Accepted Manuscript

A quantitative approach to reasoning about incomplete knowledge

Yanhong She, Xiaoli He, Yuhua Qian, Weihua Xu, Jinhai Li

PII: S0020-0255(18)30259-7
DOI: 10.1016/j.ins.2018.04.017
Reference: INS 13560

To appear in: Information Sciences

Received date: 11 June 2016

Revised date: 15 March 2018
Accepted date: 2 April 2018

Please cite this article as: Yanhong She, Xiaoli He, Yuhua Qian, Weihua Xu, Jinhai Li, A quan-
titative approach to reasoning about incomplete knowledge, Information Sciences (2018), doi:
10.1016/j.ins.2018.04.017

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

A quantitative approach to reasoning about

incomplete knowledge∗
Yanhong She1,2 , Xiaoli He2 , Yuhua Qian1 , Weihua Xu3 , Jinhai Li4
1 School of Computer and Information Technology, Shanxi University Taiyuan, 030006, China
2 College of Science, Xi’an Shiyou University Xi’an 710065, China

T
3 School of Mathematics and Statistics, Chongqing University of Technology, Chongqing 400054, China
4 Faculty of Science, Kunming University of Science and Technology, Kunming 650500, Yunnan, China

IP
e-mail:Y.H.She(yanhongshe@gmail.com,yanhongshe@xsyu.edu.cn.)

CR
ABSTRACT In this paper, we aim to present a quantitative approach to reasoning about

US
incomplete information. The study is conducted in MEL, a minimal epistemic logic relating modal
languages to uncertainty theories. The proposed approach leads to two types of epistemic truth
AN
degrees of a proposition. Some related properties are derived. By means of a more general proba-
bility distribution on the set of epistemic states, two randomized versions of epistemic truth degrees
are obtained. The connection between the notion of local probabilistic epistemic truth degree and
M

belief function is also established. Based upon the fundamental notion of the global epistemic
truth degree, the notion of epistemic similarity degree is also proposed and a kind of pseudo-metric
used for approximate reasoning in MEL is thus derived. The obtained results provide a useful
ED

supplement to the existing study in the sense that it offers a quantitative approach instead of the
qualitative manner in the literature.
PT

Keywords : Incomplete information, Quantitative approach, Epistemic truth degree, Quanti-

tative logic
CE

1 Introduction
AC

Various types of formal logical tools have been developed for the representation and reason-
ing of knowledge, including classical logics (propositional logic and predicate logic), nonclassical
logics(multiple-valued logic, fuzzy logic, modal logic, description logic, etc.)([1]-[2],[6],[10]-[13]).
They have been widely used as a formalism for knowledge representation in artificial intelligence
∗ Project supported by the National Nature Science Fund of China under Grant 61472471, 11531009, 61672332,
61322211, the Innovation Talent Promotion Plan of Shaanxi Province for Young Sci-Tech New Star (No.2017KJXX-
60) and Postdoctoral Science Foundation of China (No. 2017M610173)

1
 ACCEPTED MANUSCRIPT

and an analysis tool in computer science ([1],[11]).

Classical propositional logic([12],[19]), among others, offers an approach to reasoning bout
knowledge and beliefs based on complete information. It consists of syntactic and semantic parts.
Semantically, a valuation, which assigns 1 or 0 to any propositional variable, represents an agent’s
complete information about the real world. At the syntactic level, it is possible to express that
certain propositions are known or believed. However, it is usually not expressive enough to state
that some propositions are acknowledged as being unknown to an agent. That is, reasoning about
incomplete information requires a type of more expressive language. In [3]-[4], a minimal epistemic
logic (MEL for short) that makes it possible to reason about partial information provided by a

T
logically sophisticated agent was proposed. In MEL, atomic formulae express epistemic attitudes

IP
about formulae expressed in another propositional language. An agent can express both beliefs and
explicitly ignored facts by using modal formulae of depth 1. Semantically, incomplete knowledge

CR
about the real world possessed by an agent is presented by a non-empty subset of interpretations,
one and only one of which is, according to this agent’s beliefs, the actual state of the world. This
is what is usually called an epistemic state. Recently, theoretical studies on MEL have progressed

US
rapidly. In [5], a system MEL+ which is an extension of MEL having S5 syntax was proposed.
The semantics is not based on Kripke models with equivalence relations, but on pairs made of an
AN
interpretation (representing the real state of facts) and a non-empty set of possible interpretations
(representing an epistemic state). In [9], a generalized possibilistic logic extending both possibilistic
logic and MEL was introduced. Then in [10], two types of approaches for reasoning about ignorance
M

in GPL were proposed, based on the idea of minimal specificity and on the notion of guaranteed
possibility, respectively. Moreover, in [7]-[8], the authors have proposed to translate three-valued
ED

logics of incomplete information into MEL, provided that the third truth-value refers to the idea
of unknown Boolean truth-value.
Note that MEL just provides a qualitative approach to reasoning about incomplete informa-
PT

tion in the sense that an agent can express both beliefs and explicitly ignored facts. It lacks a
quantitative characterization. Here we wish to mention Prof. Wang for his promising study of
CE

quantitative logic ([19],[20]). In his research work, all the basic logic notions were graded and
three types of approximate reasoning patterns in the framework of propositional logic systems
were proposed. As far as MEL is concerned, owing to its propositional feature, some basic ideas
AC

underlying quantitative logic can be naturally employed. More precisely, if E 6|= 2α holds for an
epistemic state E, i.e., E 6⊆ [α], then the agent may not assert α, such a point does not reflect
the quantitative manner. It seems more reasonable to assume that if E has a larger overlap with
[α], the agent may assert α to a larger degree, moreover, how to combine the truth degrees of a
proposition relative to different epistemic states to reflect the global epistemic truth degree requires
further investigation. Such a quantitative description and the existing qualitative reasoning can
greatly improve the expressive power of MEL.
Motivated by such a consideration, we seek to introduce the notion of epistemic truth degree of

2
 ACCEPTED MANUSCRIPT

any formula in MEL, from two different types of viewpoints: local viewpoint and global viewpoint.
In local viewpoint, we mainly restrict our attention to the set of epistemic states related to the
propositional variables contained in each formula, in this sense, epistemic truth degrees of different
formula cannot be compared directly. On the contrary, in global viewpoint, we have a fixed set
of epistemic states for all formulae in MEL. From both viewpoints, by employing a more general
probability distribution on the set of epistemic states, two randomized versions of epistemic truth
degrees are also obtained. A comparative study with belief function, quantitative logic is also
performed.
The rest of the paper proceeds as follows: In Section 2, we mainly review some basic notions

T
in MEL. In Section 3, we introduce the notion of epistemic truth degree from the local viewpoint

IP
and derive some related properties. In Section 4, the notion of epistemic truth degree from the
global viewpoint is proposed, some new properties are obtained. In Section 5, a comparative study

CR
between the proposed notion, quantitative logic and belief function is performed. We complete
this paper with some concluding remarks, as shown in Section 6.

2
2.1
Preliminaries
A simple logic for reasoning bout knolwedge
US
AN
In what follows, we briefly recall the minimal epistemic logic[3],[4] (MEL for short) used for reasoning
about incomplete knowledge.
M

2.1.1 Syntax
ED

In MEL, a finite set of propositional variables P V = {p1 , p2 , · · · , pk } is used. By using Boolean

connectives ¬ and ∧, the set of Boolean propositional logic formulae (denoted by L) can be built.
The proposed syntax in MEL is obtained by encapsulating propositional language inside a lan-
PT

guage equipped with a modality denoted by 2. The intended purpose is to completely separate
propositions in L referring to the real world and propositions that refer to an agent’s epistemic
CE

state, where the symbol 2 appears.

The atoms of MEL are obtained by adding the unary connective 2 in front of all formulae in
L, that is, atomic formulae of MEL are of the form 2α, α ∈ L. The intended meaning of 2α is that
AC

an agent knows (or believes) proposition α is true, semantically speaking, α holds in each possible
world compatible with the agent’s epistemic state.
The set of MEL-formulae, denoted by φ, ψ, · · · forms a set of modal language L2 generated
recursively from the set At of atomic formulae, with the help of the same Boolean connectives
¬, ∧ : that is,
2α ∈ L2 , whenever α ∈ L; ¬φ ∈ L2 , φ ∧ ψ ∈ L2 , whenever φ, ψ ∈ L2 .
From the primitive connectives ¬ and ∧, one can define disjunction φ ∨ ψ = ¬(¬φ ∧ ¬ψ) and
the modality α = ¬2¬α, where α ∈ L, in the usual way. In MEL, modalities 2 and  only apply

3
 ACCEPTED MANUSCRIPT

to PL-formulae, contrary to usual modal logics. The intended meaning of α is that an agent
either believes α is true, or ignores whether α is true or not.
Definition 1[4] The set of axioms in MEL consists of the following formulae:
(PL) (i) φ → (ψ → φ),
(ii) (φ → (ψ → µ)) → ((φ → ψ) → (φ → µ)),
(iii) (¬φ → ¬ψ) → (ψ → φ),
(K) 2(α → β) → (2α → 2β),
(N) 2α, whenever `P L α,
(D) 2α → α.

T
Rule:

IP
(MP) If φ, φ → ψ, then ψ.
The following results shed more light on the connection between propositional logic and MEL.

CR
We use α, β, · · · to denote formulae in proposition logic while φ, ψ, · · · to denote formulae in MEL.
Proposition 1[4] (i) `P L α → β if and only if `M EL 2α → 2β,
(ii) `P L α → β if and only if `M EL α → β,
(iii) `P L α → β if and only if `M EL 2α → β.
US
AN
2.1.2 Semantics

In the semantic part of MEL, an epistemic state E is represented by a subset of mutually exclu-
sive propositional valuations. Each valuation v represents a ‘possible world’ compatible with the
M

epistemic state of the agent, and has the form v : P V → {0, 1}. Let V denote the set of valuations
in propositional logic. Sometimes, a valuation in propositional logic can also be represented as a
ED

k-dimensional vector. For α ∈ L, α is the truth function obtained by substituting each atom pi in
α by xi and interpreting ¬ as the unary operation ¬x = 1 − x on {0, 1}, and ∧ as the corresponding
minimum operation on {0, 1}2 . For any propositional valuation v ∈ V and α(p1 , · · · , pk ), we have
PT

v(α) = α(v(p1 ), · · · , v(pk )). According to the definition of an epistemic state, we have E ⊆ V. An
epistemic state is further assumed to be non-empty (otherwise the agent is inconsistent).
The satisfaction of MEL-formulae is defined recursively, as follows. For any α ∈ L, φ, ψ ∈
CE

L2 ,E(6= ∅) ⊆ V :
•E |= 2α if and only if E ⊆ [α],
AC

•E |= ¬φ if and only if E 6|= φ,

•E |= φ ∧ ψ if and only if E |= φ and E |= ψ,
where [α] symbols the set of valuations which assign 1 to α.
Considering that α = ¬2¬α, we have
E |= α if and only if E ∩ [α] 6= ∅.
Some notions such as semantic equivalence, semantic consequence, etc., can be presented in a
similar way as in propositional logic.

4
 ACCEPTED MANUSCRIPT

Since L2 is a propositional language, we can also define a standard propositional semantics. A

propositional model of a MEL-formula is an interpretation v of L2 , that is, a mapping from At to
{0, 1}. Let V2 be the set of such propositional valuations. It is clear that the propositional logic
thus defined from atoms in At, using language L2 , axioms P L and inference rule MP is sound
and complete with respect to this propositional semantics. For the logic MEL, one must restrict
to standard interpretations v that satisfy axioms K,D and N , forming a subset of V2 denoted by
V M EL .
More precisely, valuations in V M EL obey the following conditions:
v(φ) = 1 for any tautology φ ∈ L2 ,

T
(K) v(2(α → β)) ≤ v(2α → 2β),

IP
(N) v(2α) = 1, whenever `P L α,
(D) v(2α) ≤ v(α).

CR
2.1.3 Soundness and completeness theorem

It has been shown in [4] that V M EL is in one-to-one correspondence with the set of epistemic states
{E : ∅ =
US
6 E ⊆ V}, and that the epistemic semantics is equivalent to the standard propositional
semantics restricted to V M EL . The soundness and completeness of MEL follows directly from the
AN
one of propositional logic.
Proposition 2[4] (i) V M EL is in one-to-one correspondence with the set of epistemic states
{E : ∅ =
6 E ⊆ V}.
M

(ii) Given a valuation v satisfying MEL axioms, there exists an associated epistemic state
Ev = {w ∈ V| gv (V\{w}) = 0}, where gv is the necessity measure defined by gv ([α]) = v(2α),
ED

such that v(φ) = 1 (in PL semantics) if and only if Ev |= φ (in MEL semantics).
(iii) Given an epistemic state E, there exists a valuation vE such that vE (φ) = 1 (in PL
semantics) if and only if E |= φ (in MEL semantics), where vE is defined as follows: vE (2α) = 1
PT

if E ⊆ [α], and 0 otherwise.

(iv) For any valuation v satisfying MEL axioms, v = vEv ; For any epistemic state E, E = EvE .
For any set H of formulae in propositional logic, let 2H = {2α | α ∈ H}.
CE

Theorem 1[4] 2H `M EL 2β if and only if H `P L β.

Theorem 2[4] (Soundness and completeness) For any set Γ ∪ {φ} of MEL-formulae,
AC

Γ `M EL φ if and only if Γ |=M EL φ.

3 Epistemic truth degree of propositions in MEL

In this section, we aim to present a graded version of satisfaction of propositions in MEL. We begin
with the following analysis of the expressive power of MEL.

5
 ACCEPTED MANUSCRIPT

3.1 MEL provides a qualitative approach to reasoning about incomplete

information

MEL emerges as an attempt to bridge possibility theory and modal logic, two commonly
used knowledge representation frameworks. It offers a logical grounding to uncertainty theories of
incomplete information. In MEL, an agent can express beliefs and ignored facts by modal formulae
of depth 1, not objective formulae.
One remarkable feature of MEL is that it just provides a qualitative approach to reasoning
about incomplete knowledge, which can be observed from both syntactic part and semantic part

T
of MEL. In syntactics, by using formula of the form 2α, the agent can express that (s)he believes
α. Or in other words, the agent can assert beliefs in proposition α. However, under incomplete

IP
information, an agent may not assert her/his beliefs in some propositions, that is, 2α may not

CR
hold. Then in MEL, by using α, the agent expresses that α is possibly true.
We then observe that in MEL, an agent can only express complete beliefs (including beliefs
that α is true and α is false), partial beliefs of a formula in propositional logic. This is a kind of

US
ternary judgement. However, such a language cannot express that the agent believes a proposition
to an extent, when (s)he does not have enough information to believe α. α just provides a
qualitative belief, it cannot embody the idea of epistemic grade. A more intended case is that for
AN
two propositions α and β, an agent believes α to a larger degree than β, when s(he) has partial
information about both α and β.
Such a point can also be observed from the viewpoint of semantics. Recall that 2α holds if
M

and only if in all possible worlds compatible with what the agent believes, it is the case that α
holds. Mathematically, E |= 2α if and only if E ⊆ [α]. Similarly, E |= α if and only if E ∩ [α] 6= ∅.
ED

We observe that the truth of 2α, α is closely related to the set-theoretic relationship between
E and [α]. However, the complete inclusion and non-empty intersection are just two qualitative
descriptions of two crisp sets. It lacks some quantitative analysis. To state it more clearly, let us
PT

consider two formulae α, β, in the same epistemic state E. If [α] is partially included in E to a
|E∩[α]| |E∩[β]|
larger degree than [β] (numerically, |E| ≥ |E| ), then a more plausible conclusion should be
CE

that the agent has a larger degree of belief in α than β.

3.2 Epistemic truth degree of formulae in MEL from a local viewpoint

AC

The analysis in the previous subsection motivates us to present a graded version of epistemic
attitude towards formulae in MEL. As we will see below, the definition of epistemic truth degree
depends on the number of propositional variables contained in the given formula. This is what the
term “local viewpoint” explicitly refers to. Moreover, such an approach is mainly conducted at the
semantical level. In view of the soundness and completeness theorem, the semantical gradedness
also leads to the graded version of syntactic part.
Let φ ∈ L2 , we use V φM EL to denote the set of valuations satisfying MEL axioms when

6
 ACCEPTED MANUSCRIPT

restricted to the set of propositional variables contained in φ and use V φ to denote the set of
Boolean valuations on the set of propositional variables contained in φ.
Proposition 3 Let φ ∈ L2 be a formula containing n propositional variables, then |V φM EL |=
n
22 − 1.
Proof Taking into account Proposition 2, we have |V φM EL |=|{E : ∅ =
6 E ⊆ {0, 1}n }|. Then
n
6 E ⊆ {0, 1}n }| = 22 − 1.
the desired result immediately follows from the fact that |{E : ∅ =
Definition 2 Let φ ∈ L2 , define
φ
L |{v ∈ VM EL : v(φ) = 1}|
τM EL (φ) = φ
,
|VM EL |

T
L
then we call τM EL (φ) the local epistemic truth degree of φ.

IP
L
Observe from Definition 2 that τM EL (φ) indeed measures the portion of valuations in VM EL

CR
assigning 1 to φ. This leads to the notion of epistemic truth degree.
In view of the fact that there exists a one-to-one correspondence between the set of epistemic
states and that of valuations satisfying MEL axioms, Definition 2 can be equivalently stated as
follows:
US
Proposition 4 Let φ ∈ L2 be a formula containing n propositional variables, and P be an
even probability distribution on the set of epistemic states contained in {0, 1} , then
n
AN
X n
X n
L
τM EL (φ) = {P (E) | π(E, φ) = 1, ∅ =
6 E ⊆ {0, 1} } = {P (E)π(E, φ) | ∅ =
6 E ⊆ {0, 1} },

where π(E, φ) = 1 if and only if E |= φ, and 0 otherwise.

M

Proof Trivially.
Example 1 Let φ = 2p1 , ϕ = p1 , ψ = 2(p1 ∨ p2 ), compute the local epistemic truth degree
ED

of φ, ϕ and ψ.
Solution For φ, since φ contains only one propositional variable p1 , i.e., n = 1, then we have
1
from Proposition 3 that the cardinality of V φM EL is equal to 22 − 1 = 3. Since V φ = {0, 1}, we have
PT

6 E ⊆ V φ } = {E1 , E2 , E3 } with E1 = {1}, E2 = {0}, E3 = {1, 0}. According to Proposition

{E : ∅ =
2, we have altogether 3 valuations satisfying MEL axioms, i.e., vE1 , vE2 , vE3 . It can be checked
CE

easily that vE1 (2p1 ) = 1 while vE2 (2p1 ) = vE3 (2p1 ) = 0. Then we have from Definition 2 that
L 1
τM EL (φ) = 3 .

For ϕ, it can be checked easily that vE1 (p1 ) = vE3 (p1 ) = 1 while vE2 (p1 ) = 0. Then we
AC

L 2
have from Definition 2 that τM EL (ϕ) = 3 .

Similarly, for ψ, since ψ contains two propositional variables p1 , p2 , i.e., n = 2, then we have
2
from Proposition 3 that the cardinality of V ψ
M EL is equal to 2
2
− 1 = 15. It can be checked that
6 E ⊆ V ψ } = {E1 , E2 , · · · , E15 }, where
{E : ∅ =
E1 = {(1, 1)}, E2 = {(1, 0)}, E3 = {(0, 1)}, E4 = {(0, 0)}, E5 = {(1, 1), (1, 0)}, E6 = {(1, 1), (0, 1)},
E7 = {(1, 1), (0, 0)}, E8 = {(1, 0), (0, 1)}, E9 = {(1, 0), (0, 0)}, E10 = {(0, 1), (0, 0)}, E11 = {(1, 1), (1, 0),
(0, 1)}, E12 = {(1, 1), (1, 0), (0, 0)}, E13 = {(1, 1), (0, 1), (0, 0)}, E14 = {(1, 0), (0, 1), (0, 0)}, E15 =
{(1, 1), (1, 0), (0, 1), (0, 0)}. By using Proposition 3, we can show that vEi (ψ) = 1 for i = 1, 2, 3, 5, 6, 8, 11.

7
 ACCEPTED MANUSCRIPT

That is, we have altogether 7 valuations satisfying MEL axioms, then we have from Definition 2
L 7
that τM EL (ψ) = 15 .

Observe from Definition 2 and Proposition 3 that the epistemic truth degree of a formula
n
containing n propositional variables must be a fraction with the denominator 22 − 1. Then one
n
t
natural question arises: for any fraction of the form 22n −1
(0 ≤ t ≤ 22 − 1, t ∈ Z), does there exist
L t
some formula φ such that τM EL (φ) = 22n −1
? The following proposition gives a positive answer to
this question.
n
t
Proposition 5 For any proper fraction of the form 22n −1
(0 ≤ t ≤ 22 − 1, t ∈ Z), there exists
L t
some formula (say as φ) in L2 such that τM EL (φ) = 22n −1
.

T
n
Proof For any epistemic state E ⊆ {0, 1} , define

IP
^
δE = 2αE ∧ αw ,

CR
w∈E
V V
where αE is the formula satisfying [αE ] = E and αw = w(p)=1 p∧ w(p)=0 ¬p is a formula
characterizing w.

US
It has been shown in [4] that the set of epistemic models of δE is equal to the singleton {E}.
n
For integer t satisfying 0 ≤ t ≤ 22 − 1, we choose arbitrarily t epistemic states contained in
n
{0, 1} , say as E1 , E2 , · · · , Et , then we define φ = δE1 ∨ δE2 ∨ · · · ∨ δEt . It can be checked that the
AN
set of epistemic models of φ is {E1 , E2 , · · · , Et }. Moreover, considering the fact that there exists
a one-to-one correspondence between epistemic states and valuations satisfying MEL axioms, we
can thus conclude that there exists t valuations in V M EL assigning 1 to φ. Then we have from
M

L t
Definition 2 that τM EL (φ) = 22n −1
.
L
The following proposition provides a necessary and sufficient condition for τM EL (φ) = 1.
ED

L
Proposition 6 Let φ ∈ L2 , then τM EL (φ) = 1 if and only if φ is a theorem in MEL.

Proof “⇒:” Assume, without any loss of generality, that φ contains n propositional variables,
L
then τM EL (φ) = 1 implies, according to Definition 2, that v(φ) = 1 holds for each valuation
PT

v ∈ V φM EL . Considering the one-to-one correspondence between the set of valuations satisfying

MEL axioms and that of epistemic states, we then conclude from Proposition 2 that E |= φ holds
CE

for each epistemic state, that is, φ is a valid formula, then it follows from Theorem 2 that φ is a
theorem in MEL.
“⇐:” Conversely, if φ is a theorem in MEL, then according to Theorem 2, E |= φ holds for each
AC

epistemic state, considering the one-to-one correspondence between the set of valuations satisfying
MEL axioms and that of epistemic states, we obtain that v(φ) = 1 holds for each v ∈ V φM EL . Then
L
we have from Definition 2 that τM EL (φ) = 1.

Proposition 7 Let φ, ψ ∈ L2 be two formulae both containing n propositional variables, then

L L
`M EL φ → ψ implies τM EL (φ) ≤ τM EL (ψ).

Proof Since both φ, ψ contain n propositional variables, then according to Definition 2,

φ φ ψ
L |{v∈VM :v(φ)=1}| |{v∈VM :v(φ)=1}| L |{v∈VM :v(ψ)=1}|
we have τM EL (φ) =
EL
φ
|VM |
= EL
22n −1
, τM EL (ψ) =
EL
ψ
|VM |
=
EL EL

8
 ACCEPTED MANUSCRIPT

ψ
|{v∈VM :v(ψ)=1}| φ ψ
EL
22n −1
and VM EL = VM EL . Since φ → ψ is a theorem in MEL, then we have from
φ L L
Theorem 2 that v(φ) ≤ v(ψ) holds for each v ∈ VM EL , and therefore, τM EL (φ) ≤ τM EL (ψ).

Proposition 7 yields immediately the following corollary.

Corollary 1 Let φ, ψ ∈ L2 be two formulae both containing n propositional variables, if φ
L L
and ψ are logically equivalent, then τM EL (φ) = τM EL (ψ).

Remark 1 If the condition “both φ and ψ contain the same propositional variables” is with-
drawn, then the desired result may not hold. Please see the following example:
Let φ = 2p1 , ψ = 2p1 ∧ 2(p2 → p2 ), clearly, φ and ψ are logically equivalent. We have
L 1
from Example 1 that τM EL (φ) = 3. For ψ, since ψ contains two propositional variables p1 , p2 ,

T
ψ 2
2
then we have from Proposition 3 that the cardinality of VM EL is equal to 2 − 1 = 15. Since

IP
V ψ = {(1, 1), (1, 0), (0, 1), (0, 0)}, we obtain {E : ∅ =
6 E ⊆ V ψ } = {E1 , E2 , · · · , E15 }, as shown in
Example 1. We have from Proposition 2 that vEi (ψ) = 1 for i = 1, 2, 5. That is, we have

CR
L
altogether 3 valuations satisfying MEL axioms, then we have from Definition 2 that τM EL (ψ) =
ψ
|{v∈VM :v(ψ)=1}| 3
EL
ψ
|VM |
= 15 = 15 , which shows that τM
L L
EL (φ) 6≤ τM EL (ψ).
EL

US
L L
Proposition 8 Let φ ∈ L2 , then τM EL (¬φ) = 1 − τM EL (φ).
¬φ
|{v∈VM :v(¬φ)=1}|
Proof Clearly, V ¬φ M EL = V φM EL . Then by definition, τM
L
EL (¬φ) =
EL
¬φ
|VM |
=
EL
φ φ
|{v∈VM :v(φ)=0}| |{v∈VM :v(φ)=1}| L
EL
φ =1− EL
φ = 1 − τM EL (φ).
AN
|VM EL
| |VM EL
|
Proposition 9 Let φ, ψ ∈ L2 be two formulae both containing n propositional variables, then
L L L L
τM EL (φ ∨ ψ) = τM EL (φ) + τM EL (ψ) − τM EL (φ ∧ ψ).
φ∨ψ φ ψ
Proof Since φ, ψ ∈ L2 both contain n propositional variables, then VM EL = VM EL = VM EL =
M

n
22 −1. For the sake of discussion, we denote them by VM EL . By definition, we have τM
L
EL (φ∨ψ) =
|{v∈VM EL :v(φ∨ψ)=1}| |{v∈VM EL :v(φ∨ψ)=1}| |{v∈VM EL :v(φ)=1}| |{v∈VM EL :v(ψ)=1}| |{v∈VM EL :v(φ∧ψ)=1}|
|VM EL | = 22n −1
= 22n −1
+ 22n −1
− 22n −1
=
ED

L L L
τM EL (φ) + τM EL (ψ) − τM EL (φ ∧ ψ).

Remark 2 As in Proposition 7, the precondition in Proposition 9 cannot be withdrawn,

otherwise, the desired result may not hold.
PT

The notion of epistemic truth degree can be generalized by means of more general probability
distributions on the set of epistemic states.
CE

Definition 3 Let φ ∈ L2 and P be a probability distribution on the set of epistemic states.

Define
LP
τM EL (φ) = ΣE⊆V φ {P (E) | E |= φ},
AC

LP
then we call τM EL (φ) the local probabilistic epistemic truth degree of φ.

For MEL-formula of the form 2α with α ∈ L, we can present a graded version of Definition 3
by further employing the inclusion degree relative to any epistemic state.
Definition 4 Let 2α ∈ L2 and P be a probability distribution on the set of epistemic states.
Define
LP G |E ∩ [α]|
τM EL (2α) = ΣE⊆V 2α {P (E) },
|E|
where [α] = {v ∈ V 2α | v(α) = 1}.

9
 ACCEPTED MANUSCRIPT

LP LP G
We have the following proposition stating the relationship between τM EL (φ) and τM EL (φ).
LP LP G
Proposition 10 (i) τM EL (2α) ≤ τM EL (2α),
LP LP G
(ii) τM EL (2α) = τM EL (2α) if and only if [α] is a singleton.

Example 2 Let φ = 2(p1 ∨ p2 ) and P be an even probability distribution on the set of

2
epistemic states contained in {0, 1} . Compute the local probabilistic epistemic truth degree of φ.
Solution Since φ contains two propositional variables p1 , p2 , i.e., n = 2, then we have from
2
Proposition 2 that the cardinality of V φM EL is equal to 22 −1 = 15. Since V φ = {(1, 1), (1, 0), (0, 1), (0, 0)},
6 E ⊆ V φ } = {E1 , E2 , · · · , E15 }, as shown in Example 1. Denote p1 ∨ p2 by α, an
we have {E : ∅ =
|Ei ∩[α]|
easy verification shows that [α] = {(1, 0), (0, 1), (1, 1)} and =1 for i = 1, 2, 3, 5, 6, 8, 11. For

T
|Ei |
|E4 ∩[α]| |E7 ∩[α]| |E9 ∩[α]|
the other epistemic states, |E4 | = 0, |E7 | = |E9 | = |E|E 10 ∩[α]|
10 |
= 12 ,|E12 ∩[α]|
|E12 | = |E|E
13 ∩[α]|
13 |
=

IP
|E14 ∩[α]| |E15 ∩[α]| 7+3× 2 +3× 3 + 34
1 2

|E14 | = 23 , |E15 | = 34 , then we have from Definition 4 that τM

LP G
EL (φ) = 222 −1
= 4.3

CR
3.3 Epistemic truth degree in MEL vs truth degree in QL

Recall that in the quantitative logic (QL for short) originated in [20], Wang and Zhou introduced

US
the concept of the degree of the truth in the framework of so called many-valued propositional
logic systems with the intention of measuring to what extent a given formula is true. Some further
studies along this research line can be found in [21] and [25]. Since MEL is also a propositional
AN
logic with the formulae of the form 2α being its atoms, a close relationship between epistemic
truth degree in MEL and that in QL is thus expected. In what follows, we will examine this issue
in detail.
M

Definition 5[20] Let α(p1 , · · · , pn ) be a formula in classical propositional logic, define

|α−1 (1)|
τ2 (α) = ,
ED

2n

then we call τ2 (α) the truth degree of α.

The following results concerning the notion of truth degree in QL are summarized in [20].
PT

Proposition 11 The set of truth degrees of formulae in classical propositional logic is equal
to { 2mn , n = 1, 2, · · · , m = 0, 1, · · · , 2n }.
CE

Proposition 12 Let α, β be two formulae in classical propositional logic, then

(1) τ2 (α) = 1 if and only if α is a theorem (or equivalently, a tautology) in classical propositional
logic; τ2 (α) = 0 if and only if α is a refutable formula (or equivalently, a contradiction) in classical
AC

propositional logic,
(2) τ2 (¬α) = 1 − τ2 (α),
(3) τ2 (α ∨ β) = τ2 (α) + τ2 (β) − τ2 (α ∧ β),
(4) ` α → β implies that τ2 (α) ≤ τ2 (β).
There exist similarities as well as differences between the notion of truth degree in quantitative
logic and that of epistemic truth degree in MEL.
Similarities: (i) Both are proposed with the aim of measuring the extent to which any formula
is true.

10
 ACCEPTED MANUSCRIPT

(ii) They enjoy similar properties. For instance, the (epistemic) truth degree of a formula is
equal to 1 if and only if it is a theorem in classical propositional logic (MEL); the (epistemic) truth
degree of a formula is equal to 0 if and only if it is a refutable formula, et al.
Differences: (i) The epistemic truth degree is proposed in the context of incomplete infor-
mation while the notion of truth degree was introduced in the context of complete information.
n
m
(ii) The epistemic truth degrees of formulae in MEL are of the form 22n −1
(0 ≤ m ≤ 22 −1, m ∈
m
Z) while the truth degrees of formulae in QL enjoy the form 2n (0 ≤ m ≤ 2n , m ∈ Z).
(iii) In QL, some properties, e.g., τ2 (α ∨ β) = τ2 (α) + τ2 (β) − τ2 (α ∧ β), hold without any
restriction on α, β. On the contrary, in MEL, τ2 (φ ∨ ψ) = τ2 (φ) + τ2 (ψ) − τ2 (φ ∧ ψ) does not hold

T
generally. Proposition 9 tells that when both φ and ψ have the same set of propositional variables,

IP
then the desired property holds.
The following proposition establishes the relationship between the notion of truth degree in

CR
QL and that of epistemic truth degree in MEL.
Proposition 13 Let α be a formula in propositional logic containing n propositional variables
m L 2m −1
p1 , · · · , pn . If τ2 (α) = 2n , then τM EL (2α) = .
Proof According to Definition 5, τ2 (α)
US
22n −1

= 2mn means that |α−1 (1)| = m, that is, there are

altogether m valuations assigning truth value 1 to α. For α−1 (1), there are 2m − 1 non-empty
AN
subsets, say as E1 , · · · , E2m −1 . Since Ei ⊆ [α] = α−1 , vEi (2α) = 1 holds according to Proposition
0
2, i = 1, · · · , 2m − 1. For epistemic state E which are not subset of E, we have vE 0 (2α) = 0, that
is, E1 , · · · , E2m −1 are exactly those epistemic states at which 2α is true. Considering the one-to-
M

one correspondence relationship between epistemic states and valuations satisfying MEL axioms,
we conclude that there are altogether 2m − 1 valuations satisfying axioms, then by Definition 2,
2m −1
ED

L
τM EL (2α) = 22n −1
.
Proposition 13 yields the following corollary.
L
Corollary 2 Let α be a refutable formula in propositional logic, then we have τM EL (2α) = 0;
PT

and vice versa.

3.4 Connection with belief function theory

CE

In this subsection, it will be shown that the notion of epistemic truth degree can be interpreted
from the viewpoint of belief function[16] .
AC

Recall that a belief function is a mapping Bel from 2U to the unit interval [0, 1] and satisfies
the following axioms:
(F1) Bel(∅) = 0,
(F2) Bel(U ) = 1,
(F3) For every positive integer n and every collection A1 , · · · , An ⊆ U,Bel(A1 ∪A2 ∪· · ·∪An ) ≥
Σi Bel(Ai ) − Σi<j Bel(Ai ∩ Aj ) + · · · + (−1)n+1 Bel(A1 ∩ · · · ∩ An ).
Axioms (F1) and (F2) may be considered as normalization conditions. Axiom (F3) is a weaker
version of the commonly known additivity axiom of probability functions. It is referred to as the

11
 ACCEPTED MANUSCRIPT

axiom of superadditivity.
A belief function can be equivalently defined by another mapping, m : 2U → [0, 1], which is
called a basic probability assignment and satisfies two axioms:
(M1) m(∅) = 0,
(M2) ΣX⊆U m(X) = 1.
A subset A ⊆ U with m(A) > 0 is called a focal element. Using the basic probability assign-
ment, belief of Y can be expressed as:
(M3) Bel(Y ) = ΣX⊆Y m(X).
The following proposition shows that the notion of epistemic truth degree can be interpreted

T
in terms of belief function.

IP
Proposition 14 Let α ∈ L be a formula containing n propositional variables, P be a proba-
nLP
bility distribution on the set of epistemic states contained in {0, 1} and τM EL be the probabilistic

CR
n
epistemic truth degree mapping. Define a mapping Bel : 2{0,1} → [0, 1] by

LP n
Bel(X) = τM EL (2αX ), X ⊆ {0, 1} , (1)

US
where αX is the formula satisfying [αX ] = X, then Bel is a belief function on 2{0,1} .
LP
Proof It follows from (1) and Definition 3 that Bel(X) = τM
n

EL (2αX ) = Σ{P (E) : E |= 2αX } =

AN
n
Σ{P (E) : E ⊆ X}, then we have from (M3) that Bel is a belief function on 2{0,1} .
n
Proposition 15 Let Bel be a a belief function on 2{0,1} . Define a mapping τ : L2 → [0, 1] by
M

τ (2α) = Bel(Xα ), α ∈ L, (2)

where Xα = [α], then there exists a probability distribution P on the set of epistemic states
ED

n LP
contained in {0, 1} such that τ coincides with τM EL induced by P .

Proof Since Bel is a belief function, there exists a probability distribution P (satisfying
n
P (∅) = 0) on the set of epistemic states contained in {0, 1} such that Bel(Y ) = ΣX⊆Y P (X). Then
PT

we have from (2) and (M3) that τ (2α) = Bel(Xα ) = Σ{P (Y ) | Y ⊆ Xα } = Σ{P (Y ) | Y |= 2α},
LP
which shows that τ is equal to τM EL with the restriction on the atomic formulae in L2 .
CE

Indeed, Proposition 15 holds for each MEL-formula obtained by using only the connective ∧,
we omit the detailed proof here.
AC

3.5 A new definition of epistemic truth degree from the global viewpoint

In the definition of epistemic truth degree, for a formula containing n propositional variables, a
n
set of epistemic states of the form {E | ∅ =
6 E ⊆ {0, 1} } is taken into consideration. That is, the
definition of epistemic truth degree depends on the number of propositional variables contained
in the concerned formula. In other words, such a definition is given from a local viewpoint. As a
consequence, the epistemic truth degrees of different formulae cannot be compared directly owing
to the fact that their epistemic settings are different.

12
 ACCEPTED MANUSCRIPT

The above-mentioned fact motivates us to present a new definition of epistemic truth degree
from a global manner. In what follows, we always assume that MEL is built upon a fixed set of n
propositional variables and we use V M EL to denote the set of valuations satisfying MEL axioms.
Definition 6 Let φ ∈ L2 , define

G |v ∈ VM EL | v(φ) = 1|
τM EL (φ) = ,
22n − 1
G
then we call τM EL (φ) the global epistemic truth degree of φ.

Observe from Definition 6 that for different formulae, the proposed new definition of global
epistemic truth degree measures the portion of valuations assigning 1 to A in the same set V M EL .

T
This is what the term “global” refers to.

IP
Example 3 Let φ = 2p1 , compute the global epistemic truth degree of φ.
Solution For valuation v ∈ V M EL , we have from Proposition 2 that there exists an epistemic

CR
n
state E ⊆ {0, 1} such that v = vE . If v(φ) = 1, i.e., vE (2p1 ) = 1, then we have E ⊆ [p1 ].
Moreover, since [p1 ] = {(1, x2 , · · · , xn )} with xi ∈ {0, 1}, it can be easily computed that |[p1 ]| =
n−1
2n−1 , then the number of epistemic states satisfying E ⊆ [p1 ] is equal to 22
Definition 6, we have G
τM EL (φ) = 22
n−1

22n −1
= 1
22n−1 +1 US .
L
A similar proof shows that the previously proved propositions for τM G
, and thus, by

EL also hold for τM EL .

AN
G
Moreover, τM EL enjoys a more desired property like that in probability theory.
G G G G
Proposition 16 Let φ, ψ ∈ L2 , then τM EL (φ ∨ ψ) = τM EL (φ) + τM EL (ψ) − τM EL (φ ∧ ψ).

Proof Since for any valuation v ∈ V M EL , v(φ ∨ ψ) = v(φ) + v(ψ) − v(φ ∧ ψ), then by Definition
M

G G G G
6, we have τM EL (φ ∨ ψ) = τM EL (φ) + τM EL (ψ) − τM EL (φ ∧ ψ).

The set of global epistemic truth degrees enjoys the following form.
ED

Proposition 17 Let H denote the set of global epistemic truth degrees of formulae in MEL,
n
then H = { 22nk−1 | k = 0, 1, · · · , 22 − 1}.
Proposition 18 Let φ, ϕ, ψ ∈ L2 , then
PT

G G G
(i) If τM EL (φ) ≥ α, τM EL (φ → ϕ) ≥ β, then τM EL (ϕ) ≥ α + β − 1,
G G G
(ii) If τM EL (φ → ϕ) ≥ α, τM EL (ϕ → ψ) ≥ β, then τM EL (φ → ψ) ≥ α + β − 1.
G G G G G
CE

Proof (i) τM EL (φ → ϕ) = τM EL (¬φ ∨ ϕ) = 1 − τM EL (φ) + τM EL (ϕ) − τM EL (¬φ ∧ ϕ) ≥ β,

G G
and so, τM EL (ϕ) ≥ β + τM EL (φ) − 1 ≥ α + β − 1.

(ii) It can be proved similarly.

AC

The proposed notion of global epistemic truth degree can be further generalized by using a
more general probability distribution on the set of epistemic states, as in Definition 4.
Definition 7 Let P be a probability distribution on the set of epistemic states contained in
n
{0, 1} . Define
GP
τM EL (φ) = ΣE⊆{0,1}n {P (E) | E |= φ},

GP
then we call τM EL (φ) probabilistic global epistemic truth degree of φ.

Proposition 19 There exists a one-to-one correspondence between the set of probabilistic

n
global epistemic truth degree functions on L2 and that of belief functions on 2{0,1} .

13
 ACCEPTED MANUSCRIPT

GP GP
Proof Given an epistemic truth degree function τM EL , then by Definition 7, τM EL cor-

responds to a probability distribution on the set of epistemic states, which is indeed the basic
probability assignment m in theory of belief functions. Consequently, m uniquely determines a
n
belief function on 2{0,1} . This shows that an epistemic truth degree function uniquely determines
a belief function.
n
Conversely, given a belief function Bel on the set 2{0,1} , then Bel uniquely determines a
basic probability assignment function m, which is indeed a general probability distribution. By
GP
Definition 7, such a probability distribution can lead to a truth degree function τM EL .

T
4 Epistemic similarity degree between formulae in MEL

IP
In MEL, an important result states that 2α and 2β are semantically equivalent if and only if

CR
α, β are logically equivalent in propositional logic. That is, in context of incomplete information,
an agent can assert belief in α and β simultaneously if and only if α, β are logically equivalent in
propositional logic. For instance, 2p1 and 2(p1 ∧ (p2 → p2 )). In this section, by grading the notion

US
of logical equivalence, we aim to present graded versions of similarities between any two formulae
in MEL.
AN
In what follows, we adopt the global view of epistemic truth degree to introduce the notion of
epistemic similarity degree.
Definition 8 Let φ, ϕ ∈ L2 , define
M

G
ξM EL (φ, ϕ) = τM EL ((φ → ϕ) ∧ (ϕ → φ)),

and we then call ξM EL (φ, ϕ) the epistemic similarity degree between φ and ϕ.
ED

Proposition 20 Let φ, ϕ, ψ ∈ L2 ,
(i) ξM EL (φ, ϕ) = 1 if and only if φ ≈ ϕ,
PT

(ii) ξM EL (φ, ϕ) = 0 if and only if φ ≈ ¬ϕ,

(iii) ξM EL (φ, ϕ) + ξM EL (φ, ¬ϕ) = 1,
(iv) ξM EL (φ, ϕ) + ξM EL (ϕ, ψ) ≤ 1 + ξM EL (φ, ψ).
CE

Proof Both (i) and (ii) trivially hold.

(iii) According to Definition 6 and Definition 8, it suffices to show that for each v ∈ V M EL ,
AC

either v((φ → ϕ) ∧ (ϕ → φ)) = 1 or v((φ → ¬ϕ) ∧ (¬ϕ → φ)) = 1 holds. Indeed, we have
v(φ), v(ϕ) ∈ {0, 1} that there are altogether four cases: (1) (v(φ), v(ϕ)) = (1, 1), (2) (v(φ), v(ϕ)) =
(0, 0), (3) (v(φ), v(ϕ)) = (0, 1), (4) (v(φ), v(ϕ)) = (1, 0). Clearly, in case (1) and case (2), v((φ →
ϕ) ∧ (ϕ → φ)) = 1 holds, and in case (3) and (4), v((φ → ¬ϕ) ∧ (¬ϕ → φ)) = 1 holds, as desired.
(iv) Denote
X = {v ∈ V M EL | v((φ → ϕ) ∧ (ϕ → φ)) = 1},
Y = {v ∈ V M EL | v((ϕ → ψ) ∧ (ψ → ϕ)) = 1},
Z = {v ∈ V M EL | v((φ → ψ) ∧ (ψ → φ)) = 1}.

14
 ACCEPTED MANUSCRIPT

Clearly, X ∩ Y ⊆ Z. According to Definition 6 and Definition 8, (iv) can be equivalently

written as
|X| |Y | |Z|
+ ≤ 1 + 2n ,
22n − 1 22n − 1 2 −1
|X| |Y | |X∪Y | |X∩Y | |X∪Y | |X∩Y | |Z|
Since 22n −1
+ 22n −1
= 22n −1
+ 22n −1
, moreover, 22n −1
≤ 1 and 22n −1
≤ 22n −1
, then the desired
result holds immediately.
Example 4 Compute the similarity degree of 2p1 and 2p2 .
n
Solution Let φ = ((2p1 → 2p2 ) ∧ (2p2 → 2p1 )) and E ⊆ {0, 1} , it can be checked easily
that E |= φ if and only if either E |= 2(p1 ∧ p2 ) or E |= ¬2p1 ∧ ¬2p2 holds.

T
(i) Suppose that the former, i.e., E |= 2(p1 ∧p2 ), holds, then E ⊆ {(1, 1, x3 , · · · , xn ) | x3 , · · · , xn ∈ {0, 1}},
n−2
since {(1, 1, x3 , · · · , xn ) | x3 , · · · , xn ∈ {0, 1}} = 2n−2 , there are altogether 22 − 1 types of such

IP
epistemic states.
(ii) Suppose that the latter, i.e., E |= ¬2p1 ∧¬2p2 , holds, then E 6⊆ {(1, x2 , x3 , · · · , xn ) | x2 , · · · , xn ∈ {0, 1}}

CR
and E 6⊆ {(x1 , 1, x3 , · · · , xn ) | x1 , x3 , · · · , xn ∈ {0, 1}}, a trivial computation shows that there are
n−1 n−2
altogether 22 +1
− 1 − 22 types of such epistemic states.
n−1 +1 n−2 n−2

22
n−1 +1

22n −1
−2
.
G
Then, according to Definition 8, we have ξM EL (2p1 , 2p2 ) = τM

US
EL (φ) =

Considering the fact that there exists a one-to-one correspondence between the set of epistemic
22 −1−22
22n −1
+22 −1
=
AN
states and the set of valuations satisfying MEL axioms, the notion of epistemic similarity degree
can be equivalently stated as follows.
Proposition 21 Let P be an even probability distribution on the set of epistemic states, that
M

n 1
is, for each ∅ =
6 E ⊆ {0, 1} , P (E) = 22n −1
, then for any φ, ψ ∈ L2 , ξM EL (φ, ψ) = Σ{P (E) | E |=
φ ∧ ψ or E |= ¬φ ∧ ¬ψ}.
ED

4.1 Epistemic similarity degree in MEL vs similarity degree in QL

Let α, β ∈ L, recall first α and β are said to be logically equivalent if v(α) = v(β) holds for
PT

each propositional valuation v. In quantitative logic, this notion has been graded to introduce the
degree of the similarity between α and β.
CE

Definition 9[19] Let α, β ∈ L, define

ξ(α, β) = τ2 ((α → β) ∧ (β → α)),

AC

then we call ξ(α, β) the similarity degree between α and β.

Observe from Definition 9 that for α, β ∈ L containing n propositional variables, ξ(α, β) is a
m
fraction of the form 2n .

The following proposition states the relationship between the notion of epistemic similarity
degree in MEL and that of similarity degree in QL.
m 2m −1
Proposition 22 Let α, β ∈ L, if ξ(α, β) = 2n , then ξM EL (2α, 2β) ≥ 22n −1
.
m
Proof According to Definition 9 and Definition 5, ξ(α, β) = 2n implies that |[(α → β) ∧ (β →
α)]| = m. Denote U = [(α → β) ∧ (β → α)], then an easy verification shows that U = (α−1 (1) ∩

15
 ACCEPTED MANUSCRIPT

−1 −1
β (1)) ∪ (α−1 (0) ∩ β (0)). Moreover, for each epistemic state contained in U , it can be checked
that E |= (2α → 2β) ∧ (2β → 2α). We can prove it by considering the following two cases:
−1
Case 1: E ⊆ (α−1 (1) ∩ β (1)), then we have E |= 2α, E |= 2β, consequently, E |= (2α →
2β) ∧ (2β → 2α).
−1
Case 2: E ∩ (α−1 (0) ∩ β (0)) 6= ∅, then we have E 6|= 2α, E 6|= 2β, consequently, E |=
(2α → 2β) ∧ (2β → 2α).
The above argument shows that there exist at least 2m − 1 epistemic states at which (2α →
2β) ∧ (2β → 2α) is true. In view of the fact that there is a one-to-one correspondence between
the set of epistemic states and that of valuations satisfying MEL axioms, we conclude that there

T
exist at least 2m − 1 valuations in V M EL which assign truth value 1 to (2α → 2β) ∧ (2β → 2α).

IP
2m −1
Then by Definition 8, we have that ξ(2α, 2β) ≥ 22n −1
.
Proposition 22 yields the following corollary.

CR
Corollary 3 Let α, β ∈ L, if ξ(α, β) = 1, then ξM EL (2α, 2β) = 1.
m
Proof Let ξ(α, β) = 2n , then we have from ξ(α, β) = 1 that m = 2n , which, together with
n
2m −1 22 −1
Proposition 22, implies that ξM EL (2α, 2β) ≥
Proposition 23 Let α ∈ L, if τ (α) = m
2n , US
22n −1
= 22n −1

then ξM EL (2α, α) =

Proof It can be proved in a similar way as in Proposition 22.
= 1. Therefore, ξM EL (2α, 2β) = 1.
n
2m +22 −m −2
22n −1
.
AN
In quantitative logic, the notion of similarity degree can induce a kind of pseudo-metric in the
following way.
Definition 10 Let α, β ∈ L, define
M

ρ(α, β) = 1 − ξ(α, β).

ED

Consequently, a kind of logic metric space (F (S), ρ) is thus obtained. By means of the pesudo-
metric ρ, Wang and Zhou then introduced and investigated approximate reasoning[19] from various
points of view.
PT

Definition 11 Let Γ be a theory, i.e., Γ ⊆ F (S), and β ∈ F (S), ε > 0, if

ρ(β, D(Γ)) = inf {ρ(β, α) | α ∈ D(Γ)} < ε,

CE

we then call β is a conclusion of Γ with error less than ε.

In MEL, we can similarly define a notion of epistemic pseudo-metric on the set of logical
AC

formulae in MEL.
Definition 12 Let φ, ψ ∈ L2 , define

ρM EL (φ, ψ) = 1 − ξM EL (φ, ψ).

Owing to Proposition 20, it can be checked that ρM EL is indeed a pseudo-metric on the set of
logical formulae in MEL.
Definition 13 Let Γ be a theory, i.e., Γ ⊆ L2 , and φ ∈ L2 ,ε > 0, if

ρM EL (φ, D(Γ)) = inf {ρM EL (φ, ψ) | ψ ∈ D(Γ)} < ε,

16
 ACCEPTED MANUSCRIPT

we then call φ is a conclusion of Γ with error less than ε.

In what follows, we explore the relationship between approximation reasoning pattern in quan-
titative logic and that in MEL.
2n (1−ε)
−1
Proposition 24 Let Γ ⊆ L, α ∈ L, if ρ(α, D(Γ)) < ε, then ρM EL (2α, D(2Γ)) < 1− 2 2 2n −1
,
where 2Γ = {2β | β ∈ Γ}.
Proof If ρ(α, D(Γ)) < ε, then by Definition 13, there exists a formulae β ∈ D(Γ) such that
ρM EL (α, β) < ε, which together with Definition 12 implies that ξM EL (α, β) > 1 − ε. Combining
n (1−ε)
22 −1
with Proposition 22, we obtain ξM EL (2α, 2β) > 22n −1
. Moreover, it follows from β ∈ D(Γ)
n (1−ε)
22 −1
(i.e., Γ ` β) and Theorem 1 that 2Γ ` 2β. And therefore, ρM EL (2α, D(2Γ)) < 1 − 22n −1
.

T
Proposition 24 yields the following corollary.

IP
Corollary 4 Let Γ ⊆ F (S), α ∈ F (S), if ρ(α, D(Γ)) = 0, then ρM EL (2α, D(2Γ)) = 0.
Proof If ρ(α, D(Γ)) = 0, then ρM EL (α, D(Γ)) < ε holds for arbitrarily chosen positive number

CR
2n (1−ε)
−1
ε. By Proposition 24, we have ρM EL (2α, D(2Γ)) < 1− 2 22n −1
, which, together with limε→0 1−
n
22 (1−ε) −1
2n
2 −1
= 0, implies immediately that ρM EL (2α, D(2Γ)) = 0.

5 Concluding remarks US
AN
In this paper, we have made a modest attempt to present a quantitative analysis of a kind of
epistemic logic MEL used for reasoning about incomplete information. More precisely, by grading
some basic notions in MEL, we have introduced the notion of epistemic truth degree in context of
M

incomplete information, then based upon such a fundamental notion, some derived notions such
as epistemic similarity degree and epistemic pseudo-metric are also investigated.
The obtained results in this paper are the natural generalizations of those in [17-18] in the
ED

sense that rough set theory is one commonly used tool to deal with incomplete information. That
is, the present study is conducted in a more general setting.
PT

Some interesting issues along this research line are worthy of further study. For instance,
the proposed notion of epistemic truth degree can be generalized by using the notion of inclusion
degree in [14], [22], [24] and [26]. What is more, how to apply MEL to concept learning ([15],[23])
CE

from the viewpoint of cognitive computing needs further investigation.

AC

References
[1] Abramsky S, Gabbay D M, Maibaum T S E, editors. Handbook of Logic in Computer Science,
Vol. 1-3. Clarendon Press-Oxford, 1992.

[2] Baader F. The description logic handbook: Theory, implementation and applications. Cam-
bridge university press, 2003.

[3] Banerjee M, Dubois D. A simple modal logic for reasoning about revealed beliefs//Symbolic
and Quantitative Approaches to Reasoning with Uncertainty. Springer Berlin Heidelberg,
2009: 805-816.

17
 ACCEPTED MANUSCRIPT

[4] Banerjee M, Dubois D. A simple logic for reasoning about incomplete knowledge. International
Journal of Approximate Reasoning, 2014, 55(2): 639-653.

[5] Banerjee M, Dubois D, Godo L. Possibilistic vs. relational semantics for logics of incomplete
information//Information Processing and Management of Uncertainty in Knowledge-Based
Systems. Springer International Publishing, 2014: 335-344.

[6] Chellas B F. Modal logic: an introduction. Cambridge: Cambridge University Press, 1980.

[7] Ciucci D, Dubois D, Lawry J. Borderline vs. unknown: comparing three-valued representations
of imperfect information. International Journal of Approximate Reasoning, 2014, 55(9): 1866-

T
1889.

IP
[8] Ciucci D, Dubois D. A modal theorem-preserving translation of a class of three-valued logics
of incomplete information. Journal of Applied Non-Classical Logics, 2013, 23(4): 321-352.

CR
[9] Dubois D, Prade H, Schockaert S. Stable models in generalized possibilistic logic//13th Inter-
national Conference on the Principles of Knowledge Representation and Reasoning, KR 2012.
2012.

US
[10] Dubois D, Prade H, Steven Schockaert: Reasoning about Uncertainty and Explicit Ignorance
in Generalized Possibilistic Logic. ECAI 2014: 261-266
AN
[11] Gabbay D M, Hogger C J, Robinson H A, editors. Handbook of Logic in Artificial Intelligence
and Logic Programming, Vol. 1-4. Clarendon Press-Oxford, 1994.

[12] Hamilton A G. Logic for mathematicians. Cambridge University Press, 1988.

M

[13] Hájek P. Metamathematics of fuzzy logic. Springer Science & Business Media, 1998.
ED

[14] Liang J Y, Xu Z B, Li Y X. Inclusion degree and measures of rough set data analysis. Chinese
Journal of Computers-Chinese edition, 2001, 24(5): 544-547.

[15] Li J H, Mei C L, Xu W H, et al. Concept learning via granular computing: a cognitive

PT

viewpoint. Information Sciences, 2015, 298: 447-467.

[16] Shafer G. A mathematical theory of evidence. Princeton: Princeton university press, 1976.
CE

[17] She Y H, He X L, Wang G J. Rough truth degrees of formulas and approximate reasoning in
rough logic. Fundamenta Informaticae, 2011, 107(1): 67-83.
AC

[18] She Y H, He X L. Rough approximation operators on R0-algebras (nilpotent minimum alge-

bras) with an application in formal logic L∗ . Information Sciences, 2014, 277: 71-89.

[19] Wang G J, Zhou H J. Introduction to mathematical logic and resolution principle. Science
Press, 2009.

[20] Wang G J, Zhou H J. Quantitative logic. Information Sciences, 2009, 179(3): 226-247.

[21] Wu H B. The generalized truth degree of quantitative logic in the logic system (n-valued
NM-logic system). Computers & Mathematics with Applications, 2010, 59(8): 2587-2596.

18
 ACCEPTED MANUSCRIPT

[22] Xu Z B, Liang J Y, Dang C Y, et al. Inclusion degree: a perspective on measures for rough
set data analysis. Information Sciences, 2002, 141(3): 227-236.

[23] Xu W H, Pang J Z, Luo S Q. A novel cognitive system model and approach to transformation
of information granules. International Journal of Approximate Reasoning, 2014, 55(3): 853-
866.

[24] Yao Y Y, Deng X F. Quantitative rough sets based on subsethood measures. Information
Sciences, 2014, 267: 306-322.

[25] Zhou H J, Wang G J. Borel probabilistic and quantitative logic. Science China Information

T
Sciences, 2011, 54(9): 1843-1854.

IP
[26] Zhang H Y, Zhang W X. Hybrid monotonic inclusion measure and its use in measuring
similarity and distance between fuzzy sets. Fuzzy Sets and Systems, 2009, 160(1): 107-118.

CR
US
AN
M
ED
PT
CE
AC

19