the question ?{A, ∼A} is launched, is captured by the fact that
the instruction than handles the introduction of new premises (second elements
of members of A) depends only on the stage of the psp and the first element of
the member of A. This suggests the following instruction:
New If A is the target of the unmarked target line i, pp(A, B) or pp(A, ∼B),
and h?{B, ∼B}, Ci ∈ A (where C ∈ {B, ∼B}), then one may add:
k C i; New
It is obvious from a line justified by New which member of A was launched
and in view of which target it was launched.
The present extension requires that one redefines: A is a dead end iff A is
(in the present propositional context) a literal and A is not a positive part of
a premise or of a direct answer to the first element of a member of A. This
redefinition has direct effects on I-marking.
Let us consider an example. The main problem is {?{p, ∼p}}, the premise
set contains (q ∧ r) ⊃ p, ∼s ∨ q, s, and h?{q ⊃ r, ∼(q ⊃ r)}, q ⊃ ri ∈ A.
9
I presuppose consistency in the present paper (up to this section). So the second
elements of the members of A should form a consistent set. I do however use a for-
mulation that is sufficiently general to be adjusted to the inconsistent case. Moreover,
even if the union of the set of premises and the set of new premises is inconsistent,
the present psps do not lead to triviality, as may be seen from [8].

1 {?{p, ∼p}} Main R19
2 [p] p Target R19
3 (q ∧ r) ⊃ p Prem
4 [q ∧ r] p 3; ⊃E R19
5 [q, r] p 4; C∧E S9
R10
6 {?{q, ∼q}, ?{r, ∼r}} 2, 5; DP I10
7 [r] r Target I10
8 q ⊃ r 7; New
9 [q] r 8; C⊃E I10
10 [q] p 5, 9; Trans R19
11 {?{q, ∼q}} 2, 10; DP R18
12 [q] q Target R18
13 ∼s ∨ q Prem
14 [s] q 13; ∨E S14
R18
15 {?{s, ∼s}} 12, 14; DP R17
16 [s] s Target R17
17 s Prem
18 q 14, 17; Trans
19 p 10, 18; Trans
The application of Trans in line 10 is triggered by the S-mark of line 5 and
has a rather dramatic effect: five lines are marked at stage 10.
5 Variants and Extensions
Instead of deriving sets of questions by DP, one might allow for, or even restrict
the procedure to, the derivation of singleton problems. Problem lines are then
more easily unmarked—problems are more easily ‘reused’ at later stages.10
Many
other variants of the specific instructions are possible.
More interesting variants are obtained by modifying the underlying logic.
They are not discussed here because all realistic applications require the pred-
icative level. Moreover, many different (combinations of) logics may be required,
depending on the context. Thus CL may be replaced by an inconsistency-
adaptive logic in order to explicate psps (as explicanda) which presuppose that
the premises are interpreted as consistently as possible. Similarly, if CL is re-
placed by an adaptive logic of inductive generalization (see for example [9]), the
psps will be able to answer problems of the form ?{(∀α)A, ∼(∀α)A} in view of
data and (possibly) background knowledge.11
The procedure from Sect. 2 may
also be transformed to a proof-search method for full CL by introducing instruc-
tions for the predicative matters together with an instruction handling the EFQ
rule from [11].
10
Compare this to: according to Definition 3, a target line may be unmarked at stage
s although it was introduced in view of a question line that is marked at stage s.
11
It is obviously determined by the underlying logic (or logics) whether a set of premises
together with the set of directly answerable questions and their answers is sufficient
to answer a specific problem.

6 Some Comments
It can be shown (i) that the procedure leads from a premise set Γ to a solution
of {?{M, ∼M}} (for contingent M) iff there is a consistent Γ′
⊆ Γ such that
Γ′
⊢CL M or Γ′
⊢CL ∼M and (ii) that the procedure stops iff this condition is
not fulfilled. This is exactly what one needs with respect to the intended domain
of application. The extension to the predicate level is as expected. It can also
be shown that there is a specific sense in which each line in a psp (as defined by
the present procedure) is (at the stage at which it is introduced) sensible with
respect to the solution of the main problem.
The procedure is not maximally efficient. The intention was to define a pro-
cedure that (i) can explicate actual problem-solving processes, (ii) avoids steps
that are useless with respect to the main problem, the premises and the directly
answerable questions (see the last paragraph of Sect. 3), and (iii) can be eas-
ily generalized to other logics. Especially adaptive logics (maximal consistent
interpretations, inductive generalization, abduction, . . . ) are interesting in this
respect because they arguably are the underlying logics of actual problem-solving
processes. The adjustment of the procedure to such logics is straightforward, but
fully beyond the scope of the present paper. Apparently any logic can be defined
by a set of prospective rules of inference. These rules are then handled by in-
structions that are determined by the sensible relations between premise sets and
conclusions. Finally, the ideas underlying each kind of mark are independent of
CL and can be adjusted to other logics.
I do not think that the described procedure is even maximally efficient in
its kind. It may possibly be made more efficient by modifying the marking def-
initions or by introducing further kinds of marks. The subject clearly requires
for further research. Moreover, some problem-solving processes (explicanda) may
require logics that are very different from CL. However, such logics are within
reach, as adaptive logics and their dynamic proofs illustrate. In other words,
what is required are different logics. The requirement does not undermine the
present plot, which boils down to this: turning the suitable logics into adequate
proof-search procedures provides the formal road to problem solving.
References
1. Diderik Batens. Meaning, acceptance, and dialectics. In Joseph C. Pitt, editor,
Change and Progress in Modern Science, pages 333–360. Reidel, Dordrecht, 1985.
2. Diderik Batens. Do we need a hierarchical model of science? In John Earman,
editor, Inference, Explanation, and Other Frustrations. Essays in the Philosophy
of Science, pages 199–215. University of California Press, 1992.
3. Diderik Batens. Menselijke kennis. Pleidooi voor een bruikbare rationaliteit.
Garant, Leuven/Apeldoorn, 1992.
4. Diderik Batens. A survey of inconsistency-adaptive logics. In Diderik Batens,
Chris Mortensen, Graham Priest, and Jean Paul Van Bendegem, editors, Frontiers
of Paraconsistent Logic, pages 49–73. Research Studies Press, Baldock, UK, 2000.
5. Diderik Batens. A general characterization of adaptive logics. Logique et Analyse,
173–175:45–68, 2001. Appeared 2003.

6. Diderik Batens. On a partial decision method for dynamic proofs. In Hendrik
Decker, Jørgen Villadsen, and Toshiharu Waragai, editors, PCL 2002. Paracon-
sistent Computational Logic, pages 91–108. (= Datalogiske Skrifter vol. 95), 2002.
Also available as cs.LO/0207090 at http://arxiv.org/archive/cs/intro.html.
7. Diderik Batens. Notes on problem solving. Technical report, 2003.
8. Diderik Batens. A paraconsistent proof procedure based on classical logic. To
appear in the proceedings of WCP3 (see http://www.cle.unicamp.br/wcp3/ for an
abstract).
9. Diderik Batens and Lieven Haesaert. On classical adaptive logics of induction.
Logique et Analyse, 173–175:255–290, 2001. Appeared 2003.
10. Diderik Batens and Joke Meheus. On the logic and pragmatics of
the process of explanation. In Mika Kiikeri and Petri Ylikoski, edi-
tors, Explanatory Connections. Electronic Essays Dedicated to Matti Sintonen.
http://www.valt.helsinki.fi/kfil/matti/, 2001. 22 pp.
11. Diderik Batens and Dagmar Provijn. Pushing the search paths in the proofs. A
study in proof heuristics. Logique et Analyse, 173–175:113–134, 2001. Appeared
2003.
12. George S. Boolos and Richard J. Jeffrey. Computability and Logic. Cambridge
University Press, 1989. (Third edition).
13. Jaakko Hintikka. Inquiry as Inquiry: A Logic of Scientific Discovery. Kluwer,
Dordrecht, 1999.
14. Joke Meheus. Wetenschappelijke ontdekking en creativiteit. Een poging tot theo-
rievorming op basis van een conceptuele, methodologische en logische studie. PhD
thesis, Universiteit Gent (Belgium), 1997. Unpublished PhD thesis.
15. Joke Meheus. Deductive and ampliative adaptive logics as tools in the study of
creativity. Foundations of Science, 4:325–336, 1999.
16. Joke Meheus. Model-based reasoning in creative processes. In Lorenzo Magnani,
Nancy Nersessian, and Paul Thagard, editors, Model-Based Reasoning in Scientific
Discovery, pages 199–217. Kluwer/Plenum, Dordrecht, 1999.
17. Joke Meheus. Analogical reasoning in creative problem solving processes: Logico-
philosophical perspectives. In Fernand Hallyn, editor, Metaphor and Analogy in
the Sciences, pages 17–34. Kluwer, Dordrecht, 2000.
18. Joke Meheus and Diderik Batens. Steering problem solving between cliff incoher-
ence and cliff solitude. Philosophica, 58:153–187, 1996.
19. Thomas Nickles. What is a problem that we may solve it? Synthese, 47:85–118,
1981.
20. Thomas Nickles. Beyond divorce: Current status of the discovery debate. Philos-
ophy of Science, 52:177–206, 1985.
21. Thomas Nickles. Reconstructing science: discovery and experiment. In Diderik
Batens and Jean Paul Van Bendegem, editors, Theory and Experiment. Recent
Insights and New Perspectives on Their Relation, pages 33–54. Reidel, Dordrecht,
1988.
22. Thomas Nickles, editor. Scientific Discovery, Logic, and Rationality. Reidel, Dor-
drecht, 1980.
23. Raymond M. Smullyan. First Order Logic. Dover, New York, 1995. Original
edition: Springer, 1968.
24. Andrzej Wiśniewski. The Posing of Questions. Logical Foundations of Erotetic
Inferences. Kluwer, Dordrecht, 1995.
25. Andrzej Wiśniewski. The logic of questions as a theory of erotetic arguments.
Synthese, 109:1–25, 1996.
26. Andrzej Wiśniewski. Questions and inferences. Logique at Analyse, in print.

A Formal Approach To Problem Solving

More Related Content

Similar to A Formal Approach To Problem Solving (20)

More from Nancy Rinehart (20)

Recently uploaded (20)

A Formal Approach To Problem Solving